Merge pull request #15909 from open-webui/dev

0.6.19

.env.example

@@ -7,6 +7,15 @@ OPENAI_API_KEY=''
 
 # AUTOMATIC1111_BASE_URL="http://localhost:7860"
 
+# For production, you should only need one host as
+# fastapi serves the svelte-kit built frontend and backend from the same host and port.
+# To test with CORS locally, you can set something like
+# CORS_ALLOW_ORIGIN='http://localhost:5173;http://localhost:8080'
+CORS_ALLOW_ORIGIN='*'
+
+# For production you should set this to match the proxy configuration (127.0.0.1)
+FORWARDED_ALLOW_IPS='*'
+
 # DO NOT TRACK
 SCARF_NO_ANALYTICS=true
 DO_NOT_TRACK=true

.gitattributes

@@ -1 +1,49 @@
-*.sh text eol=lf
+# TypeScript
+*.ts text eol=lf
+*.tsx text eol=lf
+
+# JavaScript
+*.js text eol=lf
+*.jsx text eol=lf
+*.mjs text eol=lf
+*.cjs text eol=lf
+
+# Svelte
+*.svelte text eol=lf
+
+# HTML/CSS
+*.html text eol=lf
+*.css text eol=lf
+*.scss text eol=lf
+*.less text eol=lf
+
+# Config files and JSON
+*.json text eol=lf
+*.jsonc text eol=lf
+*.yml text eol=lf
+*.yaml text eol=lf
+*.toml text eol=lf
+
+# Shell scripts
+*.sh text eol=lf
+
+# Markdown & docs
+*.md text eol=lf
+*.mdx text eol=lf
+*.txt text eol=lf
+
+# Git-related
+.gitattributes text eol=lf
+.gitignore text eol=lf
+
+# Prettier and other dotfiles
+.prettierrc text eol=lf
+.prettierignore text eol=lf
+.eslintrc text eol=lf
+.eslintignore text eol=lf
+.stylelintrc text eol=lf
+.editorconfig text eol=lf
+
+# Misc
+*.env text eol=lf
+*.lock text eol=lf

.github/workflows/docker-build.yaml

@@ -14,16 +14,18 @@ env:
 
 jobs:
   build-main-image:
-    runs-on: ${{ matrix.platform == 'linux/arm64' && 'ubuntu-24.04-arm' || 'ubuntu-latest' }}
+    runs-on: ${{ matrix.runner }}
     permissions:
       contents: read
       packages: write
     strategy:
       fail-fast: false
       matrix:
-        platform:
-          - linux/amd64
-          - linux/arm64
+        include:
+          - platform: linux/amd64
+            runner: ubuntu-latest
+          - platform: linux/arm64
+            runner: ubuntu-24.04-arm
 
     steps:
       # GitHub Packages requires the entire repository name to be in lowercase
@@ -111,16 +113,18 @@ jobs:
           retention-days: 1
 
   build-cuda-image:
-    runs-on: ${{ matrix.platform == 'linux/arm64' && 'ubuntu-24.04-arm' || 'ubuntu-latest' }}
+    runs-on: ${{ matrix.runner }}
     permissions:
       contents: read
       packages: write
     strategy:
       fail-fast: false
       matrix:
-        platform:
-          - linux/amd64
-          - linux/arm64
+        include:
+          - platform: linux/amd64
+            runner: ubuntu-latest
+          - platform: linux/arm64
+            runner: ubuntu-24.04-arm
 
     steps:
       # GitHub Packages requires the entire repository name to be in lowercase
@@ -210,17 +214,122 @@ jobs:
           if-no-files-found: error
           retention-days: 1
 
-  build-ollama-image:
-    runs-on: ${{ matrix.platform == 'linux/arm64' && 'ubuntu-24.04-arm' || 'ubuntu-latest' }}
+  build-cuda126-image:
+    runs-on: ${{ matrix.runner }}
     permissions:
       contents: read
       packages: write
     strategy:
       fail-fast: false
       matrix:
-        platform:
-          - linux/amd64
-          - linux/arm64
+        include:
+          - platform: linux/amd64
+            runner: ubuntu-latest
+          - platform: linux/arm64
+            runner: ubuntu-24.04-arm
+
+    steps:
+      # GitHub Packages requires the entire repository name to be in lowercase
+      # although the repository owner has a lowercase username, this prevents some people from running actions after forking
+      - name: Set repository and image name to lowercase
+        run: |
+          echo "IMAGE_NAME=${IMAGE_NAME,,}" >>${GITHUB_ENV}
+          echo "FULL_IMAGE_NAME=ghcr.io/${IMAGE_NAME,,}" >>${GITHUB_ENV}
+        env:
+          IMAGE_NAME: '${{ github.repository }}'
+
+      - name: Prepare
+        run: |
+          platform=${{ matrix.platform }}
+          echo "PLATFORM_PAIR=${platform//\//-}" >> $GITHUB_ENV
+
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata for Docker images (cuda126 tag)
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.FULL_IMAGE_NAME }}
+          tags: |
+            type=ref,event=branch
+            type=ref,event=tag
+            type=sha,prefix=git-
+            type=semver,pattern={{version}}
+            type=semver,pattern={{major}}.{{minor}}
+            type=raw,enable=${{ github.ref == 'refs/heads/main' }},prefix=,suffix=,value=cuda126
+          flavor: |
+            latest=${{ github.ref == 'refs/heads/main' }}
+            suffix=-cuda126,onlatest=true
+
+      - name: Extract metadata for Docker cache
+        id: cache-meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.FULL_IMAGE_NAME }}
+          tags: |
+            type=ref,event=branch
+            ${{ github.ref_type == 'tag' && 'type=raw,value=main' || '' }}
+          flavor: |
+            prefix=cache-cuda126-${{ matrix.platform }}-
+            latest=false
+
+      - name: Build Docker image (cuda126)
+        uses: docker/build-push-action@v5
+        id: build
+        with:
+          context: .
+          push: true
+          platforms: ${{ matrix.platform }}
+          labels: ${{ steps.meta.outputs.labels }}
+          outputs: type=image,name=${{ env.FULL_IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=true
+          cache-from: type=registry,ref=${{ steps.cache-meta.outputs.tags }}
+          cache-to: type=registry,ref=${{ steps.cache-meta.outputs.tags }},mode=max
+          build-args: |
+            BUILD_HASH=${{ github.sha }}
+            USE_CUDA=true
+            USE_CUDA_VER=cu126
+
+      - name: Export digest
+        run: |
+          mkdir -p /tmp/digests
+          digest="${{ steps.build.outputs.digest }}"
+          touch "/tmp/digests/${digest#sha256:}"
+
+      - name: Upload digest
+        uses: actions/upload-artifact@v4
+        with:
+          name: digests-cuda126-${{ env.PLATFORM_PAIR }}
+          path: /tmp/digests/*
+          if-no-files-found: error
+          retention-days: 1
+
+  build-ollama-image:
+    runs-on: ${{ matrix.runner }}
+    permissions:
+      contents: read
+      packages: write
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - platform: linux/amd64
+            runner: ubuntu-latest
+          - platform: linux/arm64
+            runner: ubuntu-24.04-arm
 
     steps:
       # GitHub Packages requires the entire repository name to be in lowercase
@@ -420,6 +529,62 @@ jobs:
         run: |
           docker buildx imagetools inspect ${{ env.FULL_IMAGE_NAME }}:${{ steps.meta.outputs.version }}
 
+  merge-cuda126-images:
+    runs-on: ubuntu-latest
+    needs: [build-cuda126-image]
+    steps:
+      # GitHub Packages requires the entire repository name to be in lowercase
+      # although the repository owner has a lowercase username, this prevents some people from running actions after forking
+      - name: Set repository and image name to lowercase
+        run: |
+          echo "IMAGE_NAME=${IMAGE_NAME,,}" >>${GITHUB_ENV}
+          echo "FULL_IMAGE_NAME=ghcr.io/${IMAGE_NAME,,}" >>${GITHUB_ENV}
+        env:
+          IMAGE_NAME: '${{ github.repository }}'
+
+      - name: Download digests
+        uses: actions/download-artifact@v4
+        with:
+          pattern: digests-cuda126-*
+          path: /tmp/digests
+          merge-multiple: true
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata for Docker images (default latest tag)
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.FULL_IMAGE_NAME }}
+          tags: |
+            type=ref,event=branch
+            type=ref,event=tag
+            type=sha,prefix=git-
+            type=semver,pattern={{version}}
+            type=semver,pattern={{major}}.{{minor}}
+            type=raw,enable=${{ github.ref == 'refs/heads/main' }},prefix=,suffix=,value=cuda126
+          flavor: |
+            latest=${{ github.ref == 'refs/heads/main' }}
+            suffix=-cuda126,onlatest=true
+
+      - name: Create manifest list and push
+        working-directory: /tmp/digests
+        run: |
+          docker buildx imagetools create $(jq -cr '.tags | map("-t " + .) | join(" ")' <<< "$DOCKER_METADATA_OUTPUT_JSON") \
+            $(printf '${{ env.FULL_IMAGE_NAME }}@sha256:%s ' *)
+
+      - name: Inspect image
+        run: |
+          docker buildx imagetools inspect ${{ env.FULL_IMAGE_NAME }}:${{ steps.meta.outputs.version }}
+
   merge-ollama-images:
     runs-on: ubuntu-latest
     needs: [build-ollama-image]

.github/workflows/format-build-frontend.yaml

@@ -32,7 +32,7 @@ jobs:
           node-version: '22'
 
       - name: Install Dependencies
-        run: npm install
+        run: npm install --force
 
       - name: Format Frontend
         run: npm run format
@@ -59,7 +59,7 @@ jobs:
           node-version: '22'
 
       - name: Install Dependencies
-        run: npm ci
+        run: npm ci --force
 
       - name: Run vitest
         run: npm run test:frontend

.gitignore

@@ -1,3 +1,5 @@
+x.py
+yarn.lock
 .DS_Store
 node_modules
 /build
@@ -12,7 +14,8 @@ vite.config.ts.timestamp-*
 __pycache__/
 *.py[cod]
 *$py.class
-
+.nvmrc
+CLAUDE.md
 # C extensions
 *.so
 

.prettierrc

@@ -5,5 +5,6 @@
 	"printWidth": 100,
 	"plugins": ["prettier-plugin-svelte"],
 	"pluginSearchDirs": ["."],
-	"overrides": [{ "files": "*.svelte", "options": { "parser": "svelte" } }]
+	"overrides": [{ "files": "*.svelte", "options": { "parser": "svelte" } }],
+	"endOfLine": "lf"
 }

CHANGELOG.md

@@ -5,6 +5,323 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.6.19] - 2025-08-09
+
+### Added
+
+- ✨ **Modernized Sidebar and Major UI Refinements**: The main navigation sidebar has been completely redesigned with a modern, cleaner aesthetic, featuring a sticky header and footer to keep key controls accessible. Core sidebar logic, like the pinned models list, was also refactored into dedicated components for better performance and maintainability.
+- 🪄 **Guided Response Regeneration**: The "Regenerate" button has been transformed into a powerful new menu. You can now guide the AI's next attempt by suggesting changes in a text prompt, or use one-click options like "Try Again," "Add Details," or "More Concise" to instantly refine and reshape the response to better fit your needs.
+- 🛠️ **Improved Tool Call Handling for GPT-OSS Models**: Implemented robust handling for tool calls specifically for GPT-OSS models, ensuring proper function execution and integration.
+- 🛑 **Stop Button for Merge Responses**: Added a dedicated stop button to immediately halt the generation of merged AI responses, providing users with more control over ongoing outputs.
+- 🔄 **Experimental SCIM 2.0 Support**: Implemented SCIM 2.0 (System for Cross-domain Identity Management) protocol support, enabling enterprise-grade automated user and group provisioning from identity providers like Okta, Azure AD, and Google Workspace for seamless user lifecycle management. Configuration is managed securely via environment variables.
+- 🗂️ **Amazon S3 Vector Support**: You can now use Amazon S3 Vector as a high-performance vector database for your Retrieval-Augmented Generation (RAG) workflows. This provides a scalable, cloud-native storage option for users deeply integrated into the AWS ecosystem, simplifying infrastructure and enabling enterprise-scale knowledge management.
+- 🗄️ **Oracle 23ai Vector Search Support**: Added support for Oracle 23ai's new vector search capabilities as a supported vector database, providing a robust and scalable option for managing large-scale documents and integrating vector search with existing business data at the database level.
+- ⚡ **Qdrant Performance and Configuration Enhancements**: The Qdrant client has been significantly improved with faster data retrieval logic for 'get' and 'query' operations. New environment variables ('QDRANT_TIMEOUT', 'QDRANT_HNSW_M') provide administrators with finer control over query timeouts and HNSW index parameters, enabling better performance tuning for large-scale deployments.
+- 🔐 **Encrypted SQLite Database with SQLCipher**: You can now encrypt your entire SQLite database at rest using SQLCipher. By setting the 'DATABASE_TYPE' to 'sqlite+sqlcipher' and providing a 'DATABASE_PASSWORD', all data is transparently encrypted, providing an essential security layer for protecting sensitive information in self-hosted deployments. Note that this requires additional system libraries and the 'sqlcipher3-wheels' Python package.
+- 🚀 **Efficient Redis Connection Management**: Implemented a shared connection pool cache to reuse Redis connections, dramatically reducing the number of active clients. This prevents connection exhaustion errors, improves performance, and ensures greater stability in high-concurrency deployments and those using Redis Sentinel.
+- ⚡ **Batched Response Streaming for High Performance**: Dramatically improve performance and stability during high-speed response streaming by batching multiple tokens together before sending them to the client. A new 'Stream Delta Chunk Size' advanced parameter can be set per-model or in user/chat settings, significantly reducing CPU load on the server, Redis, and client, and preventing connection issues in high-concurrency environments.
+- ⚙️ **Global Batched Streaming Configuration**: Administrators can now set a system-wide default for response streaming using the new 'CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE' environment variable. This allows for global performance tuning, while still letting per-model or per-chat settings override the default for more granular control.
+- 🔎 **Advanced Chat Search with Status Filters**: Quickly find any conversation with powerful new search filters. You can now instantly narrow down your chats using prefixes like 'pinned:true', 'shared:true', and 'archived:true' directly in the search bar. An intelligent dropdown menu assists you by suggesting available filter options as you type, streamlining your workflow and making chat management more efficient than ever.
+- 🛂 **Granular Chat Controls Permissions**: Administrators can now manage chat settings with greater detail. The main "Chat Controls" permission now acts as a master switch, while new granular toggles for "Valves", "System Prompts", and "Advanced Parameters" allow for more specific control over which sections are visible to users inside the panel.
+- ✍️ **Formatting Toolbar for Chat Input**: Introduced a dedicated formatting toolbar for the rich text chat input field, providing users with more accessible options for text styling and editing, configurable via interface settings.
+- 📑 **Tabbed View for Multi-Model Responses**: You can now enable a new tabbed interface to view responses from multiple models. Instead of side-scrolling cards, this compact view organizes each model's response into its own tab, making it easier to compare outputs and saving vertical space. This feature can be toggled on or off in Interface settings.
+- ↕️ **Reorder Pinned Models via Drag-and-Drop**: You can now organize your pinned models in the sidebar by simply dragging and dropping them into your preferred order. This custom layout is saved automatically, giving you more flexible control over your workspace.
+- 📌 **Quick Model Unpin Shortcut**: You can now quickly unpin a model by holding the Shift key and hovering over it to reveal an instant unpin button, streamlining your workspace customization.
+- ⚡ **Improved Chat Input Performance**: The chat input is now significantly more responsive, especially when pasting or typing large amounts of text. This was achieved by implementing a debounce mechanism for the auto-save feature, which prevents UI lag and ensures a smooth, uninterrupted typing experience.
+- ✍️ **Customizable Floating Quick Actions with Tool Support**: Take full control of your text interaction workflow with new customizable floating quick actions. In Settings, you can create, edit, or disable these actions and even integrate tools using the '{{TOOL:tool_id}}' syntax in your prompts, enabling powerful one-click automations on selected text. This is in addition to using placeholders like '{{CONTENT}}' and '{{INPUT_CONTENT}}' for custom text transformations.
+- 🔒 **Admin Workspace Privacy Control**: Introduced the 'ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS' environment variable (defaults to 'True') allowing administrators to control their access privileges to workspace items (Knowledge, Models, Prompts, Tools). When disabled, administrators adhere to the same access control rules as regular users, enhancing data separation for multi-tenant deployments.
+- 🗄️ **Comprehensive Model Configuration Management**: Administrators can now export the entire model configuration to a file and use a new declarative sync endpoint to manage models in bulk. This powerful feature enables seamless backups, migrations, and state replication across multiple instances.
+- 📦 **Native Redis Cluster Mode Support**: Added full support for connecting to Redis in cluster mode, allowing for scalable and highly available Redis deployments beyond Sentinel-managed setups. New environment variables 'REDIS_CLUSTER' and 'WEBSOCKET_REDIS_CLUSTER' enable the use of 'redis.cluster.RedisCluster' clients.
+- 📊 **Granular OpenTelemetry Metrics Configuration**: Introduced dedicated environment variables and enhanced configuration options for OpenTelemetry metrics, allowing for separate OTLP endpoints, basic authentication credentials, and protocol (HTTP/gRPC) specifically for metrics export, independent of trace settings. This provides greater flexibility for integrating with diverse observability stacks.
+- 🪵 **Granular OpenTelemetry Logging Configuration**: Enhanced the OpenTelemetry logging integration by introducing dedicated environment variables for logs, allowing separate OTLP endpoints, basic authentication credentials, and protocol (HTTP/gRPC) specifically for log export, independent of general OTel settings. The application's default Python logger now leverages this configuration to automatically send logs to your OTel endpoint when enabled via 'ENABLE_OTEL_LOGS'.
+- 📁 **Enhanced Folder Chat Management with Sorting and Time Blocks**: The chat list within folders now supports comprehensive sorting options by title and updated time, along with intelligent time-based grouping (e.g., "Today," "Yesterday") similar to the main chat view, making navigation and organization of project-specific conversations significantly easier.
+- ⚙️ **Configurable Datalab Marker API & Advanced Processing Options**: Enhanced Datalab Marker API integration, allowing administrators to configure custom API base URLs for self-hosting and to specify comprehensive processing options via a new 'additional_config' JSON parameter. This replaces the deprecated language selection feature and provides granular control over document extraction, with streamlined API endpoint resolution for more robust self-hosted deployments.
+- 🧑‍💼 **Export All Users to CSV**: Administrators can now export a complete list of all users to a CSV file directly from the Admin Panel's database settings. This provides a simple, one-click way to generate user data for auditing, reporting, or management purposes.
+- 🛂 **Customizable OAuth 'sub' Claim**: Administrators can now use the 'OAUTH_SUB_CLAIM_OVERRIDE' environment variable to specify which claim from the identity provider should be used as the unique user identifier ('sub'). This provides greater flexibility and control for complex enterprise authentication setups where modifying the IDP's default claims is not possible.
+- 👁️ **Password Visibility Toggle for Input Fields**: Password fields across the application (login, registration, user management, and account settings) now utilize a new 'SensitiveInput' component, providing a consistent toggle to reveal/hide passwords for improved usability and security.
+- 🛂 **Optional "Confirm Password" on Sign-Up**: To help prevent password typos during account creation, administrators can now enable a "Confirm Password" field on the sign-up page. This feature is disabled by default and can be activated via an environment variable for enhanced user experience.
+- 💬 **View Full Chat from User Feedback**: Administrators can now easily navigate to the full conversation associated with a user feedback entry directly from the feedback modal, streamlining the review and troubleshooting process.
+- 🎚️ **Intuitive Hybrid Search BM25-Weight Slider**: The numerical input for the BM25-Weight parameter in Hybrid Search has been replaced with an interactive slider, offering a more intuitive way to adjust the balance between lexical and semantic search. A "Default/Custom" toggle and clearer labels enhance usability and understanding of this key parameter.
+- ⚙️ **Enhanced Bulk Function Synchronization**: The API endpoint for synchronizing functions has been significantly improved to reliably handle bulk updates. This ensures that importing and managing large libraries of functions is more robust and error-free for administrators.
+- 🖼️ **Option to Disable Image Compression in Channels**: Introduced a new setting under Interface options to allow users to force-disable image compression specifically for images posted in channels, ensuring higher resolution for critical visual content.
+- 🔗 **Custom CORS Scheme Support**: Introduced a new environment variable 'CORS_ALLOW_CUSTOM_SCHEME' that allows administrators to define custom URL schemes (e.g., 'app://') for CORS origins, enabling greater flexibility for local development or desktop client integrations.
+- ♿ **Translatable and Accessible Banners**: Enhanced banner elements with translatable badge text and proper ARIA attributes (aria-label, aria-hidden) for SVG icons, significantly improving accessibility and screen reader compatibility.
+- ⚠️ **OAuth Configuration Warning for Missing OPENID_PROVIDER_URL**: Added a proactive startup warning that notifies administrators when OAuth providers (Google, Microsoft, or GitHub) are configured but the essential 'OPENID_PROVIDER_URL' environment variable is missing. This prevents silent OAuth logout failures and guides administrators to complete their setup correctly.
+- ♿ **Major Accessibility Enhancements**: Key parts of the interface have been made significantly more accessible. The user profile menu is now fully navigable via keyboard, essential controls in the Playground now include proper ARIA labels for screen readers, and decorative images have been hidden from assistive technologies to reduce audio clutter. Menu buttons also feature enhanced accessibility with 'aria-label', 'aria-hidden' for SVGs, and 'aria-pressed' for toggle buttons.
+- ⚙️ **General Backend Refactoring**: Implemented various backend improvements to enhance performance, stability, and security, ensuring a more resilient and reliable platform for all users, including refining logging output to be cleaner and more efficient by conditionally including 'extra_json' fields and improving consistent metadata handling in vector database operations, and laying preliminary scaffolding for future analytics features.
+- 🌐 **Localization & Internationalization Improvements**: Refined and expanded translations for Catalan, Danish, Korean, Persian, Polish, Simplified Chinese, and Spanish, ensuring a more fluent and native experience for global users across all supported languages.
+
+### Fixed
+
+- 🛡️ **Hardened Channel Message Security**: Fixed a key permission flaw that allowed users with channel access to edit or delete messages belonging to others. The system now correctly enforces that users can only modify their own messages, protecting data integrity in shared channels.
+- 🛡️ **Hardened OAuth Security by Removing JWT from URL**: Fixed a critical security vulnerability where the authentication token was exposed in the URL after a successful OAuth login. The token is now transferred via a browser cookie, preventing potential leaks through browser history or server logs and protecting user sessions.
+- 🛡️ **Hardened Chat Completion API Security**: The chat completion API endpoint now includes an explicit ownership check, ensuring non-admin users cannot access chats that do not belong to them and preventing potential unauthorized access.
+- 🛠️ **Resilient Model Loading**: Fixed an issue where a failure in loading the model list (e.g., from a misconfigured provider) would prevent the entire user interface, including the admin panel, from loading. The application now gracefully handles these errors, ensuring the UI remains accessible.
+- 🔒 **Resolved FIPS Self-Test Failure**: Fixed a critical issue that prevented Open WebUI from running on FIPS-compliant systems, specifically resolving the "FATAL FIPS SELFTEST FAILURE" error related to OpenSSL and SentenceTransformers, restoring compatibility with secure environments.
+- 📦 **Redis Cluster Connection Restored**: Fixed an issue where the backend was unable to connect to Redis in cluster mode, now ensuring seamless integration with scalable Redis cluster deployments.
+- 📦 **PGVector Connection Stability**: Fixed an issue where read-only operations could leave database transactions idle, preventing potential connection errors and improving overall database stability and resource management.
+- 🛠️ **OpenAPI Tool Integration for Array Parameters Fixed**: Resolved a critical bug where external tools using array parameters (e.g., for tags) would fail when used with OpenAI models. The system now correctly generates the required 'items' property in the function schema, restoring functionality and preventing '400 Bad Request' errors.
+- 🛠️ **Tool Creation for Users Restored**: Fixed a bug in the code editor where status messages were incorrectly prepended to tool scripts, causing a syntax error upon saving. All authorized users can now reliably create and save new tools.
+- 📁 **Folder Knowledge Processing Restored**: Fixed a bug where files uploaded to folder and model knowledge bases were not being extracted or analyzed for Retrieval-Augmented Generation (RAG) when the 'Max Upload Count' setting was empty, ensuring seamless document processing and knowledge augmentation.
+- 🧠 **Custom Model Knowledge Base Updates Recognized**: Fixed a bug where custom models linked to to knowledge bases did not automatically recognize newly added files to those knowledge bases. Models now correctly incorporate the latest information from updated knowledge collections.
+- 📦 **Comprehensive Redis Key Prefixing**: Corrected hardcoded prefixes to ensure the REDIS_KEY_PREFIX is now respected across all WebSocket and task management keys. This prevents data collisions in multi-instance deployments and improves compatibility with Redis cluster mode.
+- ✨ **More Descriptive OpenAI Router Errors**: The OpenAI-compatible API router now propagates detailed upstream error messages instead of returning a generic 'Bad Request'. This provides clear, actionable feedback for developers and API users, making it significantly easier to debug and resolve issues with model requests.
+- 🔐 **Hardened OIDC Signout Flow**: The OpenID Connect signout process now verifies that the 'OPENID_PROVIDER_URL' is configured before attempting to communicate with it, preventing potential errors and ensuring a more reliable logout experience.
+- 🍓 **Raspberry Pi Compatibility Restored**: Pinned the pyarrow library to version 20.0.0, resolving an "Illegal Instruction" crash on ARM-based devices like the Raspberry Pi and ensuring stable operation on this hardware.
+- 📁 **Folder System Prompt Variables Restored**: Fixed a bug where prompt variables (e.g., '{{CURRENT_DATETIME}}') were not being rendered in Folder-level System Prompts. This restores an important capability for creating dynamic, context-aware instructions for all chats within a project folder.
+- 📝 **Note Access in Knowledge Retrieval Fixed**: Corrected a permission oversight in knowledge retrieval, ensuring users can always use their own notes as a source for RAG without needing explicit sharing permissions.
+- 🤖 **Title Generation Compatibility for GPT-5 Models**: Added support for 'gpt-5' models in the payload handler, which correctly converts the deprecated 'max_tokens' parameter to 'max_completion_tokens'. This resolves title generation failures and ensures seamless operation with the latest generation of models.
+- ⚙️ **Correct API 'finish_reason' in Streaming Responses**: Fixed an issue where intermediate 'reasoning_content' chunks in streaming API responses incorrectly reported a 'finish_reason' of 'stop'. The 'finish_reason' is now correctly set to 'null' for these chunks, ensuring compatibility with third-party applications that rely on this field.
+- 📈 **Evaluation Pages Stability**: Resolved a crash on the Leaderboard and Feedbacks pages when processing legacy feedback entries that were missing a 'rating' field. The system now gracefully handles this older data, ensuring both pages load reliably for all users.
+- 🤝 **Reliable Collaborative Session Cleanup**: Fixed an asynchronous bug in the real-time collaboration engine that prevented document sessions from being properly cleaned up after all users had left. This ensures greater stability and resource management for features like Collaborative Notes.
+- 🧠 **Enhanced Memory Stability and Security**: Refactored memory update and delete operations to strictly enforce user ownership, preventing potential data integrity issues. Additionally, improved error handling for memory queries now provides clearer feedback when no memories exists.
+- 🧑‍⚖️ **Restored Admin Access to User Feedback**: Fixed a permission issue that blocked administrators from viewing or editing user feedback they didn't create, ensuring they can properly manage all evaluations across the platform.
+- 🔐 **PGVector Encryption Fix for Metadata**: Corrected a SQL syntax error in the experimental 'PGVECTOR_PGCRYPTO' feature that prevented encrypted metadata from being saved. Document uploads to encrypted PGVector collections now work as intended.
+- 🔍 **Serply Web Search Integration Restored**: Fixed an issue where incorrect parameters were passed to the Serply web search engine, restoring its functionality for RAG and web search workflows.
+- 🔍 **Resilient Web Search Processing**: Web search retrieval now gracefully handles search results that are missing a 'snippet', preventing crashes and ensuring that RAG workflows complete successfully even with incomplete data from search engines.
+- 🖼️ **Table Pasting in Rich Text Input Displayed Correctly**: Fixed an issue where pasting table text into the rich text input would incorrectly display it as code. Tables are now properly rendered as expected, improving content formatting and user experience.
+- ✍️ **Rich Text Input TypeError Resolution**: Addressed a potential 'TypeError: ue.getWordAtDocPos is not a function' in 'MessageInput.svelte' by refactoring how the 'getWordAtDocPos' function is accessed and referenced from 'RichTextInput.svelte', ensuring stable rich text input behavior, especially after production restarts.
+- ✏️ **Manual Code Block Creation in Chat Restored**: Fixed an issue where typing three backticks and then pressing Shift+Enter would incorrectly remove the backticks when "Enter to Send" mode was active. This ensures users can reliably create multi-line code blocks manually.
+- 🎨 **Consistent Dark Mode Background**: Fixed an issue where the application background could incorrectly flash or remain white during page loads and refreshes in dark mode, ensuring a seamless and consistent visual experience.
+- 🎨 **'Her' Theme Rendering Fixed**: Corrected a bug that caused the "Her" theme to incorrectly render as a dark theme in some situations. The theme now reliably applies its intended light appearance across all sessions.
+- 📜 **Corrected Markdown Table Line Break Rendering**: Fixed an issue where line breaks ('<br>') within Markdown tables were displayed as raw HTML instead of being rendered correctly. This ensures that tables with multi-line cell content are now displayed as intended.
+- 🚦 **Corrected App Configuration for Pending Users**: Fixed an issue where users awaiting approval could incorrectly load the full application interface, leading to a confusing or broken UI. This ensures that only fully approved users receive the standard app 'config', resulting in a smoother and more reliable onboarding experience.
+- 🔄 **Chat Cloning Now Includes Tags, Folder Status, and Pinned Status**: When cloning a chat or shared chat, its associated tags, folder organization, and pinned status are now correctly replicated, ensuring consistent chat management.
+- ⚙️ **Enhanced Backend Reliability**: Resolved a potential crash in knowledge base retrieval when referencing a deleted note. Additionally, chat processing was refactored to ensure model information is saved more reliably, enhancing overall system stability.
+- ⚙️ **Floating 'Ask/Explain' Modal Stability**: Fixed an issue that spammed the console with errors when navigating away while a model was generating a response in the floating 'Ask' or 'Explain' modals. In-flight requests are now properly cancelled, improving application stability.
+- ⚡ **Optimized User Count Checks**: Improved performance for user count and existence checks across the application by replacing resource-intensive 'COUNT' queries with more efficient 'EXISTS' queries, reducing database load.
+- 🔐 **Hardened OpenTelemetry Exporter Configuration**: The OTLP HTTP exporter no longer uses a potentially insecure explicit flag, improving security by relying on the connection URL's protocol (HTTP/HTTPS) to ensure transport safety.
+- 📱 **Mobile User Menu Closing Behavior Fixed**: Resolved an issue where the user menu would remain open on mobile devices after selecting an option, ensuring the menu correctly closes and returns focus to the main interface for a smoother mobile experience.
+- 📱 **OnBoarding Page Display Fixed on Mobile**: Resolved an issue where buttons on the OnBoarding page were not consistently visible on certain mobile browsers, ensuring a functional and complete user experience across devices.
+- ↕️ **Improved Pinned Models Drag-and-Drop Behavior**: The drag-and-drop functionality for reordering pinned models is now explicitly disabled on mobile devices, ensuring better usability and preventing potential UI conflicts or unexpected behavior.
+- 📱 **PWA Rotation Behavior Corrected**: The Progressive Web App now correctly respects the device's screen orientation lock, preventing unwanted rotation and ensuring a more native mobile experience.
+- ✏️ **Improved Chat Title Editing Behavior**: Changes to a chat title are now reliably saved when the user clicks away or presses Enter, replacing a less intuitive behavior that could accidentally discard edits. This makes renaming chats a smoother and more predictable experience.
+- ✏️ **Underscores Allowed in Prompt Commands**: Fixed the validation for prompt commands to correctly allow the use of underscores ('\_'), aligning with documentation examples and improving flexibility in naming custom prompts.
+- 💡 **Title Generation Button Behavior Fixed**: Resolved an issue where clicking the "Generate Title" button while editing a chat or note title would incorrectly save the title before generation could start. The focus is now managed correctly, ensuring a smooth and predictable user experience.
+- ✏️ **Consistent Chat Input Height**: Fixed a minor visual bug where the chat input field's height would change slightly when toggling the "Rich Text Input for Chat" setting, ensuring a more stable and consistent layout.
+- 🙈 **Admin UI Toggle Stability**: Fixed a visual glitch in the Admin settings where toggle switches could briefly display an incorrect state on page load, ensuring the UI always accurately reflects the saved settings.
+- 🙈 **Community Sharing Button Visibility**: The "Share to Community" button on the feedback page is now correctly hidden when the Enable Community Sharing feature is disabled in the admin settings, ensuring the UI respects the configured sharing policy.
+- 🙈 **"Help Us Translate" Link Visibility**: The "Help us translate" link in settings is now correctly hidden in deployments with specific license configurations, ensuring a cleaner interface for enterprise users.
+- 🔗 **Robust Tool Server URL Handling**: Fixed an issue where providing a full URL for a tool server's OpenAPI specification resulted in an invalid path. The system now correctly handles both absolute URLs and relative paths, improving configuration flexibility.
+- 🔧 **Improved Azure URL Detection**: The logic for identifying Azure OpenAI endpoints has been made more robust, ensuring all valid Azure URLs are now correctly detected for a smoother connection setup.
+- ⚙️ **Corrected Direct Connection Save Logic**: Fixed a bug in the Admin Connections settings page by removing a redundant save action for 'Direct Connections', leading to more reliable and predictable behavior when updating settings.
+- 🔗 **Corrected "Discover" Links**: The "Discover" links for models, prompts, tools, and functions now point to their specific, relevant pages on openwebui.com, improving content discovery for users.
+- ⏱️ **Refined Display of AI Thought Duration**: Adjusted the display logic for AI thought (reasoning) durations to more accurately show very short thought times as "less than a second," improving clarity in AI process feedback.
+- 📜 **Markdown Line Break Rendering Refinement**: Improved handling of line breaks within Markdown rendering for better visual consistency.
+- 🛠️ **Corrected OpenTelemetry Docker Compose Example**: The docker-compose.otel.yaml file has been fixed and enhanced by removing duplicates, adding necessary environment variables, and hardening security settings, ensuring a more reliable out-of-box observability setup.
+- 🛠️ **Development Script CORS Fix**: Corrected the CORS origin URL in the local development script (dev.sh) by removing the trailing slash, ensuring a more reliable and consistent setup for developers.
+- ⬆️ **OpenTelemetry Libraries Updated**: Upgraded all OpenTelemetry-related libraries to their latest versions, ensuring better performance, stability, and compatibility for observability.
+
+### Changed
+
+- ❗ **Docling Integration Upgraded to v1 API (Breaking Change)**: The integration with the Docling document processing engine has been updated to its new, stable '/v1' API. This is required for compatibility with Docling version 1.0.0 and newer. As a result, older versions of Docling are no longer supported. Users who rely on Docling for document ingestion **must upgrade** their docling-serve instance to ensure continued functionality.
+- 🗣️ **Admin-First Whisper Language Priority**: The global WHISPER_LANGUAGE setting now acts as a strict override for audio transcriptions. If set, it will be used for all speech-to-text tasks, ignoring any language specified by the user on a per-request basis. This gives administrators more control over transcription consistency.
+- ✂️ **Datalab Marker API Language Selection Removed**: The separate language selection option for the Datalab Marker API has been removed, as its functionality is now integrated and superseded by the more comprehensive 'additional_config' parameter. Users should transition to using 'additional_config' for relevant language and processing settings.
+- 📄 **Documentation and Releases Links Visibility**: The "Documentation" and "Releases" links in the user menu are now visible only to admin users, streamlining the user interface for non-admin roles.
+
+## [0.6.18] - 2025-07-19
+
+### Fixed
+
+- 🚑 **Users Not Loading in Groups**: Resolved an issue where user list was not displaying within user groups, restoring full visibility and management of group memberships for teams and admins.
+
+## [0.6.17] - 2025-07-19
+
+### Added
+
+- 📂 **Dedicated Folder View with Chat List**: Clicking a folder now reveals a brand-new landing page showcasing a list of all chats within that folder, making navigation simpler and giving teams immediate visibility into project-specific conversations.
+- 🆕 **Streamlined Folder Creation Modal**: Creating a new folder is now a seamless, unified experience with a dedicated modal that visually and functionally matches the edit folder flow, making workspace organization more intuitive and error-free for all users.
+- 🗃️ **Direct File Uploads to Folder Knowledge**: You can now upload files straight to a folder’s knowledge—empowering you to enrich project spaces by adding resources and documents directly, without the need to pre-create knowledge bases beforehand.
+- 🔎 **Chat Preview in Search**: When searching chats, instantly preview results in context without having to open them—making discovery, auditing, and recall dramatically quicker, especially in large, active teams.
+- 🖼️ **Image Upload and Inline Insertion in Notes**: Notes now support inserting images directly among your text, letting you create rich, visually structured documentation, brainstorms, or reports in a more natural and engaging way—no more images just as attachments.
+- 📱 **Enhanced Note Selection Editing and Q&A**: Select any portion of your notes to either edit just the highlighted part or ask focused questions about that content—streamlining workflows, boosting productivity, and making reviews or AI-powered enhancements more targeted.
+- 📝 **Copy Notes as Rich Text**: Copy entire notes—including all formatting, images, and structure—directly as rich text for seamless pasting into emails, reports, or other tools, maintaining clarity and consistency outside the WebUI.
+- ⚡ **Fade-In Streaming Text Experience**: Live-generated responses now elegantly fade in as the AI streams them, creating a more natural and visually engaging reading experience; easily toggled off in Interface settings if you prefer static displays.
+- 🔄 **Settings for Follow-Up Prompts**: Fine-tune your follow-up prompt experience—with new controls, you can choose to keep them visible or have them inserted directly into the message input instead of auto-submitting, giving you more flexibility and control over your workflow.
+- 🔗 **Prompt Variable Documentation Quick Link**: Access documentation for prompt variables in one click from the prompt editor modal—shortening the learning curve and making advanced prompt-building more accessible.
+- 📈 **Active and Total User Metrics for Telemetry**: Gain valuable insights into usage patterns and platform engagement with new metrics tracking active and total users—enhancing auditability and planning for large organizations.
+- 🏷️ **Traceability with Log Trace and Span IDs**: Each log entry now carries detailed trace and span IDs, making it much easier for admins to pinpoint and resolve issues across distributed systems or in complex troubleshooting.
+- 👥 **User Group Add/Remove Endpoints**: Effortlessly add or remove users from groups with new, improved endpoints—giving admins and team leads faster, clearer control over collaboration and permissions.
+- ⚙️ **Note Settings and Controls Streamlined**: The main “Settings” for notes are now simply called “Controls”, and note files now reside in a dedicated controls section, decluttering navigation and making it easier to find and configure note-related options.
+- 🚀 **Faster Admin User Page Loads**: The user list endpoint for admins has been optimized to exclude heavy profile images, speeding up load times for large teams and reducing waiting during administrative tasks.
+- 📡 **Chat ID Header Forwarding**: Ollama and OpenAI router requests now include the chat ID in request headers, enabling better request correlation and debugging capabilities across AI model integrations.
+- 🧠 **Enhanced Reasoning Tag Processing**: Improved and expanded reasoning tag parsing to handle various tag formats more robustly, including standard XML-style tags and custom delimiters, ensuring better AI reasoning transparency and debugging capabilities.
+- 🔐 **OAuth Token Endpoint Authentication Method**: Added configurable OAuth token endpoint authentication method support, providing enhanced flexibility and security options for enterprise OAuth integrations and identity provider compatibility.
+- 🛡️ **Redis Sentinel High Availability Support**: Comprehensive Redis Sentinel failover implementation with automatic master discovery, intelligent retry logic for connection failures, and seamless operation during master node outages—eliminating single points of failure and ensuring continuous service availability in production deployments.
+- 🌐 **Localization & Internationalization Improvements**: Refined and expanded translations for Simplified Chinese, Traditional Chinese, French, German, Korean, and Polish, ensuring a more fluent and native experience for global users across all supported languages.
+
+### Fixed
+
+- 🏷️ **Hybrid Search Functionality Restored**: Hybrid search now works seamlessly again—enabling more accurate, relevant, and comprehensive knowledge discovery across all RAG-powered workflows.
+- 🚦 **Note Chat - Edit Button Disabled During AI Generation**: The edit button when chatting with a note is now disabled while the AI is responding—preventing accidental edits and ensuring workflow clarity during chat sessions.
+- 🧹 **Cleaner Database Credentials**: Database connection no longer duplicates ‘@’ in credentials, preventing potential connection issues and ensuring smoother, more reliable integrations.
+- 🧑‍💻 **File Deletion Now Removes Related Vector Data**: When files are deleted from storage, they are now purged from the vector database as well, ensuring clean data management and preventing clutter or stale search results.
+- 📁 **Files Modal Translation Issues Fixed**: All modal dialog strings—including “Using Entire Document” and “Using Focused Retrieval”—are now fully translated for a more consistent and localized UI experience.
+- 🚫 **Drag-and-Drop File Upload Disabled for Unsupported Models**: File upload by drag-and-drop is disabled when using models that do not support attachments—removing confusion and preventing workflow interruptions.
+- 🔑 **Ollama Tool Calls Now Reliable**: Fixed issues with Ollama-based tool calls, ensuring uninterrupted AI augmentation and tool use for every chat.
+- 📄 **MIME Type Help String Correction**: Cleaned up mimetype help text by removing extraneous characters, providing clearer guidance for file upload configurations.
+- 📝 **Note Editor Permission Fix**: Removed unnecessary admin-only restriction from note chat functionality, allowing all authorized users to access note editing features as intended.
+- 📋 **Chat Sources Handling Improved**: Fixed sources handling logic to prevent duplicate source assignments in chat messages, ensuring cleaner and more accurate source attribution during conversations.
+- 😀 **Emoji Generation Error Handling**: Improved error handling in audio router and fixed metadata structure for emoji generation tasks, preventing crashes and ensuring more reliable emoji generation functionality.
+- 🔒 **Folder System Prompt Permission Enforcement**: System prompt fields in folder edit modal are now properly hidden for users without system prompt permissions, ensuring consistent security policy enforcement across all folder management interfaces.
+- 🌐 **WebSocket Redis Lock Timeout Type Conversion**: Fixed proper integer type conversion for WebSocket Redis lock timeout configuration with robust error handling, preventing potential configuration errors and ensuring stable WebSocket connections.
+- 📦 **PostHog Dependency Added**: Added PostHog 5.4.0 library to resolve ChromaDB compatibility issues, ensuring stable vector database operations and preventing library version conflicts during deployment.
+
+### Changed
+
+- 👀 **Tiptap Editor Upgraded to v3**: The underlying rich text editor has been updated for future-proofing, though some supporting libraries remain on v2 for compatibility. For now, please install dependencies using 'npm install --force' to avoid installation errors.
+- 🚫 **Removed Redundant or Unused Strings and Elements**: Miscellaneous unused, duplicate, or obsolete code and translations have been cleaned up to maintain a streamlined and high-performance experience.
+
+## [0.6.16] - 2025-07-14
+
+### Added
+
+- 🗂️ **Folders as Projects**: Organize your workflow with folder-based projects—set folder-level system prompts and associate custom knowledge, bringing seamless, context-rich management to teams and users handling multiple initiatives or clients.
+- 📁 **Instant Folder-Based Chat Creation**: Start a new chat directly from any folder; just click and your new conversation is automatically embedded in the right project context—no more manual dragging or setup, saving time and eliminating mistakes.
+- 🧩 **Prompt Variables with Automatic Input Modal**: Prompts containing variables now display a clean, auto-generated input modal that **autofocuses on the first field** for instant value entry—just select the prompt and fill in exactly what’s needed, reducing friction and guesswork.
+- 🔡 **Variable Input Typing in Prompts**: Define input types for prompt variables (e.g., text, textarea, number, select, color, date, map and more), giving everyone a clearer and more precise prompt-building experience for advanced automation or workflows.
+- 🚀 **Base Model List Caching**: Cache your base model list to speed up model selection and reduce repeated API calls; toggle this in Admin Settings > Connections for responsive model management even in large or multi-provider setups.
+- ⏱️ **Configurable Model List Cache TTL**: Take control over model list caching with the new MODEL_LIST_CACHE_TTL environment variable. Set a custom cache duration in seconds to balance performance and freshness, reducing API requests in stable environments or ensuring rapid updates when models change frequently.
+- 🔖 **Reference Notes as Knowledge or in Chats**: Use any note as knowledge for a model or folder, or reference it directly from chat—integrate living documentation into your Retrieval Augmented Generation workflows or discussions, bridging knowledge and action.
+- 📝 **Chat Directly with Notes (Experimental)**: Ask questions about any note, and directly edit or update notes from within a chat—unlock direct AI-powered brainstorming, summarization, and cleanup, like having your own collaborative AI canvas.
+- 🤝 **Collaborative Notes with Multi-User Editing**: Share notes with others and collaborate live—multiple users can edit a note in real-time, boosting cooperative knowledge building and workflow documentation.
+- 🛡️ **Collaborative Note Permissions**: Control who can view or edit each note with robust sharing permissions, ensuring privacy or collaboration per your organizational needs.
+- 🔗 **Copy Link to Notes**: Quickly copy and share direct links to notes for easier knowledge transfer within your team or external collaborators.
+- 📋 **Task List Support in Notes**: Add, organize, and manage checklists or tasks inside your notes—plan projects, track to-dos, and keep everything actionable in a single space.
+- 🧠 **AI-Generated Note Titles**: Instantly generate relevant and concise titles for your notes using AI—keep your knowledge library organized without tedious manual editing.
+- 🔄 **Full Undo/Redo Support in Notes**: Effortlessly undo or redo your latest note changes—never fear mistakes or accidental edits while collaborating or writing.
+- 📝 **Enhanced Note Word/Character Counter**: Always know the size of your notes with built-in counters, making it easier to adhere to length guidelines for shared or published content.
+- 🖊️ **Floating & Bubble Formatting Menus in Note Editor**: Access text formatting tools through both a floating menu and an intuitive bubble menu directly in the note editor—making rich text editing faster, more discoverable, and easier than ever.
+- ✍️ **Rich Text Prompt Insertion**: A new setting allows prompts to be inserted directly into the chat box as fully-formatted rich text, preserving Markdown elements like headings, lists, and bold text for a more intuitive and visually consistent editing experience.
+- 🌐 **Configurable Database URL**: WebUI now supports more flexible database configuration via new environment variables—making deployment and scaling simpler across various infrastructure setups.
+- 🎛️ **Completely Frontend-Handled File Upload in Temporary Chats**: When using temporary chats, file extraction now occurs fully in your browser with zero files sent to the backend, further strengthening privacy and giving you instant feedback.
+- 🔄 **Enhanced Banner and Chat Command Visibility**: Banner handling and command feedback in chat are now clearer and more contextually visible, making alerts, suggestions, and automation easier to spot and interact with for all users.
+- 📱 **Mobile Experience Polished**: The "new chat" button is back in mobile, plus core navigation and input controls have been smoothed out for better usability on phones and tablets.
+- 📄 **OpenDocument Text (.odt) Support**: Seamlessly upload and process .odt files from open-source office suites like LibreOffice and OpenOffice, expanding your ability to build knowledge from a wider range of document formats.
+- 📑 **Enhanced Markdown Document Splitting**: Improve knowledge retrieval from Markdown files with a new header-aware splitting strategy. This method intelligently chunks documents based on their header structure, preserving the original context and hierarchy for more accurate and relevant RAG results.
+- 📚 **Full Context Mode for Knowledge Bases**: When adding a knowledge base to a folder or custom model, you can now toggle full context mode for the entire knowledge base. This bypasses the usual chunking and retrieval process, making it perfect for leaner knowledge bases.
+- 🕰️ **Configurable OAuth Timeout**: Enhance login reliability by setting a custom timeout (OAUTH_TIMEOUT) for all OAuth providers (Google, Microsoft, GitHub, OIDC), preventing authentication failures on slow or restricted networks.
+- 🎨 **Accessibility & High-Contrast Theme Enhancements**: Major accessibility overhaul with significant updates to the high-contrast theme. Improved focus visibility, ARIA labels, and semantic HTML ensure core components like the chat interface and model selector are fully compliant and readable for visually impaired users.
+- ↕️ **Resizable System Prompt Fields**: Conveniently resize system prompt input fields to comfortably view and edit lengthy or complex instructions, improving the user experience for advanced model configuration.
+- 🔧 **Granular Update Check Control**: Gain finer control over outbound connections with the new ENABLE_VERSION_UPDATE_CHECK flag. This allows administrators to disable version update checks independently of the full OFFLINE_MODE, perfect for environments with restricted internet access that still need to download embedding models.
+- 🗃️ **Configurable Qdrant Collection Prefix**: Enhance scalability by setting a custom QDRANT_COLLECTION_PREFIX. This allows multiple Open WebUI instances to share a single Qdrant cluster safely, ensuring complete data isolation between separate deployments without conflicts.
+- ⚙️ **Improved Default Database Performance**: Enhanced out-of-the-box performance by setting smarter database connection pooling defaults, reducing API response times for users on non-SQLite databases without requiring manual configuration.
+- 🔧 **Configurable Redis Key Prefix**: Added support for the REDIS_KEY_PREFIX environment variable, allowing multiple Open WebUI instances to share a Redis cluster with isolated key namespaces for improved multi-tenancy.
+- ➡️ **Forward User Context to Reranker**: For advanced RAG integrations, user information (ID, name, email, role) can now be forwarded as HTTP headers to external reranking services, enabling personalized results or per-user access control.
+- ⚙️ **PGVector Connection Pooling**: Enhance performance and stability for PGVector-based RAG by enabling and configuring the database connection pool. New environment variables allow fine-tuning of pool size, timeout, and overflow settings to handle high-concurrency workloads efficiently.
+- ⚙️ **General Backend Refactoring**: Extensive refactoring delivers a faster, more reliable, and robust backend experience—improving chat speed, model management, and day-to-day reliability.
+- 🌍 **Expanded & Improved Translations**: Enjoy a more accessible and intuitive experience thanks to comprehensive updates and enhancements for Chinese (Simplified and Traditional), German, French, Catalan, Irish, and Spanish translations throughout the interface.
+
+### Fixed
+
+- 🛠️ **Rich Text Input Stability and Performance**: Multiple improvements ensure faster, cleaner text editing and rendering with reduced glitches—especially supporting links, color picking, checkbox controls, and code blocks in notes and chats.
+- 📷 **Seamless iPhone Image Uploads**: Effortlessly upload photos from iPhones and other devices using HEIC format—images are now correctly recognized and processed, eliminating compatibility issues.
+- 🔄 **Audio MIME Type Registration**: Issues with audio file content types have been resolved, guaranteeing smoother, error-free uploads and playback for transcription or note attachments.
+- 🖍️ **Input Commands Now Always Visible**: Input commands (like prompts or knowledge) dynamically adjust their height on small screens, ensuring nothing is cut off and every tool remains easily accessible.
+- 🛑 **Tool Result Rendering**: Fixed display problems with tool results, providing fast, clear feedback when using external or internal tools.
+- 🗂️ **Table Alignment in Markdown**: Markdown tables are now rendered and aligned as expected, keeping reports and documentation readable.
+- 🖼️ **Thread Image Handling**: Fixed an issue where messages containing only images in threads weren’t displayed correctly.
+- 🗝️ **Note Access Control Security**: Tightened access control logic for notes to guarantee that shared or collaborative notes respect all user permissions and privacy safeguards.
+- 🧾 **Ollama API Compatibility**: Fixed model parameter naming in the API to ensure uninterrupted compatibility for all Ollama endpoints.
+- 🛠️ **Detection for 'text/html' Files**: Files loaded with docling/tika are now reliably detected as the correct type, improving knowledge ingestion and document parsing.
+- 🔐 **OAuth Login Stability**: Resolved a critical OAuth bug that caused login failures on subsequent attempts after logging out. The user session is now completely cleared on logout, ensuring reliable and secure authentication across all supported providers (Google, Microsoft, GitHub, OIDC).
+- 🚪 **OAuth Logout and Redirect Reliability**: The OAuth logout process has been made more robust. Logout requests now correctly use proxy environment variables, ensuring they succeed in corporate networks. Additionally, the custom WEBUI_AUTH_SIGNOUT_REDIRECT_URL is now properly respected for all OAuth/OIDC configurations, ensuring a seamless sign-out experience.
+- 📜 **Banner Newline Rendering**: Banners now correctly render newline characters, ensuring that multi-line announcements and messages are displayed with their intended formatting.
+- ℹ️ **Consistent Model Description Rendering**: Model descriptions now render Markdown correctly in the main chat interface, matching the formatting seen in the model selection dropdown for a consistent user experience.
+- 🔄 **Offline Mode Update Check Display**: Corrected a UI bug where the "Checking for Updates..." message would display indefinitely when the application was set to offline mode.
+- 🛠️ **Tool Result Encoding**: Fixed a bug where tool calls returning non-ASCII characters would fail, ensuring robust handling of international text and special characters in tool outputs.
+
+## [0.6.15] - 2025-06-16
+
+### Added
+
+- 🖼️ **Global Image Compression Option**: Effortlessly set image compression globally so all image uploads and outputs are optimized, speeding up load times and saving bandwidth—perfect for teams dealing with large files or limited network resources.
+- 🎤 **Custom Speech-to-Text Content-Type for Transcription**: Define custom content types for audio transcription, ensuring compatibility with diverse audio sources and unlocking smoother, more accurate transcriptions in advanced setups.
+- 🗂️ **LDAP Group Synchronization (Experimental)**: Automatically sync user groups from your LDAP directory directly into Open WebUI for seamless enterprise access management—simplifies identity integration and governance across your organization.
+- 📈 **OpenTelemetry Metrics via OTLP Exporter (Experimental)**: Gain enterprise-grade analytics and monitor your AI usage in real time with experimental OpenTelemetry Metrics support—connect to any OTLP-compatible backend for instant insights into performance, load, and user interactions.
+- 🕰️ **See User Message Timestamps on Hover (Chat Bubble UI)**: Effortlessly check when any user message was sent by hovering over it in Chat Bubble mode—no more switching screens or digging through logs for context.
+- 🗂️ **Leaderboard Sorting Options**: Sort the leaderboard directly in the UI for a clearer, more actionable view of top performers, models, or tools—making analysis and recognition quick and easy for teams.
+- 🏆 **Evaluation Details Modal in Feedbacks and Leaderboard**: Dive deeper with new modals that display detailed evaluation information when reviewing feedbacks and leaderboard rankings—accelerates learning, progress tracking, and quality improvement.
+- 🔄 **Support for Multiple Pages in External Document Loaders**: Effortlessly extract and work with content spanning multiple pages in external documents, giving you complete flexibility for in-depth research and document workflows.
+- 🌐 **New Accessibility Enhancements Across the Interface**: Benefit from significant accessibility improvements—tab navigation, ARIA roles/labels, better high-contrast text/modes, accessible modals, and more—making Open WebUI more usable and equitable for everyone, including those using assistive technologies.
+- ⚡ **Performance & Stability Upgrades Across Frontend and Backend**: Enjoy a smoother, more reliable experience with numerous behind-the-scenes optimizations and refactoring on both frontend and backend—resulting in faster load times, fewer errors, and even greater stability throughout your workflows.
+- 🌏 **Updated and Expanded Localizations**: Enjoy improved, up-to-date translations for Finnish, German (now with model pinning features), Korean, Russian, Simplified Chinese, Spanish, and more—making every interaction smoother, clearer, and more intuitive for international users.
+
+### Fixed
+
+- 🦾 **Ollama Error Messages More Descriptive**: Receive clearer, more actionable error messages when something goes wrong with Ollama models—making troubleshooting and user support faster and more effective.
+- 🌐 **Bypass Webloader Now Works as Expected**: Resolved an issue where the "bypass webloader" feature failed to function correctly, ensuring web search bypasses operate smoothly and reliably for lighter, faster query results.
+- 🔍 **Prevent Redundant Documents in Citation List**: The expanded citation list no longer shows duplicate documents, offering a cleaner, easier-to-digest reference experience when reviewing sources in knowledge and research workflows.
+- 🛡️ **Trusted Header Email Matching is Now Case-Insensitive**: Fixed a critical authentication issue where email case sensitivity could cause secure headers to mismatch, ensuring robust, seamless login and session management in all environments.
+- ⚙️ **Direct Tool Server Input Accepts Empty Strings**: You can now submit direct tool server commands without unexpected errors when passing empty-string values, improving integration and automation efficiency.
+- 📄 **Citation Page Number for Page 1 is Now Displayed**: Corrected an oversight where references for page 1 documents were missing the page number; citations are now always accurate and fully visible.
+- 📒 **Notes Access Restored**: Fixed an issue where some users could not access their notes—everyone can now view and manage their notes reliably, ensuring seamless documentation and workflow continuity.
+- 🛑 **OAuth Callback Double-Slash Issue Resolved**: Fixed rare cases where an extra slash in OAuth callbacks caused failed logins or redirects, making third-party login integrations more reliable.
+
+### Changed
+
+- 🔑 **Dedicated Permission for System Prompts**: System prompt access is now controlled by its own specific permission instead of being grouped with general chat controls, empowering admins with finer-grained management over who can view or modify system prompts for enhanced security and workflow customization.
+- 🛠️ **YouTube Transcript API and python-pptx Updated**: Enjoy better performance, reliability, and broader compatibility thanks to underlying library upgrades—less friction with media-rich and presentation workflows.
+
+### Removed
+
+- 🗑️ **Console Logging Disabled in Production**: All 'console.log' and 'console.debug' statements are now disabled in production, guaranteeing improved security and cleaner browser logs for end users by removing extraneous technical output.
+
+## [0.6.14] - 2025-06-10
+
+### Added
+
+- 🤖 **Automatic "Follow Up" Suggestions**: Open WebUI now intelligently generates actionable "Follow Up" suggestions automatically with each message you send, helping you stay productive and inspired without interrupting your flow; you can always disable this in Settings if you prefer a distraction-free experience.
+- 🧩 **OpenAI-Compatible Embeddings Endpoint**: Introducing a fully OpenAI-style '/api/embeddings' endpoint—now you can plug in OpenAI-style embeddings workflows with zero hassle, making integrations with external tools and platforms seamless and familiar.
+- ↗️ **Model Pinning for Quick Access**: Pin your favorite or most-used models to the sidebar for instant selection—no more scrolling through long model lists; your go-to models are always visible and ready for fast access.
+- 📌 **Selector Model Item Menu**: Each model in the selector now features a menu where you can easily pin/unpin to the sidebar and copy a direct link—simplifying collaboration and staying organized in even the busiest environments.
+- 🛑 **Reliable Stop for Ongoing Chats in Multi-Replica Setups**: Stopping or cancelling an in-progress chat now works reliably even in clustered deployments—ensuring every user can interrupt AI output at any time, no matter your scale.
+- 🧠 **'Think' Parameter for Ollama Models**: Leverage new 'think' parameter support for Ollama—giving you advanced control over AI reasoning process and further tuning model behavior for your unique use cases.
+- 💬 **Picture Description Modes for Docling**: Customize how images are described/extracted by Docling Loader for smarter, more detailed, and workflow-tailored image understanding in your document pipelines.
+- 🛠 **Settings Modal Deep Linking**: Every tab in Settings now has its own route—making direct navigation and sharing of precise settings faster and more intuitive.
+- 🎤 **Audio HTML Component Token**: Easily embed and play audio directly in your chats, improving voice-based workflows and making audio content instantly accessible and manageable from any conversation.
+- 🔑 **Support for Secret Key File**: Now you can specify 'WEBUI_SECRET_KEY_FILE' for more secure and flexible key management—ideal for advanced deployments and tighter security standards.
+- 💡 **Clarity When Cloning Prompts**: Cloned workspace prompts are clearly labelled with "(Clone)" and IDs have "-clone", keeping your prompt library organized and preventing accidental overwrites.
+- 📝 **Dedicated User Role Edit Modal**: Updating user roles now reliably opens a dedicated edit user modal instead of cycling through roles—making it safer and more clear to manage team permissions.
+- 🏞️ **Better Handling & Storage of Interpreter-Generated Images**: Code interpreter-generated images are now centrally stored and reliably loaded from the database or cloud storage, ensuring your artifacts are always available.
+- 🚀 **Pinecone & Vector Search Optimizations**: Applied latest best practices from Pinecone for smarter timeouts, intelligent retry control, improved connection pooling, faster DNS, and concurrent batch handling—giving you more reliable, faster document search and RAG performance without manual tweaks.
+- ⚙️ **Ollama Advanced Parameters Unified**: 'keep_alive' and 'format' options are now integrated into the advanced params section—edit everything from the model editor for flexible model control.
+- 🛠️ **CUDA 12.6 Docker Image Support**: Deploy to NVIDIA GPUs with capability 7.0 and below (e.g., V100, GTX1080) via new cuda126 image—broadening your hardware options for scalable AI workloads.
+- 🔒 **Experimental Table-Level PGVector Data Encryption**: Activate pgcrypto encryption support for pgvector to secure your vector search table contents, giving organizations enhanced compliance and data protection—perfect for enterprise or regulated environments.
+- 👁 **Accessibility Upgrades Across Interface**: Chat buttons and close controls are now labelled and structured for optimal accessibility support, ensuring smoother operation with assistive technologies.
+- 🎨 **High-Contrast Mode Expansions**: High-contrast accessibility mode now also applies to menu items, tabs, and search input fields, offering a more readable experience for all users.
+- 🛠️ **Tooltip & Translation Clarity**: Improved translation and tooltip clarity, especially over radio buttons, making the UI more understandable for all users.
+- 🔠 **Global Localization & Translation Improvements**: Hefty upgrades to Traditional Chinese, Simplified Chinese, Hebrew, Russian, Irish, German, and Danish translation packs—making the platform feel native and intuitive for even more users worldwide.
+- ⚡ **General Backend Stability & Security Enhancements**: Refined numerous backend routines to minimize memory use, improve performance, and streamline integration with external APIs—making the entire platform more robust and secure for daily work.
+
+### Fixed
+
+- 🏷 **Feedback Score Display Improved**: Addressed overflow and visibility issues with feedback scores for more readable and accessible evaluations.
+- 🗂 **Admin Settings Model Edits Apply Immediately**: Changes made in the Model Editor within Admin Settings now take effect instantly, eliminating confusion during model management.
+- 🔄 **Assigned Tools Update Instantly on New Chats**: Models assigned with specific tools now consistently update and are available in every new chat—making tool workflows more predictable and robust.
+- 🛠 **Document Settings Saved Only on User Action**: Document settings now save only when you press the Save button, reducing accidental changes and ensuring greater control.
+- 🔊 **Voice Recording on Older iOS Devices Restored**: Voice input is now fully functional on older iOS devices, keeping voice workflows accessible to all users.
+- 🔒 **Trusted Email Header Session Security**: User sessions now strictly verify the trusted email header matches the logged-in user's email, ensuring secure authentication and preventing accidental session switching.
+- 🔒 **Consistent User Signout on Email Mismatch**: When the trusted email in the header changes, you will now be properly signed out and redirected, safeguarding your session's integrity.
+- 🛠 **General Error & Content Validation Improvements**: Smarter error handling means clearer messages and fewer unnecessary retries—making batch uploads, document handling, and knowledge indexing more resilient.
+- 🕵️ **Better Feedback on Chat Title Edits**: Error messages now show clearly if problems occur while editing chat titles.
+
+## [0.6.13] - 2025-05-30
+
+### Added
+
+- 🟦 **Azure OpenAI Embedding Support**: You can now select Azure OpenAI endpoints for text embeddings, unlocking seamless integration with enterprise-scale Azure AI for powerful RAG and knowledge workflows—no more workarounds, connect and scale effortlessly.
+- 🧩 **Smarter Custom Parameter Handling**: Instantly enjoy more flexible model setup—any JSON pasted into custom parameter fields is now parsed automatically, so you can define rich, nested parameters without tedious manual adjustment. This streamlines advanced configuration for all models and accelerates experimentation.
+- ⚙️ **General Backend Refactoring**: Significant backend improvements deliver a cleaner codebase for better maintainability, faster performance, and even greater platform reliability—making all your workflows run more smoothly.
+- 🌏 **Localization Upgrades**: Experience highly improved user interface translations and clarity in Simplified, Traditional Chinese, Korean, and Finnish, offering a more natural, accurate, and accessible experience for global users.
+
+### Fixed
+
+- 🛡️ **Robust Message Handling on Chat Load**: Fixed an issue where chat pages could fail to load if a referenced message was missing or undefined; now, chats always load smoothly and missing IDs no longer disrupt your workflow.
+- 📝 **Correct Prompt Access Control**: Ensured that the prompt access controls register properly, restoring reliable permissioning and safeguarding your prompt workflows.
+- 🛠 **Open WebUI-Specific Params No Longer Sent to Models**: Fixed a bug that sent internal WebUI parameters to APIs, ensuring only intended model options are transmitted—restoring predictable, error-free model operation.
+- 🧠 **Refined Memory Error Handling**: Enhanced stability during memory-related operations, so even uncommon memory errors are gracefully managed without disrupting your session—resulting in a more reliable, worry-free experience.
+
 ## [0.6.12] - 2025-05-29
 
 ### Added

Dockerfile

@@ -30,7 +30,7 @@ WORKDIR /app
 RUN apk add --no-cache git
 
 COPY package.json package-lock.json ./
-RUN npm ci
+RUN npm ci --force
 
 COPY . .
 ENV APP_BUILD_HASH=${BUILD_HASH}

LICENSE_HISTORY

@@ -0,0 +1,53 @@
+All code and materials created before commit `60d84a3aae9802339705826e9095e272e3c83623` are subject to the following copyright and license:
+
+Copyright (c) 2023-2025 Timothy Jaeryang Baek
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+All code and materials created before commit `a76068d69cd59568b920dfab85dc573dbbb8f131` are subject to the following copyright and license:
+
+MIT License
+
+Copyright (c) 2023 Timothy Jaeryang Baek
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -12,6 +12,8 @@
 
 **Open WebUI is an [extensible](https://docs.openwebui.com/features/plugin/), feature-rich, and user-friendly self-hosted AI platform designed to operate entirely offline.** It supports various LLM runners like **Ollama** and **OpenAI-compatible APIs**, with **built-in inference engine** for RAG, making it a **powerful AI deployment solution**.
 
+Passionate about open-source AI? [Join our team →](https://careers.openwebui.com/)
+
 ![Open WebUI Demo](./demo.gif)
 
 > [!TIP]  
@@ -29,6 +31,8 @@ For more information, be sure to check out our [Open WebUI Documentation](https:
 
 - 🛡️ **Granular Permissions and User Groups**: By allowing administrators to create detailed user roles and permissions, we ensure a secure user environment. This granularity not only enhances security but also allows for customized user experiences, fostering a sense of ownership and responsibility amongst users.
 
+- 🔄 **SCIM 2.0 Support**: Enterprise-grade user and group provisioning through SCIM 2.0 protocol, enabling seamless integration with identity providers like Okta, Azure AD, and Google Workspace for automated user lifecycle management.
+
 - 📱 **Responsive Design**: Enjoy a seamless experience across Desktop PC, Laptop, and Mobile devices.
 
 - 📱 **Progressive Web App (PWA) for Mobile**: Enjoy a native app-like experience on your mobile device with our PWA, providing offline access on localhost and a seamless user interface.
@@ -73,17 +77,17 @@ Want to learn more about Open WebUI's features? Check out our [Open WebUI docume
       </a>
     </td>
     <td>
-      N8N • Does your interface have a backend yet?<br>Try <a href="https://n8n.io/">n8n</a>
+      <a href="https://n8n.io/">n8n</a> • Does your interface have a backend yet?<br>Try <a href="https://n8n.io/">n8n</a>
     </td>
   </tr>
   <tr>
     <td>
-      <a href="https://warp.dev/open-webui" target="_blank">
-        <img src="https://docs.openwebui.com/sponsors/logos/warp.png" alt="n8n" style="width: 8rem; height: 8rem; border-radius: .75rem;" />
+      <a href="https://tailscale.com/blog/self-host-a-local-ai-stack/?utm_source=OpenWebUI&utm_medium=paid-ad-placement&utm_campaign=OpenWebUI-Docs" target="_blank">
+        <img src="https://docs.openwebui.com/sponsors/logos/tailscale.png" alt="Tailscale" style="width: 8rem; height: 8rem; border-radius: .75rem;" />
       </a>
     </td>
     <td>
-      Wrap • The intelligent terminal for developers
+      <a href="https://tailscale.com/blog/self-host-a-local-ai-stack/?utm_source=OpenWebUI&utm_medium=paid-ad-placement&utm_campaign=OpenWebUI-Docs">Tailscale</a> • Connect self-hosted AI to any device with Tailscale
     </td>
   </tr>
 </table>
@@ -181,6 +185,8 @@ After installation, you can access Open WebUI at [http://localhost:3000](http://
 
 We offer various installation alternatives, including non-Docker native installation methods, Docker Compose, Kustomize, and Helm. Visit our [Open WebUI Documentation](https://docs.openwebui.com/getting-started/) or join our [Discord community](https://discord.gg/5rJgQTnV4s) for comprehensive guidance.
 
+Look at the [Local Development Guide](https://docs.openwebui.com/getting-started/advanced-topics/development) for instructions on setting up a local development environment.
+
 ### Troubleshooting
 
 Encountering connection issues? Our [Open WebUI Documentation](https://docs.openwebui.com/troubleshooting/) has got you covered. For further assistance and to join our vibrant community, visit the [Open WebUI Discord](https://discord.gg/5rJgQTnV4s).

backend/dev.sh

@@ -1,2 +1,3 @@
+export CORS_ALLOW_ORIGIN="http://localhost:5173"
 PORT="${PORT:-8080}"
-uvicorn open_webui.main:app --port $PORT --host 0.0.0.0 --forwarded-allow-ips '*' --reload
+uvicorn open_webui.main:app --port $PORT --host 0.0.0.0 --forwarded-allow-ips '*' --reload

backend/open_webui/config.py

@@ -7,18 +7,21 @@ import redis
 
 from datetime import datetime
 from pathlib import Path
-from typing import Generic, Optional, TypeVar
+from typing import Generic, Union, Optional, TypeVar
 from urllib.parse import urlparse
 
 import requests
 from pydantic import BaseModel
 from sqlalchemy import JSON, Column, DateTime, Integer, func
+from authlib.integrations.starlette_client import OAuth
+
 
 from open_webui.env import (
     DATA_DIR,
     DATABASE_URL,
     ENV,
     REDIS_URL,
+    REDIS_KEY_PREFIX,
     REDIS_SENTINEL_HOSTS,
     REDIS_SENTINEL_PORT,
     FRONTEND_BUILD_DIR,
@@ -165,9 +168,19 @@ class PersistentConfig(Generic[T]):
         self.config_path = config_path
         self.env_value = env_value
         self.config_value = get_config_value(config_path)
+
         if self.config_value is not None and ENABLE_PERSISTENT_CONFIG:
-            log.info(f"'{env_name}' loaded from the latest database entry")
-            self.value = self.config_value
+            if (
+                self.config_path.startswith("oauth.")
+                and not ENABLE_OAUTH_PERSISTENT_CONFIG
+            ):
+                log.info(
+                    f"Skipping loading of '{env_name}' as OAuth persistent config is disabled"
+                )
+                self.value = env_value
+            else:
+                log.info(f"'{env_name}' loaded from the latest database entry")
+                self.value = self.config_value
         else:
             self.value = env_value
 
@@ -210,16 +223,27 @@ class PersistentConfig(Generic[T]):
 
 class AppConfig:
     _state: dict[str, PersistentConfig]
-    _redis: Optional[redis.Redis] = None
+    _redis: Union[redis.Redis, redis.cluster.RedisCluster] = None
+    _redis_key_prefix: str
 
     def __init__(
-        self, redis_url: Optional[str] = None, redis_sentinels: Optional[list] = []
+        self,
+        redis_url: Optional[str] = None,
+        redis_sentinels: Optional[list] = [],
+        redis_cluster: Optional[bool] = False,
+        redis_key_prefix: str = "open-webui",
     ):
         super().__setattr__("_state", {})
+        super().__setattr__("_redis_key_prefix", redis_key_prefix)
         if redis_url:
             super().__setattr__(
                 "_redis",
-                get_redis_connection(redis_url, redis_sentinels, decode_responses=True),
+                get_redis_connection(
+                    redis_url,
+                    redis_sentinels,
+                    redis_cluster,
+                    decode_responses=True,
+                ),
             )
 
     def __setattr__(self, key, value):
@@ -230,7 +254,7 @@ class AppConfig:
             self._state[key].save()
 
             if self._redis:
-                redis_key = f"open-webui:config:{key}"
+                redis_key = f"{self._redis_key_prefix}:config:{key}"
                 self._redis.set(redis_key, json.dumps(self._state[key].value))
 
     def __getattr__(self, key):
@@ -239,7 +263,7 @@ class AppConfig:
 
         # If Redis is available, check for an updated value
         if self._redis:
-            redis_key = f"open-webui:config:{key}"
+            redis_key = f"{self._redis_key_prefix}:config:{key}"
             redis_value = self._redis.get(redis_key)
 
             if redis_value is not None:
@@ -288,6 +312,9 @@ JWT_EXPIRES_IN = PersistentConfig(
 # OAuth config
 ####################################
 
+ENABLE_OAUTH_PERSISTENT_CONFIG = (
+    os.environ.get("ENABLE_OAUTH_PERSISTENT_CONFIG", "True").lower() == "true"
+)
 
 ENABLE_OAUTH_SIGNUP = PersistentConfig(
     "ENABLE_OAUTH_SIGNUP",
@@ -347,6 +374,24 @@ MICROSOFT_CLIENT_TENANT_ID = PersistentConfig(
     os.environ.get("MICROSOFT_CLIENT_TENANT_ID", ""),
 )
 
+MICROSOFT_CLIENT_LOGIN_BASE_URL = PersistentConfig(
+    "MICROSOFT_CLIENT_LOGIN_BASE_URL",
+    "oauth.microsoft.login_base_url",
+    os.environ.get(
+        "MICROSOFT_CLIENT_LOGIN_BASE_URL", "https://login.microsoftonline.com"
+    ),
+)
+
+MICROSOFT_CLIENT_PICTURE_URL = PersistentConfig(
+    "MICROSOFT_CLIENT_PICTURE_URL",
+    "oauth.microsoft.picture_url",
+    os.environ.get(
+        "MICROSOFT_CLIENT_PICTURE_URL",
+        "https://graph.microsoft.com/v1.0/me/photo/$value",
+    ),
+)
+
+
 MICROSOFT_OAUTH_SCOPE = PersistentConfig(
     "MICROSOFT_OAUTH_SCOPE",
     "oauth.microsoft.scope",
@@ -413,6 +458,18 @@ OAUTH_SCOPES = PersistentConfig(
     os.environ.get("OAUTH_SCOPES", "openid email profile"),
 )
 
+OAUTH_TIMEOUT = PersistentConfig(
+    "OAUTH_TIMEOUT",
+    "oauth.oidc.oauth_timeout",
+    os.environ.get("OAUTH_TIMEOUT", ""),
+)
+
+OAUTH_TOKEN_ENDPOINT_AUTH_METHOD = PersistentConfig(
+    "OAUTH_TOKEN_ENDPOINT_AUTH_METHOD",
+    "oauth.oidc.token_endpoint_auth_method",
+    os.environ.get("OAUTH_TOKEN_ENDPOINT_AUTH_METHOD", None),
+)
+
 OAUTH_CODE_CHALLENGE_METHOD = PersistentConfig(
     "OAUTH_CODE_CHALLENGE_METHOD",
     "oauth.oidc.code_challenge_method",
@@ -425,6 +482,12 @@ OAUTH_PROVIDER_NAME = PersistentConfig(
     os.environ.get("OAUTH_PROVIDER_NAME", "SSO"),
 )
 
+OAUTH_SUB_CLAIM = PersistentConfig(
+    "OAUTH_SUB_CLAIM",
+    "oauth.oidc.sub_claim",
+    os.environ.get("OAUTH_SUB_CLAIM", None),
+)
+
 OAUTH_USERNAME_CLAIM = PersistentConfig(
     "OAUTH_USERNAME_CLAIM",
     "oauth.oidc.username_claim",
@@ -516,13 +579,20 @@ def load_oauth_providers():
     OAUTH_PROVIDERS.clear()
     if GOOGLE_CLIENT_ID.value and GOOGLE_CLIENT_SECRET.value:
 
-        def google_oauth_register(client):
+        def google_oauth_register(client: OAuth):
             client.register(
                 name="google",
                 client_id=GOOGLE_CLIENT_ID.value,
                 client_secret=GOOGLE_CLIENT_SECRET.value,
                 server_metadata_url="https://accounts.google.com/.well-known/openid-configuration",
-                client_kwargs={"scope": GOOGLE_OAUTH_SCOPE.value},
+                client_kwargs={
+                    "scope": GOOGLE_OAUTH_SCOPE.value,
+                    **(
+                        {"timeout": int(OAUTH_TIMEOUT.value)}
+                        if OAUTH_TIMEOUT.value
+                        else {}
+                    ),
+                },
                 redirect_uri=GOOGLE_REDIRECT_URI.value,
             )
 
@@ -537,27 +607,32 @@ def load_oauth_providers():
         and MICROSOFT_CLIENT_TENANT_ID.value
     ):
 
-        def microsoft_oauth_register(client):
+        def microsoft_oauth_register(client: OAuth):
             client.register(
                 name="microsoft",
                 client_id=MICROSOFT_CLIENT_ID.value,
                 client_secret=MICROSOFT_CLIENT_SECRET.value,
-                server_metadata_url=f"https://login.microsoftonline.com/{MICROSOFT_CLIENT_TENANT_ID.value}/v2.0/.well-known/openid-configuration?appid={MICROSOFT_CLIENT_ID.value}",
+                server_metadata_url=f"{MICROSOFT_CLIENT_LOGIN_BASE_URL.value}/{MICROSOFT_CLIENT_TENANT_ID.value}/v2.0/.well-known/openid-configuration?appid={MICROSOFT_CLIENT_ID.value}",
                 client_kwargs={
                     "scope": MICROSOFT_OAUTH_SCOPE.value,
+                    **(
+                        {"timeout": int(OAUTH_TIMEOUT.value)}
+                        if OAUTH_TIMEOUT.value
+                        else {}
+                    ),
                 },
                 redirect_uri=MICROSOFT_REDIRECT_URI.value,
             )
 
         OAUTH_PROVIDERS["microsoft"] = {
             "redirect_uri": MICROSOFT_REDIRECT_URI.value,
-            "picture_url": "https://graph.microsoft.com/v1.0/me/photo/$value",
+            "picture_url": MICROSOFT_CLIENT_PICTURE_URL.value,
             "register": microsoft_oauth_register,
         }
 
     if GITHUB_CLIENT_ID.value and GITHUB_CLIENT_SECRET.value:
 
-        def github_oauth_register(client):
+        def github_oauth_register(client: OAuth):
             client.register(
                 name="github",
                 client_id=GITHUB_CLIENT_ID.value,
@@ -566,7 +641,14 @@ def load_oauth_providers():
                 authorize_url="https://github.com/login/oauth/authorize",
                 api_base_url="https://api.github.com",
                 userinfo_endpoint="https://api.github.com/user",
-                client_kwargs={"scope": GITHUB_CLIENT_SCOPE.value},
+                client_kwargs={
+                    "scope": GITHUB_CLIENT_SCOPE.value,
+                    **(
+                        {"timeout": int(OAUTH_TIMEOUT.value)}
+                        if OAUTH_TIMEOUT.value
+                        else {}
+                    ),
+                },
                 redirect_uri=GITHUB_CLIENT_REDIRECT_URI.value,
             )
 
@@ -582,9 +664,19 @@ def load_oauth_providers():
         and OPENID_PROVIDER_URL.value
     ):
 
-        def oidc_oauth_register(client):
+        def oidc_oauth_register(client: OAuth):
             client_kwargs = {
                 "scope": OAUTH_SCOPES.value,
+                **(
+                    {
+                        "token_endpoint_auth_method": OAUTH_TOKEN_ENDPOINT_AUTH_METHOD.value
+                    }
+                    if OAUTH_TOKEN_ENDPOINT_AUTH_METHOD.value
+                    else {}
+                ),
+                **(
+                    {"timeout": int(OAUTH_TIMEOUT.value)} if OAUTH_TIMEOUT.value else {}
+                ),
             }
 
             if (
@@ -613,6 +705,23 @@ def load_oauth_providers():
             "register": oidc_oauth_register,
         }
 
+    configured_providers = []
+    if GOOGLE_CLIENT_ID.value:
+        configured_providers.append("Google")
+    if MICROSOFT_CLIENT_ID.value:
+        configured_providers.append("Microsoft")
+    if GITHUB_CLIENT_ID.value:
+        configured_providers.append("GitHub")
+
+    if configured_providers and not OPENID_PROVIDER_URL.value:
+        provider_list = ", ".join(configured_providers)
+        log.warning(
+            f"⚠️  OAuth providers configured ({provider_list}) but OPENID_PROVIDER_URL not set - logout will not work!"
+        )
+        log.warning(
+            f"Set OPENID_PROVIDER_URL to your OAuth provider's OpenID Connect discovery endpoint to fix logout functionality."
+        )
+
 
 load_oauth_providers()
 
@@ -622,6 +731,17 @@ load_oauth_providers()
 
 STATIC_DIR = Path(os.getenv("STATIC_DIR", OPEN_WEBUI_DIR / "static")).resolve()
 
+try:
+    if STATIC_DIR.exists():
+        for item in STATIC_DIR.iterdir():
+            if item.is_file() or item.is_symlink():
+                try:
+                    item.unlink()
+                except Exception as e:
+                    pass
+except Exception as e:
+    pass
+
 for file_path in (FRONTEND_BUILD_DIR / "static").glob("**/*"):
     if file_path.is_file():
         target_path = STATIC_DIR / file_path.relative_to(
@@ -701,12 +821,6 @@ if CUSTOM_NAME:
         pass
 
 
-####################################
-# LICENSE_KEY
-####################################
-
-LICENSE_KEY = os.environ.get("LICENSE_KEY", "")
-
 ####################################
 # STORAGE PROVIDER
 ####################################
@@ -877,6 +991,18 @@ except Exception:
     pass
 OPENAI_API_BASE_URL = "https://api.openai.com/v1"
 
+
+####################################
+# MODELS
+####################################
+
+ENABLE_BASE_MODELS_CACHE = PersistentConfig(
+    "ENABLE_BASE_MODELS_CACHE",
+    "models.base_models_cache",
+    os.environ.get("ENABLE_BASE_MODELS_CACHE", "False").lower() == "true",
+)
+
+
 ####################################
 # TOOL_SERVERS
 ####################################
@@ -901,9 +1027,7 @@ TOOL_SERVER_CONNECTIONS = PersistentConfig(
 ####################################
 
 
-WEBUI_URL = PersistentConfig(
-    "WEBUI_URL", "webui.url", os.environ.get("WEBUI_URL", "http://localhost:3000")
-)
+WEBUI_URL = PersistentConfig("WEBUI_URL", "webui.url", os.environ.get("WEBUI_URL", ""))
 
 
 ENABLE_SIGNUP = PersistentConfig(
@@ -1061,6 +1185,18 @@ USER_PERMISSIONS_CHAT_CONTROLS = (
     os.environ.get("USER_PERMISSIONS_CHAT_CONTROLS", "True").lower() == "true"
 )
 
+USER_PERMISSIONS_CHAT_VALVES = (
+    os.environ.get("USER_PERMISSIONS_CHAT_VALVES", "True").lower() == "true"
+)
+
+USER_PERMISSIONS_CHAT_SYSTEM_PROMPT = (
+    os.environ.get("USER_PERMISSIONS_CHAT_SYSTEM_PROMPT", "True").lower() == "true"
+)
+
+USER_PERMISSIONS_CHAT_PARAMS = (
+    os.environ.get("USER_PERMISSIONS_CHAT_PARAMS", "True").lower() == "true"
+)
+
 USER_PERMISSIONS_CHAT_FILE_UPLOAD = (
     os.environ.get("USER_PERMISSIONS_CHAT_FILE_UPLOAD", "True").lower() == "true"
 )
@@ -1146,6 +1282,9 @@ DEFAULT_USER_PERMISSIONS = {
     },
     "chat": {
         "controls": USER_PERMISSIONS_CHAT_CONTROLS,
+        "valves": USER_PERMISSIONS_CHAT_VALVES,
+        "system_prompt": USER_PERMISSIONS_CHAT_SYSTEM_PROMPT,
+        "params": USER_PERMISSIONS_CHAT_PARAMS,
         "file_upload": USER_PERMISSIONS_CHAT_FILE_UPLOAD,
         "delete": USER_PERMISSIONS_CHAT_DELETE,
         "edit": USER_PERMISSIONS_CHAT_EDIT,
@@ -1212,6 +1351,10 @@ WEBHOOK_URL = PersistentConfig(
 
 ENABLE_ADMIN_EXPORT = os.environ.get("ENABLE_ADMIN_EXPORT", "True").lower() == "true"
 
+ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS = (
+    os.environ.get("ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS", "True").lower() == "true"
+)
+
 ENABLE_ADMIN_CHAT_ACCESS = (
     os.environ.get("ENABLE_ADMIN_CHAT_ACCESS", "True").lower() == "true"
 )
@@ -1247,19 +1390,14 @@ if THREAD_POOL_SIZE is not None and isinstance(THREAD_POOL_SIZE, str):
         THREAD_POOL_SIZE = None
 
 
-def validate_cors_origins(origins):
-    for origin in origins:
-        if origin != "*":
-            validate_cors_origin(origin)
-
-
 def validate_cors_origin(origin):
     parsed_url = urlparse(origin)
 
-    # Check if the scheme is either http or https
-    if parsed_url.scheme not in ["http", "https"]:
+    # Check if the scheme is either http or https, or a custom scheme
+    schemes = ["http", "https"] + CORS_ALLOW_CUSTOM_SCHEME
+    if parsed_url.scheme not in schemes:
         raise ValueError(
-            f"Invalid scheme in CORS_ALLOW_ORIGIN: '{origin}'. Only 'http' and 'https' are allowed."
+            f"Invalid scheme in CORS_ALLOW_ORIGIN: '{origin}'. Only 'http' and 'https' and CORS_ALLOW_CUSTOM_SCHEME are allowed."
         )
 
     # Ensure that the netloc (domain + port) is present, indicating it's a valid URL
@@ -1272,16 +1410,22 @@ def validate_cors_origin(origin):
 # To test CORS_ALLOW_ORIGIN locally, you can set something like
 # CORS_ALLOW_ORIGIN=http://localhost:5173;http://localhost:8080
 # in your .env file depending on your frontend port, 5173 in this case.
-CORS_ALLOW_ORIGIN = os.environ.get(
-    "CORS_ALLOW_ORIGIN", "*;http://localhost:5173;http://localhost:8080"
-).split(";")
+CORS_ALLOW_ORIGIN = os.environ.get("CORS_ALLOW_ORIGIN", "*").split(";")
 
-if "*" in CORS_ALLOW_ORIGIN:
+# Allows custom URL schemes (e.g., app://) to be used as origins for CORS.
+# Useful for local development or desktop clients with schemes like app:// or other custom protocols.
+# Provide a semicolon-separated list of allowed schemes in the environment variable CORS_ALLOW_CUSTOM_SCHEMES.
+CORS_ALLOW_CUSTOM_SCHEME = os.environ.get("CORS_ALLOW_CUSTOM_SCHEME", "").split(";")
+
+if CORS_ALLOW_ORIGIN == ["*"]:
     log.warning(
         "\n\nWARNING: CORS_ALLOW_ORIGIN IS SET TO '*' - NOT RECOMMENDED FOR PRODUCTION DEPLOYMENTS.\n"
     )
-
-validate_cors_origins(CORS_ALLOW_ORIGIN)
+else:
+    # You have to pick between a single wildcard or a list of origins.
+    # Doing both will result in CORS errors in the browser.
+    for origin in CORS_ALLOW_ORIGIN:
+        validate_cors_origin(origin)
 
 
 class BannerModel(BaseModel):
@@ -1413,6 +1557,35 @@ Strictly return in JSON format:
 {{MESSAGES:END:6}}
 </chat_history>"""
 
+
+FOLLOW_UP_GENERATION_PROMPT_TEMPLATE = PersistentConfig(
+    "FOLLOW_UP_GENERATION_PROMPT_TEMPLATE",
+    "task.follow_up.prompt_template",
+    os.environ.get("FOLLOW_UP_GENERATION_PROMPT_TEMPLATE", ""),
+)
+
+DEFAULT_FOLLOW_UP_GENERATION_PROMPT_TEMPLATE = """### Task:
+Suggest 3-5 relevant follow-up questions or prompts that the user might naturally ask next in this conversation as a **user**, based on the chat history, to help continue or deepen the discussion.
+### Guidelines:
+- Write all follow-up questions from the user’s point of view, directed to the assistant.
+- Make questions concise, clear, and directly related to the discussed topic(s).
+- Only suggest follow-ups that make sense given the chat content and do not repeat what was already covered.
+- If the conversation is very short or not specific, suggest more general (but relevant) follow-ups the user might ask.
+- Use the conversation's primary language; default to English if multilingual.
+- Response must be a JSON array of strings, no extra text or formatting.
+### Output:
+JSON format: { "follow_ups": ["Question 1?", "Question 2?", "Question 3?"] }
+### Chat History:
+<chat_history>
+{{MESSAGES:END:6}}
+</chat_history>"""
+
+ENABLE_FOLLOW_UP_GENERATION = PersistentConfig(
+    "ENABLE_FOLLOW_UP_GENERATION",
+    "task.follow_up.enable",
+    os.environ.get("ENABLE_FOLLOW_UP_GENERATION", "True").lower() == "true",
+)
+
 ENABLE_TAGS_GENERATION = PersistentConfig(
     "ENABLE_TAGS_GENERATION",
     "task.tags.enable",
@@ -1749,11 +1922,14 @@ MILVUS_IVF_FLAT_NLIST = int(os.environ.get("MILVUS_IVF_FLAT_NLIST", "128"))
 QDRANT_URI = os.environ.get("QDRANT_URI", None)
 QDRANT_API_KEY = os.environ.get("QDRANT_API_KEY", None)
 QDRANT_ON_DISK = os.environ.get("QDRANT_ON_DISK", "false").lower() == "true"
-QDRANT_PREFER_GRPC = os.environ.get("QDRANT_PREFER_GRPC", "False").lower() == "true"
+QDRANT_PREFER_GRPC = os.environ.get("QDRANT_PREFER_GRPC", "false").lower() == "true"
 QDRANT_GRPC_PORT = int(os.environ.get("QDRANT_GRPC_PORT", "6334"))
+QDRANT_TIMEOUT = int(os.environ.get("QDRANT_TIMEOUT", "5"))
+QDRANT_HNSW_M = int(os.environ.get("QDRANT_HNSW_M", "16"))
 ENABLE_QDRANT_MULTITENANCY_MODE = (
-    os.environ.get("ENABLE_QDRANT_MULTITENANCY_MODE", "false").lower() == "true"
+    os.environ.get("ENABLE_QDRANT_MULTITENANCY_MODE", "true").lower() == "true"
 )
+QDRANT_COLLECTION_PREFIX = os.environ.get("QDRANT_COLLECTION_PREFIX", "open-webui")
 
 # OpenSearch
 OPENSEARCH_URI = os.environ.get("OPENSEARCH_URI", "https://localhost:9200")
@@ -1785,6 +1961,52 @@ PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH = int(
     os.environ.get("PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH", "1536")
 )
 
+PGVECTOR_PGCRYPTO = os.getenv("PGVECTOR_PGCRYPTO", "false").lower() == "true"
+PGVECTOR_PGCRYPTO_KEY = os.getenv("PGVECTOR_PGCRYPTO_KEY", None)
+if PGVECTOR_PGCRYPTO and not PGVECTOR_PGCRYPTO_KEY:
+    raise ValueError(
+        "PGVECTOR_PGCRYPTO is enabled but PGVECTOR_PGCRYPTO_KEY is not set. Please provide a valid key."
+    )
+
+
+PGVECTOR_POOL_SIZE = os.environ.get("PGVECTOR_POOL_SIZE", None)
+
+if PGVECTOR_POOL_SIZE != None:
+    try:
+        PGVECTOR_POOL_SIZE = int(PGVECTOR_POOL_SIZE)
+    except Exception:
+        PGVECTOR_POOL_SIZE = None
+
+PGVECTOR_POOL_MAX_OVERFLOW = os.environ.get("PGVECTOR_POOL_MAX_OVERFLOW", 0)
+
+if PGVECTOR_POOL_MAX_OVERFLOW == "":
+    PGVECTOR_POOL_MAX_OVERFLOW = 0
+else:
+    try:
+        PGVECTOR_POOL_MAX_OVERFLOW = int(PGVECTOR_POOL_MAX_OVERFLOW)
+    except Exception:
+        PGVECTOR_POOL_MAX_OVERFLOW = 0
+
+PGVECTOR_POOL_TIMEOUT = os.environ.get("PGVECTOR_POOL_TIMEOUT", 30)
+
+if PGVECTOR_POOL_TIMEOUT == "":
+    PGVECTOR_POOL_TIMEOUT = 30
+else:
+    try:
+        PGVECTOR_POOL_TIMEOUT = int(PGVECTOR_POOL_TIMEOUT)
+    except Exception:
+        PGVECTOR_POOL_TIMEOUT = 30
+
+PGVECTOR_POOL_RECYCLE = os.environ.get("PGVECTOR_POOL_RECYCLE", 3600)
+
+if PGVECTOR_POOL_RECYCLE == "":
+    PGVECTOR_POOL_RECYCLE = 3600
+else:
+    try:
+        PGVECTOR_POOL_RECYCLE = int(PGVECTOR_POOL_RECYCLE)
+    except Exception:
+        PGVECTOR_POOL_RECYCLE = 3600
+
 # Pinecone
 PINECONE_API_KEY = os.environ.get("PINECONE_API_KEY", None)
 PINECONE_ENVIRONMENT = os.environ.get("PINECONE_ENVIRONMENT", None)
@@ -1793,6 +2015,37 @@ PINECONE_DIMENSION = int(os.getenv("PINECONE_DIMENSION", 1536))  # or 3072, 1024
 PINECONE_METRIC = os.getenv("PINECONE_METRIC", "cosine")
 PINECONE_CLOUD = os.getenv("PINECONE_CLOUD", "aws")  # or "gcp" or "azure"
 
+# ORACLE23AI (Oracle23ai Vector Search)
+
+ORACLE_DB_USE_WALLET = os.environ.get("ORACLE_DB_USE_WALLET", "false").lower() == "true"
+ORACLE_DB_USER = os.environ.get("ORACLE_DB_USER", None)  #
+ORACLE_DB_PASSWORD = os.environ.get("ORACLE_DB_PASSWORD", None)  #
+ORACLE_DB_DSN = os.environ.get("ORACLE_DB_DSN", None)  #
+ORACLE_WALLET_DIR = os.environ.get("ORACLE_WALLET_DIR", None)
+ORACLE_WALLET_PASSWORD = os.environ.get("ORACLE_WALLET_PASSWORD", None)
+ORACLE_VECTOR_LENGTH = os.environ.get("ORACLE_VECTOR_LENGTH", 768)
+
+ORACLE_DB_POOL_MIN = int(os.environ.get("ORACLE_DB_POOL_MIN", 2))
+ORACLE_DB_POOL_MAX = int(os.environ.get("ORACLE_DB_POOL_MAX", 10))
+ORACLE_DB_POOL_INCREMENT = int(os.environ.get("ORACLE_DB_POOL_INCREMENT", 1))
+
+
+if VECTOR_DB == "oracle23ai":
+    if not ORACLE_DB_USER or not ORACLE_DB_PASSWORD or not ORACLE_DB_DSN:
+        raise ValueError(
+            "Oracle23ai requires setting ORACLE_DB_USER, ORACLE_DB_PASSWORD, and ORACLE_DB_DSN."
+        )
+    if ORACLE_DB_USE_WALLET and (not ORACLE_WALLET_DIR or not ORACLE_WALLET_PASSWORD):
+        raise ValueError(
+            "Oracle23ai requires setting ORACLE_WALLET_DIR and ORACLE_WALLET_PASSWORD when using wallet authentication."
+        )
+
+log.info(f"VECTOR_DB: {VECTOR_DB}")
+
+# S3 Vector
+S3_VECTOR_BUCKET_NAME = os.environ.get("S3_VECTOR_BUCKET_NAME", None)
+S3_VECTOR_REGION = os.environ.get("S3_VECTOR_REGION", None)
+
 ####################################
 # Information Retrieval (RAG)
 ####################################
@@ -1854,10 +2107,16 @@ DATALAB_MARKER_API_KEY = PersistentConfig(
     os.environ.get("DATALAB_MARKER_API_KEY", ""),
 )
 
-DATALAB_MARKER_LANGS = PersistentConfig(
-    "DATALAB_MARKER_LANGS",
-    "rag.datalab_marker_langs",
-    os.environ.get("DATALAB_MARKER_LANGS", ""),
+DATALAB_MARKER_API_BASE_URL = PersistentConfig(
+    "DATALAB_MARKER_API_BASE_URL",
+    "rag.datalab_marker_api_base_url",
+    os.environ.get("DATALAB_MARKER_API_BASE_URL", ""),
+)
+
+DATALAB_MARKER_ADDITIONAL_CONFIG = PersistentConfig(
+    "DATALAB_MARKER_ADDITIONAL_CONFIG",
+    "rag.datalab_marker_additional_config",
+    os.environ.get("DATALAB_MARKER_ADDITIONAL_CONFIG", ""),
 )
 
 DATALAB_MARKER_USE_LLM = PersistentConfig(
@@ -1897,6 +2156,12 @@ DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = PersistentConfig(
     == "true",
 )
 
+DATALAB_MARKER_FORMAT_LINES = PersistentConfig(
+    "DATALAB_MARKER_FORMAT_LINES",
+    "rag.datalab_marker_format_lines",
+    os.environ.get("DATALAB_MARKER_FORMAT_LINES", "false").lower() == "true",
+)
+
 DATALAB_MARKER_OUTPUT_FORMAT = PersistentConfig(
     "DATALAB_MARKER_OUTPUT_FORMAT",
     "rag.datalab_marker_output_format",
@@ -1945,6 +2210,40 @@ DOCLING_DO_PICTURE_DESCRIPTION = PersistentConfig(
     os.getenv("DOCLING_DO_PICTURE_DESCRIPTION", "False").lower() == "true",
 )
 
+DOCLING_PICTURE_DESCRIPTION_MODE = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_MODE",
+    "rag.docling_picture_description_mode",
+    os.getenv("DOCLING_PICTURE_DESCRIPTION_MODE", ""),
+)
+
+
+docling_picture_description_local = os.getenv("DOCLING_PICTURE_DESCRIPTION_LOCAL", "")
+try:
+    docling_picture_description_local = json.loads(docling_picture_description_local)
+except json.JSONDecodeError:
+    docling_picture_description_local = {}
+
+
+DOCLING_PICTURE_DESCRIPTION_LOCAL = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_LOCAL",
+    "rag.docling_picture_description_local",
+    docling_picture_description_local,
+)
+
+docling_picture_description_api = os.getenv("DOCLING_PICTURE_DESCRIPTION_API", "")
+try:
+    docling_picture_description_api = json.loads(docling_picture_description_api)
+except json.JSONDecodeError:
+    docling_picture_description_api = {}
+
+
+DOCLING_PICTURE_DESCRIPTION_API = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_API",
+    "rag.docling_picture_description_api",
+    docling_picture_description_api,
+)
+
+
 DOCUMENT_INTELLIGENCE_ENDPOINT = PersistentConfig(
     "DOCUMENT_INTELLIGENCE_ENDPOINT",
     "rag.document_intelligence_endpoint",
@@ -2021,6 +2320,27 @@ RAG_FILE_MAX_SIZE = PersistentConfig(
     ),
 )
 
+FILE_IMAGE_COMPRESSION_WIDTH = PersistentConfig(
+    "FILE_IMAGE_COMPRESSION_WIDTH",
+    "file.image_compression_width",
+    (
+        int(os.environ.get("FILE_IMAGE_COMPRESSION_WIDTH"))
+        if os.environ.get("FILE_IMAGE_COMPRESSION_WIDTH")
+        else None
+    ),
+)
+
+FILE_IMAGE_COMPRESSION_HEIGHT = PersistentConfig(
+    "FILE_IMAGE_COMPRESSION_HEIGHT",
+    "file.image_compression_height",
+    (
+        int(os.environ.get("FILE_IMAGE_COMPRESSION_HEIGHT"))
+        if os.environ.get("FILE_IMAGE_COMPRESSION_HEIGHT")
+        else None
+    ),
+)
+
+
 RAG_ALLOWED_FILE_EXTENSIONS = PersistentConfig(
     "RAG_ALLOWED_FILE_EXTENSIONS",
     "rag.file.allowed_extensions",
@@ -2184,6 +2504,22 @@ RAG_OPENAI_API_KEY = PersistentConfig(
     os.getenv("RAG_OPENAI_API_KEY", OPENAI_API_KEY),
 )
 
+RAG_AZURE_OPENAI_BASE_URL = PersistentConfig(
+    "RAG_AZURE_OPENAI_BASE_URL",
+    "rag.azure_openai.base_url",
+    os.getenv("RAG_AZURE_OPENAI_BASE_URL", ""),
+)
+RAG_AZURE_OPENAI_API_KEY = PersistentConfig(
+    "RAG_AZURE_OPENAI_API_KEY",
+    "rag.azure_openai.api_key",
+    os.getenv("RAG_AZURE_OPENAI_API_KEY", ""),
+)
+RAG_AZURE_OPENAI_API_VERSION = PersistentConfig(
+    "RAG_AZURE_OPENAI_API_VERSION",
+    "rag.azure_openai.api_version",
+    os.getenv("RAG_AZURE_OPENAI_API_VERSION", ""),
+)
+
 RAG_OLLAMA_BASE_URL = PersistentConfig(
     "RAG_OLLAMA_BASE_URL",
     "rag.ollama.url",
@@ -2428,6 +2764,18 @@ PERPLEXITY_API_KEY = PersistentConfig(
     os.getenv("PERPLEXITY_API_KEY", ""),
 )
 
+PERPLEXITY_MODEL = PersistentConfig(
+    "PERPLEXITY_MODEL",
+    "rag.web.search.perplexity_model",
+    os.getenv("PERPLEXITY_MODEL", "sonar"),
+)
+
+PERPLEXITY_SEARCH_CONTEXT_USAGE = PersistentConfig(
+    "PERPLEXITY_SEARCH_CONTEXT_USAGE",
+    "rag.web.search.perplexity_search_context_usage",
+    os.getenv("PERPLEXITY_SEARCH_CONTEXT_USAGE", "medium"),
+)
+
 SOUGOU_API_SID = PersistentConfig(
     "SOUGOU_API_SID",
     "rag.web.search.sougou_api_sid",
@@ -2792,6 +3140,18 @@ AUDIO_STT_MODEL = PersistentConfig(
     os.getenv("AUDIO_STT_MODEL", ""),
 )
 
+AUDIO_STT_SUPPORTED_CONTENT_TYPES = PersistentConfig(
+    "AUDIO_STT_SUPPORTED_CONTENT_TYPES",
+    "audio.stt.supported_content_types",
+    [
+        content_type.strip()
+        for content_type in os.environ.get(
+            "AUDIO_STT_SUPPORTED_CONTENT_TYPES", ""
+        ).split(",")
+        if content_type.strip()
+    ],
+)
+
 AUDIO_STT_AZURE_API_KEY = PersistentConfig(
     "AUDIO_STT_AZURE_API_KEY",
     "audio.stt.azure.api_key",
@@ -2966,3 +3326,22 @@ LDAP_VALIDATE_CERT = PersistentConfig(
 LDAP_CIPHERS = PersistentConfig(
     "LDAP_CIPHERS", "ldap.server.ciphers", os.environ.get("LDAP_CIPHERS", "ALL")
 )
+
+# For LDAP Group Management
+ENABLE_LDAP_GROUP_MANAGEMENT = PersistentConfig(
+    "ENABLE_LDAP_GROUP_MANAGEMENT",
+    "ldap.group.enable_management",
+    os.environ.get("ENABLE_LDAP_GROUP_MANAGEMENT", "False").lower() == "true",
+)
+
+ENABLE_LDAP_GROUP_CREATION = PersistentConfig(
+    "ENABLE_LDAP_GROUP_CREATION",
+    "ldap.group.enable_creation",
+    os.environ.get("ENABLE_LDAP_GROUP_CREATION", "False").lower() == "true",
+)
+
+LDAP_ATTRIBUTE_FOR_GROUPS = PersistentConfig(
+    "LDAP_ATTRIBUTE_FOR_GROUPS",
+    "ldap.server.attribute_for_groups",
+    os.environ.get("LDAP_ATTRIBUTE_FOR_GROUPS", "memberOf"),
+)

backend/open_webui/constants.py

@@ -111,6 +111,7 @@ class TASKS(str, Enum):
 
     DEFAULT = lambda task="": f"{task if task else 'generation'}"
     TITLE_GENERATION = "title_generation"
+    FOLLOW_UP_GENERATION = "follow_up_generation"
     TAGS_GENERATION = "tags_generation"
     EMOJI_GENERATION = "emoji_generation"
     QUERY_GENERATION = "query_generation"

backend/open_webui/env.py

@@ -5,7 +5,9 @@ import os
 import pkgutil
 import sys
 import shutil
+from uuid import uuid4
 from pathlib import Path
+from cryptography.hazmat.primitives import serialization
 
 import markdown
 from bs4 import BeautifulSoup
@@ -130,6 +132,7 @@ else:
         PACKAGE_DATA = {"version": "0.0.0"}
 
 VERSION = PACKAGE_DATA["version"]
+INSTANCE_ID = os.environ.get("INSTANCE_ID", str(uuid4()))
 
 
 # Function to parse each section
@@ -197,6 +200,7 @@ CHANGELOG = changelog_json
 
 SAFE_MODE = os.environ.get("SAFE_MODE", "false").lower() == "true"
 
+
 ####################################
 # ENABLE_FORWARD_USER_INFO_HEADERS
 ####################################
@@ -264,21 +268,43 @@ else:
 
 DATABASE_URL = os.environ.get("DATABASE_URL", f"sqlite:///{DATA_DIR}/webui.db")
 
+DATABASE_TYPE = os.environ.get("DATABASE_TYPE")
+DATABASE_USER = os.environ.get("DATABASE_USER")
+DATABASE_PASSWORD = os.environ.get("DATABASE_PASSWORD")
+
+DATABASE_CRED = ""
+if DATABASE_USER:
+    DATABASE_CRED += f"{DATABASE_USER}"
+if DATABASE_PASSWORD:
+    DATABASE_CRED += f":{DATABASE_PASSWORD}"
+
+DB_VARS = {
+    "db_type": DATABASE_TYPE,
+    "db_cred": DATABASE_CRED,
+    "db_host": os.environ.get("DATABASE_HOST"),
+    "db_port": os.environ.get("DATABASE_PORT"),
+    "db_name": os.environ.get("DATABASE_NAME"),
+}
+
+if all(DB_VARS.values()):
+    DATABASE_URL = f"{DB_VARS['db_type']}://{DB_VARS['db_cred']}@{DB_VARS['db_host']}:{DB_VARS['db_port']}/{DB_VARS['db_name']}"
+elif DATABASE_TYPE == "sqlite+sqlcipher" and not os.environ.get("DATABASE_URL"):
+    # Handle SQLCipher with local file when DATABASE_URL wasn't explicitly set
+    DATABASE_URL = f"sqlite+sqlcipher:///{DATA_DIR}/webui.db"
+
 # Replace the postgres:// with postgresql://
 if "postgres://" in DATABASE_URL:
     DATABASE_URL = DATABASE_URL.replace("postgres://", "postgresql://")
 
 DATABASE_SCHEMA = os.environ.get("DATABASE_SCHEMA", None)
 
-DATABASE_POOL_SIZE = os.environ.get("DATABASE_POOL_SIZE", 0)
+DATABASE_POOL_SIZE = os.environ.get("DATABASE_POOL_SIZE", None)
 
-if DATABASE_POOL_SIZE == "":
-    DATABASE_POOL_SIZE = 0
-else:
+if DATABASE_POOL_SIZE != None:
     try:
         DATABASE_POOL_SIZE = int(DATABASE_POOL_SIZE)
     except Exception:
-        DATABASE_POOL_SIZE = 0
+        DATABASE_POOL_SIZE = None
 
 DATABASE_POOL_MAX_OVERFLOW = os.environ.get("DATABASE_POOL_MAX_OVERFLOW", 0)
 
@@ -323,9 +349,22 @@ ENABLE_REALTIME_CHAT_SAVE = (
 ####################################
 
 REDIS_URL = os.environ.get("REDIS_URL", "")
+REDIS_CLUSTER = os.environ.get("REDIS_CLUSTER", "False").lower() == "true"
+
+REDIS_KEY_PREFIX = os.environ.get("REDIS_KEY_PREFIX", "open-webui")
+
 REDIS_SENTINEL_HOSTS = os.environ.get("REDIS_SENTINEL_HOSTS", "")
 REDIS_SENTINEL_PORT = os.environ.get("REDIS_SENTINEL_PORT", "26379")
 
+# Maximum number of retries for Redis operations when using Sentinel fail-over
+REDIS_SENTINEL_MAX_RETRY_COUNT = os.environ.get("REDIS_SENTINEL_MAX_RETRY_COUNT", "2")
+try:
+    REDIS_SENTINEL_MAX_RETRY_COUNT = int(REDIS_SENTINEL_MAX_RETRY_COUNT)
+    if REDIS_SENTINEL_MAX_RETRY_COUNT < 1:
+        REDIS_SENTINEL_MAX_RETRY_COUNT = 2
+except ValueError:
+    REDIS_SENTINEL_MAX_RETRY_COUNT = 2
+
 ####################################
 # UVICORN WORKERS
 ####################################
@@ -345,6 +384,10 @@ except ValueError:
 ####################################
 
 WEBUI_AUTH = os.environ.get("WEBUI_AUTH", "True").lower() == "true"
+ENABLE_SIGNUP_PASSWORD_CONFIRMATION = (
+    os.environ.get("ENABLE_SIGNUP_PASSWORD_CONFIRMATION", "False").lower() == "true"
+)
+
 WEBUI_AUTH_TRUSTED_EMAIL_HEADER = os.environ.get(
     "WEBUI_AUTH_TRUSTED_EMAIL_HEADER", None
 )
@@ -394,19 +437,105 @@ WEBUI_AUTH_COOKIE_SECURE = (
 if WEBUI_AUTH and WEBUI_SECRET_KEY == "":
     raise ValueError(ERROR_MESSAGES.ENV_VAR_NOT_FOUND)
 
+ENABLE_COMPRESSION_MIDDLEWARE = (
+    os.environ.get("ENABLE_COMPRESSION_MIDDLEWARE", "True").lower() == "true"
+)
+
+
+####################################
+# SCIM Configuration
+####################################
+
+SCIM_ENABLED = os.environ.get("SCIM_ENABLED", "False").lower() == "true"
+SCIM_TOKEN = os.environ.get("SCIM_TOKEN", "")
+
+####################################
+# LICENSE_KEY
+####################################
+
+LICENSE_KEY = os.environ.get("LICENSE_KEY", "")
+
+LICENSE_BLOB = None
+LICENSE_BLOB_PATH = os.environ.get("LICENSE_BLOB_PATH", DATA_DIR / "l.data")
+if LICENSE_BLOB_PATH and os.path.exists(LICENSE_BLOB_PATH):
+    with open(LICENSE_BLOB_PATH, "rb") as f:
+        LICENSE_BLOB = f.read()
+
+LICENSE_PUBLIC_KEY = os.environ.get("LICENSE_PUBLIC_KEY", "")
+
+pk = None
+if LICENSE_PUBLIC_KEY:
+    pk = serialization.load_pem_public_key(
+        f"""
+-----BEGIN PUBLIC KEY-----
+{LICENSE_PUBLIC_KEY}
+-----END PUBLIC KEY-----
+""".encode(
+            "utf-8"
+        )
+    )
+
+
+####################################
+# MODELS
+####################################
+
+MODELS_CACHE_TTL = os.environ.get("MODELS_CACHE_TTL", "1")
+if MODELS_CACHE_TTL == "":
+    MODELS_CACHE_TTL = None
+else:
+    try:
+        MODELS_CACHE_TTL = int(MODELS_CACHE_TTL)
+    except Exception:
+        MODELS_CACHE_TTL = 1
+
+
+####################################
+# CHAT
+####################################
+
+CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE = os.environ.get(
+    "CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE", "1"
+)
+
+if CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE == "":
+    CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE = 1
+else:
+    try:
+        CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE = int(
+            CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE
+        )
+    except Exception:
+        CHAT_RESPONSE_STREAM_DELTA_CHUNK_SIZE = 1
+
+
+####################################
+# WEBSOCKET SUPPORT
+####################################
+
 ENABLE_WEBSOCKET_SUPPORT = (
     os.environ.get("ENABLE_WEBSOCKET_SUPPORT", "True").lower() == "true"
 )
 
+
 WEBSOCKET_MANAGER = os.environ.get("WEBSOCKET_MANAGER", "")
 
 WEBSOCKET_REDIS_URL = os.environ.get("WEBSOCKET_REDIS_URL", REDIS_URL)
-WEBSOCKET_REDIS_LOCK_TIMEOUT = os.environ.get("WEBSOCKET_REDIS_LOCK_TIMEOUT", 60)
+WEBSOCKET_REDIS_CLUSTER = (
+    os.environ.get("WEBSOCKET_REDIS_CLUSTER", str(REDIS_CLUSTER)).lower() == "true"
+)
+
+websocket_redis_lock_timeout = os.environ.get("WEBSOCKET_REDIS_LOCK_TIMEOUT", "60")
+
+try:
+    WEBSOCKET_REDIS_LOCK_TIMEOUT = int(websocket_redis_lock_timeout)
+except ValueError:
+    WEBSOCKET_REDIS_LOCK_TIMEOUT = 60
 
 WEBSOCKET_SENTINEL_HOSTS = os.environ.get("WEBSOCKET_SENTINEL_HOSTS", "")
-
 WEBSOCKET_SENTINEL_PORT = os.environ.get("WEBSOCKET_SENTINEL_PORT", "26379")
 
+
 AIOHTTP_CLIENT_TIMEOUT = os.environ.get("AIOHTTP_CLIENT_TIMEOUT", "")
 
 if AIOHTTP_CLIENT_TIMEOUT == "":
@@ -504,11 +633,14 @@ else:
 # OFFLINE_MODE
 ####################################
 
+ENABLE_VERSION_UPDATE_CHECK = (
+    os.environ.get("ENABLE_VERSION_UPDATE_CHECK", "true").lower() == "true"
+)
 OFFLINE_MODE = os.environ.get("OFFLINE_MODE", "false").lower() == "true"
 
 if OFFLINE_MODE:
     os.environ["HF_HUB_OFFLINE"] = "1"
-
+    ENABLE_VERSION_UPDATE_CHECK = False
 
 ####################################
 # AUDIT LOGGING
@@ -517,6 +649,14 @@ if OFFLINE_MODE:
 AUDIT_LOGS_FILE_PATH = f"{DATA_DIR}/audit.log"
 # Maximum size of a file before rotating into a new log file
 AUDIT_LOG_FILE_ROTATION_SIZE = os.getenv("AUDIT_LOG_FILE_ROTATION_SIZE", "10MB")
+
+# Comma separated list of logger names to use for audit logging
+# Default is "uvicorn.access" which is the access log for Uvicorn
+# You can add more logger names to this list if you want to capture more logs
+AUDIT_UVICORN_LOGGER_NAMES = os.getenv(
+    "AUDIT_UVICORN_LOGGER_NAMES", "uvicorn.access"
+).split(",")
+
 # METADATA | REQUEST | REQUEST_RESPONSE
 AUDIT_LOG_LEVEL = os.getenv("AUDIT_LOG_LEVEL", "NONE").upper()
 try:
@@ -537,9 +677,27 @@ AUDIT_EXCLUDED_PATHS = [path.lstrip("/") for path in AUDIT_EXCLUDED_PATHS]
 ####################################
 
 ENABLE_OTEL = os.environ.get("ENABLE_OTEL", "False").lower() == "true"
+ENABLE_OTEL_METRICS = os.environ.get("ENABLE_OTEL_METRICS", "False").lower() == "true"
+ENABLE_OTEL_LOGS = os.environ.get("ENABLE_OTEL_LOGS", "False").lower() == "true"
+
 OTEL_EXPORTER_OTLP_ENDPOINT = os.environ.get(
     "OTEL_EXPORTER_OTLP_ENDPOINT", "http://localhost:4317"
 )
+OTEL_METRICS_EXPORTER_OTLP_ENDPOINT = os.environ.get(
+    "OTEL_METRICS_EXPORTER_OTLP_ENDPOINT", OTEL_EXPORTER_OTLP_ENDPOINT
+)
+OTEL_LOGS_EXPORTER_OTLP_ENDPOINT = os.environ.get(
+    "OTEL_LOGS_EXPORTER_OTLP_ENDPOINT", OTEL_EXPORTER_OTLP_ENDPOINT
+)
+OTEL_EXPORTER_OTLP_INSECURE = (
+    os.environ.get("OTEL_EXPORTER_OTLP_INSECURE", "False").lower() == "true"
+)
+OTEL_METRICS_EXPORTER_OTLP_INSECURE = (
+    os.environ.get("OTEL_METRICS_EXPORTER_OTLP_INSECURE", "False").lower() == "true"
+)
+OTEL_LOGS_EXPORTER_OTLP_INSECURE = (
+    os.environ.get("OTEL_LOGS_EXPORTER_OTLP_INSECURE", "False").lower() == "true"
+)
 OTEL_SERVICE_NAME = os.environ.get("OTEL_SERVICE_NAME", "open-webui")
 OTEL_RESOURCE_ATTRIBUTES = os.environ.get(
     "OTEL_RESOURCE_ATTRIBUTES", ""
@@ -547,6 +705,33 @@ OTEL_RESOURCE_ATTRIBUTES = os.environ.get(
 OTEL_TRACES_SAMPLER = os.environ.get(
     "OTEL_TRACES_SAMPLER", "parentbased_always_on"
 ).lower()
+OTEL_BASIC_AUTH_USERNAME = os.environ.get("OTEL_BASIC_AUTH_USERNAME", "")
+OTEL_BASIC_AUTH_PASSWORD = os.environ.get("OTEL_BASIC_AUTH_PASSWORD", "")
+
+OTEL_METRICS_BASIC_AUTH_USERNAME = os.environ.get(
+    "OTEL_METRICS_BASIC_AUTH_USERNAME", OTEL_BASIC_AUTH_USERNAME
+)
+OTEL_METRICS_BASIC_AUTH_PASSWORD = os.environ.get(
+    "OTEL_METRICS_BASIC_AUTH_PASSWORD", OTEL_BASIC_AUTH_PASSWORD
+)
+OTEL_LOGS_BASIC_AUTH_USERNAME = os.environ.get(
+    "OTEL_LOGS_BASIC_AUTH_USERNAME", OTEL_BASIC_AUTH_USERNAME
+)
+OTEL_LOGS_BASIC_AUTH_PASSWORD = os.environ.get(
+    "OTEL_LOGS_BASIC_AUTH_PASSWORD", OTEL_BASIC_AUTH_PASSWORD
+)
+
+OTEL_OTLP_SPAN_EXPORTER = os.environ.get(
+    "OTEL_OTLP_SPAN_EXPORTER", "grpc"
+).lower()  # grpc or http
+
+OTEL_METRICS_OTLP_SPAN_EXPORTER = os.environ.get(
+    "OTEL_METRICS_OTLP_SPAN_EXPORTER", OTEL_OTLP_SPAN_EXPORTER
+).lower()  # grpc or http
+
+OTEL_LOGS_OTLP_SPAN_EXPORTER = os.environ.get(
+    "OTEL_LOGS_OTLP_SPAN_EXPORTER", OTEL_OTLP_SPAN_EXPORTER
+).lower()  # grpc or http
 
 ####################################
 # TOOLS/FUNCTIONS PIP OPTIONS

backend/open_webui/functions.py

@@ -25,6 +25,7 @@ from open_webui.socket.main import (
 )
 
 
+from open_webui.models.users import UserModel
 from open_webui.models.functions import Functions
 from open_webui.models.models import Models
 
@@ -227,12 +228,7 @@ async def generate_function_chat_completion(
         "__task__": __task__,
         "__task_body__": __task_body__,
         "__files__": files,
-        "__user__": {
-            "id": user.id,
-            "email": user.email,
-            "name": user.name,
-            "role": user.role,
-        },
+        "__user__": user.model_dump() if isinstance(user, UserModel) else {},
         "__metadata__": metadata,
         "__request__": request,
     }
@@ -253,8 +249,13 @@ async def generate_function_chat_completion(
             form_data["model"] = model_info.base_model_id
 
         params = model_info.params.model_dump()
-        form_data = apply_model_params_to_body_openai(params, form_data)
-        form_data = apply_model_system_prompt_to_body(params, form_data, metadata, user)
+
+        if params:
+            system = params.pop("system", None)
+            form_data = apply_model_params_to_body_openai(params, form_data)
+            form_data = apply_model_system_prompt_to_body(
+                system, form_data, metadata, user
+            )
 
     pipe_id = get_pipe_id(form_data)
     function_module = get_function_module_by_id(request, pipe_id)

backend/open_webui/internal/db.py

@@ -1,3 +1,4 @@
+import os
 import json
 import logging
 from contextlib import contextmanager
@@ -62,6 +63,9 @@ def handle_peewee_migration(DATABASE_URL):
 
     except Exception as e:
         log.error(f"Failed to initialize the database connection: {e}")
+        log.warning(
+            "Hint: If your database password contains special characters, you may need to URL-encode it."
+        )
         raise
     finally:
         # Properly closing the database connection
@@ -76,25 +80,58 @@ handle_peewee_migration(DATABASE_URL)
 
 
 SQLALCHEMY_DATABASE_URL = DATABASE_URL
-if "sqlite" in SQLALCHEMY_DATABASE_URL:
+
+# Handle SQLCipher URLs
+if SQLALCHEMY_DATABASE_URL.startswith("sqlite+sqlcipher://"):
+    database_password = os.environ.get("DATABASE_PASSWORD")
+    if not database_password or database_password.strip() == "":
+        raise ValueError(
+            "DATABASE_PASSWORD is required when using sqlite+sqlcipher:// URLs"
+        )
+
+    # Extract database path from SQLCipher URL
+    db_path = SQLALCHEMY_DATABASE_URL.replace("sqlite+sqlcipher://", "")
+    if db_path.startswith("/"):
+        db_path = db_path[1:]  # Remove leading slash for relative paths
+
+    # Create a custom creator function that uses sqlcipher3
+    def create_sqlcipher_connection():
+        import sqlcipher3
+
+        conn = sqlcipher3.connect(db_path, check_same_thread=False)
+        conn.execute(f"PRAGMA key = '{database_password}'")
+        return conn
+
+    engine = create_engine(
+        "sqlite://",  # Dummy URL since we're using creator
+        creator=create_sqlcipher_connection,
+        echo=False,
+    )
+
+    log.info("Connected to encrypted SQLite database using SQLCipher")
+
+elif "sqlite" in SQLALCHEMY_DATABASE_URL:
     engine = create_engine(
         SQLALCHEMY_DATABASE_URL, connect_args={"check_same_thread": False}
     )
 else:
-    if DATABASE_POOL_SIZE > 0:
-        engine = create_engine(
-            SQLALCHEMY_DATABASE_URL,
-            pool_size=DATABASE_POOL_SIZE,
-            max_overflow=DATABASE_POOL_MAX_OVERFLOW,
-            pool_timeout=DATABASE_POOL_TIMEOUT,
-            pool_recycle=DATABASE_POOL_RECYCLE,
-            pool_pre_ping=True,
-            poolclass=QueuePool,
-        )
+    if isinstance(DATABASE_POOL_SIZE, int):
+        if DATABASE_POOL_SIZE > 0:
+            engine = create_engine(
+                SQLALCHEMY_DATABASE_URL,
+                pool_size=DATABASE_POOL_SIZE,
+                max_overflow=DATABASE_POOL_MAX_OVERFLOW,
+                pool_timeout=DATABASE_POOL_TIMEOUT,
+                pool_recycle=DATABASE_POOL_RECYCLE,
+                pool_pre_ping=True,
+                poolclass=QueuePool,
+            )
+        else:
+            engine = create_engine(
+                SQLALCHEMY_DATABASE_URL, pool_pre_ping=True, poolclass=NullPool
+            )
     else:
-        engine = create_engine(
-            SQLALCHEMY_DATABASE_URL, pool_pre_ping=True, poolclass=NullPool
-        )
+        engine = create_engine(SQLALCHEMY_DATABASE_URL, pool_pre_ping=True)
 
 
 SessionLocal = sessionmaker(

backend/open_webui/internal/wrappers.py

@@ -1,4 +1,5 @@
 import logging
+import os
 from contextvars import ContextVar
 
 from open_webui.env import SRC_LOG_LEVELS
@@ -43,24 +44,47 @@ class ReconnectingPostgresqlDatabase(CustomReconnectMixin, PostgresqlDatabase):
 
 
 def register_connection(db_url):
-    db = connect(db_url, unquote_user=True, unquote_password=True)
-    if isinstance(db, PostgresqlDatabase):
-        # Enable autoconnect for SQLite databases, managed by Peewee
+    # Check if using SQLCipher protocol
+    if db_url.startswith("sqlite+sqlcipher://"):
+        database_password = os.environ.get("DATABASE_PASSWORD")
+        if not database_password or database_password.strip() == "":
+            raise ValueError(
+                "DATABASE_PASSWORD is required when using sqlite+sqlcipher:// URLs"
+            )
+        from playhouse.sqlcipher_ext import SqlCipherDatabase
+
+        # Parse the database path from SQLCipher URL
+        # Convert sqlite+sqlcipher:///path/to/db.sqlite to /path/to/db.sqlite
+        db_path = db_url.replace("sqlite+sqlcipher://", "")
+        if db_path.startswith("/"):
+            db_path = db_path[1:]  # Remove leading slash for relative paths
+
+        # Use Peewee's native SqlCipherDatabase with encryption
+        db = SqlCipherDatabase(db_path, passphrase=database_password)
         db.autoconnect = True
         db.reuse_if_open = True
-        log.info("Connected to PostgreSQL database")
+        log.info("Connected to encrypted SQLite database using SQLCipher")
 
-        # Get the connection details
-        connection = parse(db_url, unquote_user=True, unquote_password=True)
-
-        # Use our custom database class that supports reconnection
-        db = ReconnectingPostgresqlDatabase(**connection)
-        db.connect(reuse_if_open=True)
-    elif isinstance(db, SqliteDatabase):
-        # Enable autoconnect for SQLite databases, managed by Peewee
-        db.autoconnect = True
-        db.reuse_if_open = True
-        log.info("Connected to SQLite database")
     else:
-        raise ValueError("Unsupported database connection")
+        # Standard database connection (existing logic)
+        db = connect(db_url, unquote_user=True, unquote_password=True)
+        if isinstance(db, PostgresqlDatabase):
+            # Enable autoconnect for SQLite databases, managed by Peewee
+            db.autoconnect = True
+            db.reuse_if_open = True
+            log.info("Connected to PostgreSQL database")
+
+            # Get the connection details
+            connection = parse(db_url, unquote_user=True, unquote_password=True)
+
+            # Use our custom database class that supports reconnection
+            db = ReconnectingPostgresqlDatabase(**connection)
+            db.connect(reuse_if_open=True)
+        elif isinstance(db, SqliteDatabase):
+            # Enable autoconnect for SQLite databases, managed by Peewee
+            db.autoconnect = True
+            db.reuse_if_open = True
+            log.info("Connected to SQLite database")
+        else:
+            raise ValueError("Unsupported database connection")
     return db

backend/open_webui/main.py

@@ -8,6 +8,8 @@ import shutil
 import sys
 import time
 import random
+from uuid import uuid4
+
 
 from contextlib import asynccontextmanager
 from urllib.parse import urlencode, parse_qs, urlparse
@@ -19,6 +21,7 @@ from aiocache import cached
 import aiohttp
 import anyio.to_thread
 import requests
+from redis import Redis
 
 
 from fastapi import (
@@ -33,11 +36,10 @@ from fastapi import (
     applications,
     BackgroundTasks,
 )
-
 from fastapi.openapi.docs import get_swagger_ui_html
 
 from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import JSONResponse, RedirectResponse
+from fastapi.responses import FileResponse, JSONResponse, RedirectResponse
 from fastapi.staticfiles import StaticFiles
 
 from starlette_compress import CompressMiddleware
@@ -46,6 +48,7 @@ from starlette.exceptions import HTTPException as StarletteHTTPException
 from starlette.middleware.base import BaseHTTPMiddleware
 from starlette.middleware.sessions import SessionMiddleware
 from starlette.responses import Response, StreamingResponse
+from starlette.datastructures import Headers
 
 
 from open_webui.utils import logger
@@ -54,6 +57,8 @@ from open_webui.utils.logger import start_logger
 from open_webui.socket.main import (
     app as socket_app,
     periodic_usage_pool_cleanup,
+    get_models_in_use,
+    get_active_user_ids,
 )
 from open_webui.routers import (
     audio,
@@ -80,10 +85,12 @@ from open_webui.routers import (
     tools,
     users,
     utils,
+    scim,
 )
 
 from open_webui.routers.retrieval import (
     get_embedding_function,
+    get_reranking_function,
     get_ef,
     get_rf,
 )
@@ -96,7 +103,6 @@ from open_webui.models.users import UserModel, Users
 from open_webui.models.chats import Chats
 
 from open_webui.config import (
-    LICENSE_KEY,
     # Ollama
     ENABLE_OLLAMA_API,
     OLLAMA_BASE_URLS,
@@ -111,6 +117,8 @@ from open_webui.config import (
     OPENAI_API_CONFIGS,
     # Direct Connections
     ENABLE_DIRECT_CONNECTIONS,
+    # Model list
+    ENABLE_BASE_MODELS_CACHE,
     # Thread pool size for FastAPI/AnyIO
     THREAD_POOL_SIZE,
     # Tool Server Configs
@@ -154,6 +162,7 @@ from open_webui.config import (
     # Audio
     AUDIO_STT_ENGINE,
     AUDIO_STT_MODEL,
+    AUDIO_STT_SUPPORTED_CONTENT_TYPES,
     AUDIO_STT_OPENAI_API_BASE_URL,
     AUDIO_STT_OPENAI_API_KEY,
     AUDIO_STT_AZURE_API_KEY,
@@ -205,20 +214,27 @@ from open_webui.config import (
     RAG_ALLOWED_FILE_EXTENSIONS,
     RAG_FILE_MAX_COUNT,
     RAG_FILE_MAX_SIZE,
+    FILE_IMAGE_COMPRESSION_WIDTH,
+    FILE_IMAGE_COMPRESSION_HEIGHT,
     RAG_OPENAI_API_BASE_URL,
     RAG_OPENAI_API_KEY,
+    RAG_AZURE_OPENAI_BASE_URL,
+    RAG_AZURE_OPENAI_API_KEY,
+    RAG_AZURE_OPENAI_API_VERSION,
     RAG_OLLAMA_BASE_URL,
     RAG_OLLAMA_API_KEY,
     CHUNK_OVERLAP,
     CHUNK_SIZE,
     CONTENT_EXTRACTION_ENGINE,
     DATALAB_MARKER_API_KEY,
-    DATALAB_MARKER_LANGS,
+    DATALAB_MARKER_API_BASE_URL,
+    DATALAB_MARKER_ADDITIONAL_CONFIG,
     DATALAB_MARKER_SKIP_CACHE,
     DATALAB_MARKER_FORCE_OCR,
     DATALAB_MARKER_PAGINATE,
     DATALAB_MARKER_STRIP_EXISTING_OCR,
     DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+    DATALAB_MARKER_FORMAT_LINES,
     DATALAB_MARKER_OUTPUT_FORMAT,
     DATALAB_MARKER_USE_LLM,
     EXTERNAL_DOCUMENT_LOADER_URL,
@@ -228,6 +244,9 @@ from open_webui.config import (
     DOCLING_OCR_ENGINE,
     DOCLING_OCR_LANG,
     DOCLING_DO_PICTURE_DESCRIPTION,
+    DOCLING_PICTURE_DESCRIPTION_MODE,
+    DOCLING_PICTURE_DESCRIPTION_LOCAL,
+    DOCLING_PICTURE_DESCRIPTION_API,
     DOCUMENT_INTELLIGENCE_ENDPOINT,
     DOCUMENT_INTELLIGENCE_KEY,
     MISTRAL_OCR_API_KEY,
@@ -265,6 +284,8 @@ from open_webui.config import (
     BRAVE_SEARCH_API_KEY,
     EXA_API_KEY,
     PERPLEXITY_API_KEY,
+    PERPLEXITY_MODEL,
+    PERPLEXITY_SEARCH_CONTEXT_USAGE,
     SOUGOU_API_SID,
     SOUGOU_API_SK,
     KAGI_SEARCH_API_KEY,
@@ -338,6 +359,10 @@ from open_webui.config import (
     LDAP_CA_CERT_FILE,
     LDAP_VALIDATE_CERT,
     LDAP_CIPHERS,
+    # LDAP Group Management
+    ENABLE_LDAP_GROUP_MANAGEMENT,
+    ENABLE_LDAP_GROUP_CREATION,
+    LDAP_ATTRIBUTE_FOR_GROUPS,
     # Misc
     ENV,
     CACHE_DIR,
@@ -356,10 +381,12 @@ from open_webui.config import (
     TASK_MODEL_EXTERNAL,
     ENABLE_TAGS_GENERATION,
     ENABLE_TITLE_GENERATION,
+    ENABLE_FOLLOW_UP_GENERATION,
     ENABLE_SEARCH_QUERY_GENERATION,
     ENABLE_RETRIEVAL_QUERY_GENERATION,
     ENABLE_AUTOCOMPLETE_GENERATION,
     TITLE_GENERATION_PROMPT_TEMPLATE,
+    FOLLOW_UP_GENERATION_PROMPT_TEMPLATE,
     TAGS_GENERATION_PROMPT_TEMPLATE,
     IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE,
     TOOLS_FUNCTION_CALLING_PROMPT_TEMPLATE,
@@ -370,10 +397,13 @@ from open_webui.config import (
     reset_config,
 )
 from open_webui.env import (
+    LICENSE_KEY,
     AUDIT_EXCLUDED_PATHS,
     AUDIT_LOG_LEVEL,
     CHANGELOG,
     REDIS_URL,
+    REDIS_CLUSTER,
+    REDIS_KEY_PREFIX,
     REDIS_SENTINEL_HOSTS,
     REDIS_SENTINEL_PORT,
     GLOBAL_LOG_LEVEL,
@@ -381,17 +411,23 @@ from open_webui.env import (
     SAFE_MODE,
     SRC_LOG_LEVELS,
     VERSION,
+    INSTANCE_ID,
     WEBUI_BUILD_HASH,
     WEBUI_SECRET_KEY,
     WEBUI_SESSION_COOKIE_SAME_SITE,
     WEBUI_SESSION_COOKIE_SECURE,
+    ENABLE_SIGNUP_PASSWORD_CONFIRMATION,
     WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
     WEBUI_AUTH_TRUSTED_NAME_HEADER,
     WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
+    # SCIM
+    SCIM_ENABLED,
+    SCIM_TOKEN,
+    ENABLE_COMPRESSION_MIDDLEWARE,
     ENABLE_WEBSOCKET_SUPPORT,
     BYPASS_MODEL_ACCESS_CONTROL,
     RESET_CONFIG_ON_START,
-    OFFLINE_MODE,
+    ENABLE_VERSION_UPDATE_CHECK,
     ENABLE_OTEL,
     EXTERNAL_PWA_MANIFEST_URL,
     AIOHTTP_CLIENT_SESSION_SSL,
@@ -408,6 +444,7 @@ from open_webui.utils.chat import (
     chat_completed as chat_completed_handler,
     chat_action as chat_action_handler,
 )
+from open_webui.utils.embeddings import generate_embeddings
 from open_webui.utils.middleware import process_chat_payload, process_chat_response
 from open_webui.utils.access_control import has_access
 
@@ -421,9 +458,11 @@ from open_webui.utils.auth import (
 from open_webui.utils.plugin import install_tool_and_function_dependencies
 from open_webui.utils.oauth import OAuthManager
 from open_webui.utils.security_headers import SecurityHeadersMiddleware
+from open_webui.utils.redis import get_redis_connection
 
 from open_webui.tasks import (
-    list_task_ids_by_chat_id,
+    redis_task_command_listener,
+    list_task_ids_by_item_id,
     stop_task,
     list_tasks,
 )  # Import from tasks.py
@@ -431,6 +470,9 @@ from open_webui.tasks import (
 from open_webui.utils.redis import get_sentinels_from_env
 
 
+from open_webui.constants import ERROR_MESSAGES
+
+
 if SAFE_MODE:
     print("SAFE MODE ENABLED")
     Functions.deactivate_all_functions()
@@ -474,7 +516,9 @@ https://github.com/open-webui/open-webui
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
+    app.state.instance_id = INSTANCE_ID
     start_logger()
+
     if RESET_CONFIG_ON_START:
         reset_config()
 
@@ -486,14 +530,52 @@ async def lifespan(app: FastAPI):
     log.info("Installing external dependencies of functions and tools...")
     install_tool_and_function_dependencies()
 
+    app.state.redis = get_redis_connection(
+        redis_url=REDIS_URL,
+        redis_sentinels=get_sentinels_from_env(
+            REDIS_SENTINEL_HOSTS, REDIS_SENTINEL_PORT
+        ),
+        redis_cluster=REDIS_CLUSTER,
+        async_mode=True,
+    )
+
+    if app.state.redis is not None:
+        app.state.redis_task_command_listener = asyncio.create_task(
+            redis_task_command_listener(app)
+        )
+
     if THREAD_POOL_SIZE and THREAD_POOL_SIZE > 0:
         limiter = anyio.to_thread.current_default_thread_limiter()
         limiter.total_tokens = THREAD_POOL_SIZE
 
     asyncio.create_task(periodic_usage_pool_cleanup())
 
+    if app.state.config.ENABLE_BASE_MODELS_CACHE:
+        await get_all_models(
+            Request(
+                # Creating a mock request object to pass to get_all_models
+                {
+                    "type": "http",
+                    "asgi.version": "3.0",
+                    "asgi.spec_version": "2.0",
+                    "method": "GET",
+                    "path": "/internal",
+                    "query_string": b"",
+                    "headers": Headers({}).raw,
+                    "client": ("127.0.0.1", 12345),
+                    "server": ("127.0.0.1", 80),
+                    "scheme": "http",
+                    "app": app,
+                }
+            ),
+            None,
+        )
+
     yield
 
+    if hasattr(app.state, "redis_task_command_listener"):
+        app.state.redis_task_command_listener.cancel()
+
 
 app = FastAPI(
     title="Open WebUI",
@@ -505,10 +587,14 @@ app = FastAPI(
 
 oauth_manager = OAuthManager(app)
 
+app.state.instance_id = None
 app.state.config = AppConfig(
     redis_url=REDIS_URL,
     redis_sentinels=get_sentinels_from_env(REDIS_SENTINEL_HOSTS, REDIS_SENTINEL_PORT),
+    redis_cluster=REDIS_CLUSTER,
+    redis_key_prefix=REDIS_KEY_PREFIX,
 )
+app.state.redis = None
 
 app.state.WEBUI_NAME = WEBUI_NAME
 app.state.LICENSE_METADATA = None
@@ -569,6 +655,24 @@ app.state.TOOL_SERVERS = []
 
 app.state.config.ENABLE_DIRECT_CONNECTIONS = ENABLE_DIRECT_CONNECTIONS
 
+########################################
+#
+# SCIM
+#
+########################################
+
+app.state.SCIM_ENABLED = SCIM_ENABLED
+app.state.SCIM_TOKEN = SCIM_TOKEN
+
+########################################
+#
+# MODELS
+#
+########################################
+
+app.state.config.ENABLE_BASE_MODELS_CACHE = ENABLE_BASE_MODELS_CACHE
+app.state.BASE_MODELS = []
+
 ########################################
 #
 # WEBUI
@@ -639,6 +743,11 @@ app.state.config.LDAP_CA_CERT_FILE = LDAP_CA_CERT_FILE
 app.state.config.LDAP_VALIDATE_CERT = LDAP_VALIDATE_CERT
 app.state.config.LDAP_CIPHERS = LDAP_CIPHERS
 
+# For LDAP Group Management
+app.state.config.ENABLE_LDAP_GROUP_MANAGEMENT = ENABLE_LDAP_GROUP_MANAGEMENT
+app.state.config.ENABLE_LDAP_GROUP_CREATION = ENABLE_LDAP_GROUP_CREATION
+app.state.config.LDAP_ATTRIBUTE_FOR_GROUPS = LDAP_ATTRIBUTE_FOR_GROUPS
+
 
 app.state.AUTH_TRUSTED_EMAIL_HEADER = WEBUI_AUTH_TRUSTED_EMAIL_HEADER
 app.state.AUTH_TRUSTED_NAME_HEADER = WEBUI_AUTH_TRUSTED_NAME_HEADER
@@ -664,9 +773,13 @@ app.state.config.TOP_K = RAG_TOP_K
 app.state.config.TOP_K_RERANKER = RAG_TOP_K_RERANKER
 app.state.config.RELEVANCE_THRESHOLD = RAG_RELEVANCE_THRESHOLD
 app.state.config.HYBRID_BM25_WEIGHT = RAG_HYBRID_BM25_WEIGHT
+
+
 app.state.config.ALLOWED_FILE_EXTENSIONS = RAG_ALLOWED_FILE_EXTENSIONS
 app.state.config.FILE_MAX_SIZE = RAG_FILE_MAX_SIZE
 app.state.config.FILE_MAX_COUNT = RAG_FILE_MAX_COUNT
+app.state.config.FILE_IMAGE_COMPRESSION_WIDTH = FILE_IMAGE_COMPRESSION_WIDTH
+app.state.config.FILE_IMAGE_COMPRESSION_HEIGHT = FILE_IMAGE_COMPRESSION_HEIGHT
 
 
 app.state.config.RAG_FULL_CONTEXT = RAG_FULL_CONTEXT
@@ -676,7 +789,8 @@ app.state.config.ENABLE_WEB_LOADER_SSL_VERIFICATION = ENABLE_WEB_LOADER_SSL_VERI
 
 app.state.config.CONTENT_EXTRACTION_ENGINE = CONTENT_EXTRACTION_ENGINE
 app.state.config.DATALAB_MARKER_API_KEY = DATALAB_MARKER_API_KEY
-app.state.config.DATALAB_MARKER_LANGS = DATALAB_MARKER_LANGS
+app.state.config.DATALAB_MARKER_API_BASE_URL = DATALAB_MARKER_API_BASE_URL
+app.state.config.DATALAB_MARKER_ADDITIONAL_CONFIG = DATALAB_MARKER_ADDITIONAL_CONFIG
 app.state.config.DATALAB_MARKER_SKIP_CACHE = DATALAB_MARKER_SKIP_CACHE
 app.state.config.DATALAB_MARKER_FORCE_OCR = DATALAB_MARKER_FORCE_OCR
 app.state.config.DATALAB_MARKER_PAGINATE = DATALAB_MARKER_PAGINATE
@@ -684,6 +798,7 @@ app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR = DATALAB_MARKER_STRIP_EXISTI
 app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = (
     DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
 )
+app.state.config.DATALAB_MARKER_FORMAT_LINES = DATALAB_MARKER_FORMAT_LINES
 app.state.config.DATALAB_MARKER_USE_LLM = DATALAB_MARKER_USE_LLM
 app.state.config.DATALAB_MARKER_OUTPUT_FORMAT = DATALAB_MARKER_OUTPUT_FORMAT
 app.state.config.EXTERNAL_DOCUMENT_LOADER_URL = EXTERNAL_DOCUMENT_LOADER_URL
@@ -693,6 +808,9 @@ app.state.config.DOCLING_SERVER_URL = DOCLING_SERVER_URL
 app.state.config.DOCLING_OCR_ENGINE = DOCLING_OCR_ENGINE
 app.state.config.DOCLING_OCR_LANG = DOCLING_OCR_LANG
 app.state.config.DOCLING_DO_PICTURE_DESCRIPTION = DOCLING_DO_PICTURE_DESCRIPTION
+app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE = DOCLING_PICTURE_DESCRIPTION_MODE
+app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL = DOCLING_PICTURE_DESCRIPTION_LOCAL
+app.state.config.DOCLING_PICTURE_DESCRIPTION_API = DOCLING_PICTURE_DESCRIPTION_API
 app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT = DOCUMENT_INTELLIGENCE_ENDPOINT
 app.state.config.DOCUMENT_INTELLIGENCE_KEY = DOCUMENT_INTELLIGENCE_KEY
 app.state.config.MISTRAL_OCR_API_KEY = MISTRAL_OCR_API_KEY
@@ -717,6 +835,10 @@ app.state.config.RAG_TEMPLATE = RAG_TEMPLATE
 app.state.config.RAG_OPENAI_API_BASE_URL = RAG_OPENAI_API_BASE_URL
 app.state.config.RAG_OPENAI_API_KEY = RAG_OPENAI_API_KEY
 
+app.state.config.RAG_AZURE_OPENAI_BASE_URL = RAG_AZURE_OPENAI_BASE_URL
+app.state.config.RAG_AZURE_OPENAI_API_KEY = RAG_AZURE_OPENAI_API_KEY
+app.state.config.RAG_AZURE_OPENAI_API_VERSION = RAG_AZURE_OPENAI_API_VERSION
+
 app.state.config.RAG_OLLAMA_BASE_URL = RAG_OLLAMA_BASE_URL
 app.state.config.RAG_OLLAMA_API_KEY = RAG_OLLAMA_API_KEY
 
@@ -764,6 +886,8 @@ app.state.config.BING_SEARCH_V7_ENDPOINT = BING_SEARCH_V7_ENDPOINT
 app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY = BING_SEARCH_V7_SUBSCRIPTION_KEY
 app.state.config.EXA_API_KEY = EXA_API_KEY
 app.state.config.PERPLEXITY_API_KEY = PERPLEXITY_API_KEY
+app.state.config.PERPLEXITY_MODEL = PERPLEXITY_MODEL
+app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE = PERPLEXITY_SEARCH_CONTEXT_USAGE
 app.state.config.SOUGOU_API_SID = SOUGOU_API_SID
 app.state.config.SOUGOU_API_SK = SOUGOU_API_SK
 app.state.config.EXTERNAL_WEB_SEARCH_URL = EXTERNAL_WEB_SEARCH_URL
@@ -779,6 +903,7 @@ app.state.config.FIRECRAWL_API_KEY = FIRECRAWL_API_KEY
 app.state.config.TAVILY_EXTRACT_DEPTH = TAVILY_EXTRACT_DEPTH
 
 app.state.EMBEDDING_FUNCTION = None
+app.state.RERANKING_FUNCTION = None
 app.state.ef = None
 app.state.rf = None
 
@@ -807,18 +932,37 @@ except Exception as e:
 app.state.EMBEDDING_FUNCTION = get_embedding_function(
     app.state.config.RAG_EMBEDDING_ENGINE,
     app.state.config.RAG_EMBEDDING_MODEL,
-    app.state.ef,
-    (
+    embedding_function=app.state.ef,
+    url=(
         app.state.config.RAG_OPENAI_API_BASE_URL
         if app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-        else app.state.config.RAG_OLLAMA_BASE_URL
+        else (
+            app.state.config.RAG_OLLAMA_BASE_URL
+            if app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+            else app.state.config.RAG_AZURE_OPENAI_BASE_URL
+        )
     ),
-    (
+    key=(
         app.state.config.RAG_OPENAI_API_KEY
         if app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-        else app.state.config.RAG_OLLAMA_API_KEY
+        else (
+            app.state.config.RAG_OLLAMA_API_KEY
+            if app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+            else app.state.config.RAG_AZURE_OPENAI_API_KEY
+        )
     ),
-    app.state.config.RAG_EMBEDDING_BATCH_SIZE,
+    embedding_batch_size=app.state.config.RAG_EMBEDDING_BATCH_SIZE,
+    azure_api_version=(
+        app.state.config.RAG_AZURE_OPENAI_API_VERSION
+        if app.state.config.RAG_EMBEDDING_ENGINE == "azure_openai"
+        else None
+    ),
+)
+
+app.state.RERANKING_FUNCTION = get_reranking_function(
+    app.state.config.RAG_RERANKING_ENGINE,
+    app.state.config.RAG_RERANKING_MODEL,
+    reranking_function=app.state.rf,
 )
 
 ########################################
@@ -889,10 +1033,12 @@ app.state.config.IMAGE_STEPS = IMAGE_STEPS
 #
 ########################################
 
-app.state.config.STT_OPENAI_API_BASE_URL = AUDIO_STT_OPENAI_API_BASE_URL
-app.state.config.STT_OPENAI_API_KEY = AUDIO_STT_OPENAI_API_KEY
 app.state.config.STT_ENGINE = AUDIO_STT_ENGINE
 app.state.config.STT_MODEL = AUDIO_STT_MODEL
+app.state.config.STT_SUPPORTED_CONTENT_TYPES = AUDIO_STT_SUPPORTED_CONTENT_TYPES
+
+app.state.config.STT_OPENAI_API_BASE_URL = AUDIO_STT_OPENAI_API_BASE_URL
+app.state.config.STT_OPENAI_API_KEY = AUDIO_STT_OPENAI_API_KEY
 
 app.state.config.WHISPER_MODEL = WHISPER_MODEL
 app.state.config.WHISPER_VAD_FILTER = WHISPER_VAD_FILTER
@@ -939,6 +1085,7 @@ app.state.config.ENABLE_RETRIEVAL_QUERY_GENERATION = ENABLE_RETRIEVAL_QUERY_GENE
 app.state.config.ENABLE_AUTOCOMPLETE_GENERATION = ENABLE_AUTOCOMPLETE_GENERATION
 app.state.config.ENABLE_TAGS_GENERATION = ENABLE_TAGS_GENERATION
 app.state.config.ENABLE_TITLE_GENERATION = ENABLE_TITLE_GENERATION
+app.state.config.ENABLE_FOLLOW_UP_GENERATION = ENABLE_FOLLOW_UP_GENERATION
 
 
 app.state.config.TITLE_GENERATION_PROMPT_TEMPLATE = TITLE_GENERATION_PROMPT_TEMPLATE
@@ -946,6 +1093,9 @@ app.state.config.TAGS_GENERATION_PROMPT_TEMPLATE = TAGS_GENERATION_PROMPT_TEMPLA
 app.state.config.IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE = (
     IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE
 )
+app.state.config.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE = (
+    FOLLOW_UP_GENERATION_PROMPT_TEMPLATE
+)
 
 app.state.config.TOOLS_FUNCTION_CALLING_PROMPT_TEMPLATE = (
     TOOLS_FUNCTION_CALLING_PROMPT_TEMPLATE
@@ -989,7 +1139,9 @@ class RedirectMiddleware(BaseHTTPMiddleware):
 
 
 # Add the middleware to the app
-app.add_middleware(CompressMiddleware)
+if ENABLE_COMPRESSION_MIDDLEWARE:
+    app.add_middleware(CompressMiddleware)
+
 app.add_middleware(RedirectMiddleware)
 app.add_middleware(SecurityHeadersMiddleware)
 
@@ -1083,6 +1235,10 @@ app.include_router(
 )
 app.include_router(utils.router, prefix="/api/v1/utils", tags=["utils"])
 
+# SCIM 2.0 API for identity management
+if SCIM_ENABLED:
+    app.include_router(scim.router, prefix="/api/v1/scim/v2", tags=["scim"])
+
 
 try:
     audit_level = AuditLevel(AUDIT_LOG_LEVEL)
@@ -1105,7 +1261,9 @@ if audit_level != AuditLevel.NONE:
 
 
 @app.get("/api/models")
-async def get_models(request: Request, user=Depends(get_verified_user)):
+async def get_models(
+    request: Request, refresh: bool = False, user=Depends(get_verified_user)
+):
     def get_filtered_models(models, user):
         filtered_models = []
         for model in models:
@@ -1129,7 +1287,7 @@ async def get_models(request: Request, user=Depends(get_verified_user)):
 
         return filtered_models
 
-    all_models = await get_all_models(request, user=user)
+    all_models = await get_all_models(request, refresh=refresh, user=user)
 
     models = []
     for model in all_models:
@@ -1166,7 +1324,7 @@ async def get_models(request: Request, user=Depends(get_verified_user)):
         models = get_filtered_models(models, user)
 
     log.debug(
-        f"/api/models returned filtered models accessible to the user: {json.dumps([model['id'] for model in models])}"
+        f"/api/models returned filtered models accessible to the user: {json.dumps([model.get('id') for model in models])}"
     )
     return {"data": models}
 
@@ -1177,6 +1335,37 @@ async def get_base_models(request: Request, user=Depends(get_admin_user)):
     return {"data": models}
 
 
+##################################
+# Embeddings
+##################################
+
+
+@app.post("/api/embeddings")
+async def embeddings(
+    request: Request, form_data: dict, user=Depends(get_verified_user)
+):
+    """
+    OpenAI-compatible embeddings endpoint.
+
+    This handler:
+      - Performs user/model checks and dispatches to the correct backend.
+      - Supports OpenAI, Ollama, arena models, pipelines, and any compatible provider.
+
+    Args:
+        request (Request): Request context.
+        form_data (dict): OpenAI-like payload (e.g., {"model": "...", "input": [...]})
+        user (UserModel): Authenticated user.
+
+    Returns:
+        dict: OpenAI-compatible embeddings response.
+    """
+    # Make sure models are loaded in app state
+    if not request.app.state.MODELS:
+        await get_all_models(request, user=user)
+    # Use generic dispatcher in utils.embeddings
+    return await generate_embeddings(request, form_data, user)
+
+
 @app.post("/api/chat/completions")
 async def chat_completion(
     request: Request,
@@ -1212,6 +1401,19 @@ async def chat_completion(
             request.state.direct = True
             request.state.model = model
 
+        model_info_params = (
+            model_info.params.model_dump() if model_info and model_info.params else {}
+        )
+
+        # Chat Params
+        stream_delta_chunk_size = form_data.get("params", {}).get(
+            "stream_delta_chunk_size"
+        )
+
+        # Model Params
+        if model_info_params.get("stream_delta_chunk_size"):
+            stream_delta_chunk_size = model_info_params.get("stream_delta_chunk_size")
+
         metadata = {
             "user_id": user.id,
             "chat_id": form_data.pop("chat_id", None),
@@ -1225,25 +1427,33 @@ async def chat_completion(
             "variables": form_data.get("variables", {}),
             "model": model,
             "direct": model_item.get("direct", False),
-            **(
-                {"function_calling": "native"}
-                if form_data.get("params", {}).get("function_calling") == "native"
-                or (
-                    model_info
-                    and model_info.params.model_dump().get("function_calling")
-                    == "native"
-                )
-                else {}
-            ),
+            "params": {
+                "stream_delta_chunk_size": stream_delta_chunk_size,
+                "function_calling": (
+                    "native"
+                    if (
+                        form_data.get("params", {}).get("function_calling") == "native"
+                        or model_info_params.get("function_calling") == "native"
+                    )
+                    else "default"
+                ),
+            },
         }
 
+        if metadata.get("chat_id") and (user and user.role != "admin"):
+            chat = Chats.get_chat_by_id_and_user_id(metadata["chat_id"], user.id)
+            if chat is None:
+                raise HTTPException(
+                    status_code=status.HTTP_404_NOT_FOUND,
+                    detail=ERROR_MESSAGES.DEFAULT(),
+                )
+
         request.state.metadata = metadata
         form_data["metadata"] = metadata
 
         form_data, metadata, events = await process_chat_payload(
             request, form_data, user, metadata, model
         )
-
     except Exception as e:
         log.debug(f"Error processing chat payload: {e}")
         if metadata.get("chat_id") and metadata.get("message_id"):
@@ -1263,11 +1473,30 @@ async def chat_completion(
 
     try:
         response = await chat_completion_handler(request, form_data, user)
+        if metadata.get("chat_id") and metadata.get("message_id"):
+            Chats.upsert_message_to_chat_by_id_and_message_id(
+                metadata["chat_id"],
+                metadata["message_id"],
+                {
+                    "model": model_id,
+                },
+            )
 
         return await process_chat_response(
             request, response, form_data, user, metadata, model, events, tasks
         )
     except Exception as e:
+        log.debug(f"Error in chat completion: {e}")
+        if metadata.get("chat_id") and metadata.get("message_id"):
+            # Update the chat message with the error
+            Chats.upsert_message_to_chat_by_id_and_message_id(
+                metadata["chat_id"],
+                metadata["message_id"],
+                {
+                    "error": {"content": str(e)},
+                },
+            )
+
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,
             detail=str(e),
@@ -1318,28 +1547,32 @@ async def chat_action(
 
 
 @app.post("/api/tasks/stop/{task_id}")
-async def stop_task_endpoint(task_id: str, user=Depends(get_verified_user)):
+async def stop_task_endpoint(
+    request: Request, task_id: str, user=Depends(get_verified_user)
+):
     try:
-        result = await stop_task(task_id)
+        result = await stop_task(request.app.state.redis, task_id)
         return result
     except ValueError as e:
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(e))
 
 
 @app.get("/api/tasks")
-async def list_tasks_endpoint(user=Depends(get_verified_user)):
-    return {"tasks": list_tasks()}
+async def list_tasks_endpoint(request: Request, user=Depends(get_verified_user)):
+    return {"tasks": await list_tasks(request.app.state.redis)}
 
 
 @app.get("/api/tasks/chat/{chat_id}")
-async def list_tasks_by_chat_id_endpoint(chat_id: str, user=Depends(get_verified_user)):
+async def list_tasks_by_chat_id_endpoint(
+    request: Request, chat_id: str, user=Depends(get_verified_user)
+):
     chat = Chats.get_chat_by_id(chat_id)
     if chat is None or chat.user_id != user.id:
         return {"task_ids": []}
 
-    task_ids = list_task_ids_by_chat_id(chat_id)
+    task_ids = await list_task_ids_by_item_id(request.app.state.redis, chat_id)
 
-    print(f"Task IDs for chat {chat_id}: {task_ids}")
+    log.debug(f"Task IDs for chat {chat_id}: {task_ids}")
     return {"task_ids": task_ids}
 
 
@@ -1387,11 +1620,13 @@ async def get_app_config(request: Request):
         "features": {
             "auth": WEBUI_AUTH,
             "auth_trusted_header": bool(app.state.AUTH_TRUSTED_EMAIL_HEADER),
+            "enable_signup_password_confirmation": ENABLE_SIGNUP_PASSWORD_CONFIRMATION,
             "enable_ldap": app.state.config.ENABLE_LDAP,
             "enable_api_key": app.state.config.ENABLE_API_KEY,
             "enable_signup": app.state.config.ENABLE_SIGNUP,
             "enable_login_form": app.state.config.ENABLE_LOGIN_FORM,
             "enable_websocket": ENABLE_WEBSOCKET_SUPPORT,
+            "enable_version_update_check": ENABLE_VERSION_UPDATE_CHECK,
             **(
                 {
                     "enable_direct_connections": app.state.config.ENABLE_DIRECT_CONNECTIONS,
@@ -1435,6 +1670,10 @@ async def get_app_config(request: Request):
                 "file": {
                     "max_size": app.state.config.FILE_MAX_SIZE,
                     "max_count": app.state.config.FILE_MAX_COUNT,
+                    "image_compression": {
+                        "width": app.state.config.FILE_IMAGE_COMPRESSION_WIDTH,
+                        "height": app.state.config.FILE_IMAGE_COMPRESSION_HEIGHT,
+                    },
                 },
                 "permissions": {**app.state.config.USER_PERMISSIONS},
                 "google_drive": {
@@ -1460,8 +1699,23 @@ async def get_app_config(request: Request):
                     else {}
                 ),
             }
-            if user is not None
-            else {}
+            if user is not None and (user.role in ["admin", "user"])
+            else {
+                **(
+                    {
+                        "metadata": {
+                            "login_footer": app.state.LICENSE_METADATA.get(
+                                "login_footer", ""
+                            ),
+                            "auth_logo_position": app.state.LICENSE_METADATA.get(
+                                "auth_logo_position", ""
+                            ),
+                        }
+                    }
+                    if app.state.LICENSE_METADATA
+                    else {}
+                )
+            }
         ),
     }
 
@@ -1493,9 +1747,9 @@ async def get_app_version():
 
 @app.get("/api/version/updates")
 async def get_app_latest_release_version(user=Depends(get_verified_user)):
-    if OFFLINE_MODE:
+    if not ENABLE_VERSION_UPDATE_CHECK:
         log.debug(
-            f"Offline mode is enabled, returning current version as latest version"
+            f"Version update check is disabled, returning current version as latest version"
         )
         return {"current": VERSION, "latest": VERSION}
     try:
@@ -1520,6 +1774,19 @@ async def get_app_changelog():
     return {key: CHANGELOG[key] for idx, key in enumerate(CHANGELOG) if idx < 5}
 
 
+@app.get("/api/usage")
+async def get_current_usage(user=Depends(get_verified_user)):
+    """
+    Get current usage statistics for Open WebUI.
+    This is an experimental endpoint and subject to change.
+    """
+    try:
+        return {"model_ids": get_models_in_use(), "user_ids": get_active_user_ids()}
+    except Exception as e:
+        log.error(f"Error getting usage statistics: {e}")
+        raise HTTPException(status_code=500, detail="Internal Server Error")
+
+
 ############################
 # OAuth Login & Callback
 ############################
@@ -1559,11 +1826,10 @@ async def get_manifest_json():
         return {
             "name": app.state.WEBUI_NAME,
             "short_name": app.state.WEBUI_NAME,
-            "description": "Open WebUI is an open, extensible, user-friendly interface for AI that adapts to your workflow.",
+            "description": f"{app.state.WEBUI_NAME} is an open, extensible, user-friendly interface for AI that adapts to your workflow.",
             "start_url": "/",
             "display": "standalone",
             "background_color": "#343541",
-            "orientation": "any",
             "icons": [
                 {
                     "src": "/static/logo.png",
@@ -1608,7 +1874,20 @@ async def healthcheck_with_db():
 
 
 app.mount("/static", StaticFiles(directory=STATIC_DIR), name="static")
-app.mount("/cache", StaticFiles(directory=CACHE_DIR), name="cache")
+
+
+@app.get("/cache/{path:path}")
+async def serve_cache_file(
+    path: str,
+    user=Depends(get_verified_user),
+):
+    file_path = os.path.abspath(os.path.join(CACHE_DIR, path))
+    # prevent path traversal
+    if not file_path.startswith(os.path.abspath(CACHE_DIR)):
+        raise HTTPException(status_code=404, detail="File not found")
+    if not os.path.isfile(file_path):
+        raise HTTPException(status_code=404, detail="File not found")
+    return FileResponse(file_path)
 
 
 def swagger_ui_html(*args, **kwargs):

backend/open_webui/migrations/env.py

@@ -2,8 +2,8 @@ from logging.config import fileConfig
 
 from alembic import context
 from open_webui.models.auths import Auth
-from open_webui.env import DATABASE_URL
-from sqlalchemy import engine_from_config, pool
+from open_webui.env import DATABASE_URL, DATABASE_PASSWORD
+from sqlalchemy import engine_from_config, pool, create_engine
 
 # this is the Alembic Config object, which provides
 # access to the values within the .ini file in use.
@@ -62,11 +62,38 @@ def run_migrations_online() -> None:
     and associate a connection with the context.
 
     """
-    connectable = engine_from_config(
-        config.get_section(config.config_ini_section, {}),
-        prefix="sqlalchemy.",
-        poolclass=pool.NullPool,
-    )
+    # Handle SQLCipher URLs
+    if DB_URL and DB_URL.startswith("sqlite+sqlcipher://"):
+        if not DATABASE_PASSWORD or DATABASE_PASSWORD.strip() == "":
+            raise ValueError(
+                "DATABASE_PASSWORD is required when using sqlite+sqlcipher:// URLs"
+            )
+
+        # Extract database path from SQLCipher URL
+        db_path = DB_URL.replace("sqlite+sqlcipher://", "")
+        if db_path.startswith("/"):
+            db_path = db_path[1:]  # Remove leading slash for relative paths
+
+        # Create a custom creator function that uses sqlcipher3
+        def create_sqlcipher_connection():
+            import sqlcipher3
+
+            conn = sqlcipher3.connect(db_path, check_same_thread=False)
+            conn.execute(f"PRAGMA key = '{DATABASE_PASSWORD}'")
+            return conn
+
+        connectable = create_engine(
+            "sqlite://",  # Dummy URL since we're using creator
+            creator=create_sqlcipher_connection,
+            echo=False,
+        )
+    else:
+        # Standard database connection (existing logic)
+        connectable = engine_from_config(
+            config.get_section(config.config_ini_section, {}),
+            prefix="sqlalchemy.",
+            poolclass=pool.NullPool,
+        )
 
     with connectable.connect() as connection:
         context.configure(connection=connection, target_metadata=target_metadata)

backend/open_webui/migrations/versions/d31026856c01_update_folder_table_data.py

@@ -0,0 +1,23 @@
+"""Update folder table data
+
+Revision ID: d31026856c01
+Revises: 9f0c9cd09105
+Create Date: 2025-07-13 03:00:00.000000
+
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "d31026856c01"
+down_revision = "9f0c9cd09105"
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    op.add_column("folder", sa.Column("data", sa.JSON(), nullable=True))
+
+
+def downgrade():
+    op.drop_column("folder", "data")

backend/open_webui/models/chats.py

@@ -6,12 +6,14 @@ from typing import Optional
 
 from open_webui.internal.db import Base, get_db
 from open_webui.models.tags import TagModel, Tag, Tags
+from open_webui.models.folders import Folders
 from open_webui.env import SRC_LOG_LEVELS
 
 from pydantic import BaseModel, ConfigDict
 from sqlalchemy import BigInteger, Boolean, Column, String, Text, JSON
 from sqlalchemy import or_, func, select, and_, text
 from sqlalchemy.sql import exists
+from sqlalchemy.sql.expression import bindparam
 
 ####################
 # Chat DB Schema
@@ -66,12 +68,14 @@ class ChatModel(BaseModel):
 
 class ChatForm(BaseModel):
     chat: dict
+    folder_id: Optional[str] = None
 
 
 class ChatImportForm(ChatForm):
     meta: Optional[dict] = {}
     pinned: Optional[bool] = False
-    folder_id: Optional[str] = None
+    created_at: Optional[int] = None
+    updated_at: Optional[int] = None
 
 
 class ChatTitleMessagesForm(BaseModel):
@@ -118,6 +122,7 @@ class ChatTable:
                         else "New Chat"
                     ),
                     "chat": form_data.chat,
+                    "folder_id": form_data.folder_id,
                     "created_at": int(time.time()),
                     "updated_at": int(time.time()),
                 }
@@ -147,8 +152,16 @@ class ChatTable:
                     "meta": form_data.meta,
                     "pinned": form_data.pinned,
                     "folder_id": form_data.folder_id,
-                    "created_at": int(time.time()),
-                    "updated_at": int(time.time()),
+                    "created_at": (
+                        form_data.created_at
+                        if form_data.created_at
+                        else int(time.time())
+                    ),
+                    "updated_at": (
+                        form_data.updated_at
+                        if form_data.updated_at
+                        else int(time.time())
+                    ),
                 }
             )
 
@@ -232,6 +245,10 @@ class ChatTable:
         if chat is None:
             return None
 
+        # Sanitize message content for null characters before upserting
+        if isinstance(message.get("content"), str):
+            message["content"] = message["content"].replace("\x00", "")
+
         chat = chat.chat
         history = chat.get("history", {})
 
@@ -280,6 +297,9 @@ class ChatTable:
                     "user_id": f"shared-{chat_id}",
                     "title": chat.title,
                     "chat": chat.chat,
+                    "meta": chat.meta,
+                    "pinned": chat.pinned,
+                    "folder_id": chat.folder_id,
                     "created_at": chat.created_at,
                     "updated_at": int(time.time()),
                 }
@@ -311,7 +331,9 @@ class ChatTable:
 
                 shared_chat.title = chat.title
                 shared_chat.chat = chat.chat
-
+                shared_chat.meta = chat.meta
+                shared_chat.pinned = chat.pinned
+                shared_chat.folder_id = chat.folder_id
                 shared_chat.updated_at = int(time.time())
                 db.commit()
                 db.refresh(shared_chat)
@@ -580,7 +602,7 @@ class ChatTable:
         """
         Filters chats based on a search query using Python, allowing pagination using skip and limit.
         """
-        search_text = search_text.lower().strip()
+        search_text = search_text.replace("\u0000", "").lower().strip()
 
         if not search_text:
             return self.get_chat_list_by_user_id(
@@ -596,8 +618,45 @@ class ChatTable:
             if word.startswith("tag:")
         ]
 
+        # Extract folder names - handle spaces and case insensitivity
+        folders = Folders.search_folders_by_names(
+            user_id,
+            [
+                word.replace("folder:", "")
+                for word in search_text_words
+                if word.startswith("folder:")
+            ],
+        )
+        folder_ids = [folder.id for folder in folders]
+
+        is_pinned = None
+        if "pinned:true" in search_text_words:
+            is_pinned = True
+        elif "pinned:false" in search_text_words:
+            is_pinned = False
+
+        is_archived = None
+        if "archived:true" in search_text_words:
+            is_archived = True
+        elif "archived:false" in search_text_words:
+            is_archived = False
+
+        is_shared = None
+        if "shared:true" in search_text_words:
+            is_shared = True
+        elif "shared:false" in search_text_words:
+            is_shared = False
+
         search_text_words = [
-            word for word in search_text_words if not word.startswith("tag:")
+            word
+            for word in search_text_words
+            if (
+                not word.startswith("tag:")
+                and not word.startswith("folder:")
+                and not word.startswith("pinned:")
+                and not word.startswith("archived:")
+                and not word.startswith("shared:")
+            )
         ]
 
         search_text = " ".join(search_text_words)
@@ -605,30 +664,41 @@ class ChatTable:
         with get_db() as db:
             query = db.query(Chat).filter(Chat.user_id == user_id)
 
-            if not include_archived:
+            if is_archived is not None:
+                query = query.filter(Chat.archived == is_archived)
+            elif not include_archived:
                 query = query.filter(Chat.archived == False)
 
+            if is_pinned is not None:
+                query = query.filter(Chat.pinned == is_pinned)
+
+            if is_shared is not None:
+                if is_shared:
+                    query = query.filter(Chat.share_id.isnot(None))
+                else:
+                    query = query.filter(Chat.share_id.is_(None))
+
+            if folder_ids:
+                query = query.filter(Chat.folder_id.in_(folder_ids))
+
             query = query.order_by(Chat.updated_at.desc())
 
             # Check if the database dialect is either 'sqlite' or 'postgresql'
             dialect_name = db.bind.dialect.name
             if dialect_name == "sqlite":
                 # SQLite case: using JSON1 extension for JSON searching
+                sqlite_content_sql = (
+                    "EXISTS ("
+                    "    SELECT 1 "
+                    "    FROM json_each(Chat.chat, '$.messages') AS message "
+                    "    WHERE LOWER(message.value->>'content') LIKE '%' || :content_key || '%'"
+                    ")"
+                )
+                sqlite_content_clause = text(sqlite_content_sql)
                 query = query.filter(
-                    (
-                        Chat.title.ilike(
-                            f"%{search_text}%"
-                        )  # Case-insensitive search in title
-                        | text(
-                            """
-                            EXISTS (
-                                SELECT 1 
-                                FROM json_each(Chat.chat, '$.messages') AS message 
-                                WHERE LOWER(message.value->>'content') LIKE '%' || :search_text || '%'
-                            )
-                            """
-                        )
-                    ).params(search_text=search_text)
+                    or_(
+                        Chat.title.ilike(bindparam("title_key")), sqlite_content_clause
+                    ).params(title_key=f"%{search_text}%", content_key=search_text)
                 )
 
                 # Check if there are any tags to filter, it should have all the tags
@@ -663,21 +733,19 @@ class ChatTable:
 
             elif dialect_name == "postgresql":
                 # PostgreSQL relies on proper JSON query for search
+                postgres_content_sql = (
+                    "EXISTS ("
+                    "    SELECT 1 "
+                    "    FROM json_array_elements(Chat.chat->'messages') AS message "
+                    "    WHERE LOWER(message->>'content') LIKE '%' || :content_key || '%'"
+                    ")"
+                )
+                postgres_content_clause = text(postgres_content_sql)
                 query = query.filter(
-                    (
-                        Chat.title.ilike(
-                            f"%{search_text}%"
-                        )  # Case-insensitive search in title
-                        | text(
-                            """
-                            EXISTS (
-                                SELECT 1
-                                FROM json_array_elements(Chat.chat->'messages') AS message
-                                WHERE LOWER(message->>'content') LIKE '%' || :search_text || '%'
-                            )
-                            """
-                        )
-                    ).params(search_text=search_text)
+                    or_(
+                        Chat.title.ilike(bindparam("title_key")),
+                        postgres_content_clause,
+                    ).params(title_key=f"%{search_text}%", content_key=search_text)
                 )
 
                 # Check if there are any tags to filter, it should have all the tags

backend/open_webui/models/folders.py

@@ -2,14 +2,14 @@ import logging
 import time
 import uuid
 from typing import Optional
+import re
+
+
+from pydantic import BaseModel, ConfigDict
+from sqlalchemy import BigInteger, Column, Text, JSON, Boolean, func
 
 from open_webui.internal.db import Base, get_db
-from open_webui.models.chats import Chats
-
 from open_webui.env import SRC_LOG_LEVELS
-from pydantic import BaseModel, ConfigDict
-from sqlalchemy import BigInteger, Column, Text, JSON, Boolean
-from open_webui.utils.access_control import get_permissions
 
 
 log = logging.getLogger(__name__)
@@ -29,6 +29,7 @@ class Folder(Base):
     name = Column(Text)
     items = Column(JSON, nullable=True)
     meta = Column(JSON, nullable=True)
+    data = Column(JSON, nullable=True)
     is_expanded = Column(Boolean, default=False)
     created_at = Column(BigInteger)
     updated_at = Column(BigInteger)
@@ -41,6 +42,7 @@ class FolderModel(BaseModel):
     name: str
     items: Optional[dict] = None
     meta: Optional[dict] = None
+    data: Optional[dict] = None
     is_expanded: bool = False
     created_at: int
     updated_at: int
@@ -55,12 +57,13 @@ class FolderModel(BaseModel):
 
 class FolderForm(BaseModel):
     name: str
+    data: Optional[dict] = None
     model_config = ConfigDict(extra="allow")
 
 
 class FolderTable:
     def insert_new_folder(
-        self, user_id: str, name: str, parent_id: Optional[str] = None
+        self, user_id: str, form_data: FolderForm, parent_id: Optional[str] = None
     ) -> Optional[FolderModel]:
         with get_db() as db:
             id = str(uuid.uuid4())
@@ -68,7 +71,7 @@ class FolderTable:
                 **{
                     "id": id,
                     "user_id": user_id,
-                    "name": name,
+                    **(form_data.model_dump(exclude_unset=True) or {}),
                     "parent_id": parent_id,
                     "created_at": int(time.time()),
                     "updated_at": int(time.time()),
@@ -103,7 +106,7 @@ class FolderTable:
 
     def get_children_folders_by_id_and_user_id(
         self, id: str, user_id: str
-    ) -> Optional[FolderModel]:
+    ) -> Optional[list[FolderModel]]:
         try:
             with get_db() as db:
                 folders = []
@@ -187,8 +190,8 @@ class FolderTable:
             log.error(f"update_folder: {e}")
             return
 
-    def update_folder_name_by_id_and_user_id(
-        self, id: str, user_id: str, name: str
+    def update_folder_by_id_and_user_id(
+        self, id: str, user_id: str, form_data: FolderForm
     ) -> Optional[FolderModel]:
         try:
             with get_db() as db:
@@ -197,16 +200,28 @@ class FolderTable:
                 if not folder:
                     return None
 
+                form_data = form_data.model_dump(exclude_unset=True)
+
                 existing_folder = (
                     db.query(Folder)
-                    .filter_by(name=name, parent_id=folder.parent_id, user_id=user_id)
+                    .filter_by(
+                        name=form_data.get("name"),
+                        parent_id=folder.parent_id,
+                        user_id=user_id,
+                    )
                     .first()
                 )
 
-                if existing_folder:
+                if existing_folder and existing_folder.id != id:
                     return None
 
-                folder.name = name
+                folder.name = form_data.get("name", folder.name)
+                if "data" in form_data:
+                    folder.data = {
+                        **(folder.data or {}),
+                        **form_data["data"],
+                    }
+
                 folder.updated_at = int(time.time())
 
                 db.commit()
@@ -236,18 +251,15 @@ class FolderTable:
             log.error(f"update_folder: {e}")
             return
 
-    def delete_folder_by_id_and_user_id(
-        self, id: str, user_id: str, delete_chats=True
-    ) -> bool:
+    def delete_folder_by_id_and_user_id(self, id: str, user_id: str) -> list[str]:
         try:
+            folder_ids = []
             with get_db() as db:
                 folder = db.query(Folder).filter_by(id=id, user_id=user_id).first()
                 if not folder:
-                    return False
+                    return folder_ids
 
-                if delete_chats:
-                    # Delete all chats in the folder
-                    Chats.delete_chats_by_user_id_and_folder_id(user_id, folder.id)
+                folder_ids.append(folder.id)
 
                 # Delete all children folders
                 def delete_children(folder):
@@ -255,12 +267,9 @@ class FolderTable:
                         folder.id, user_id
                     )
                     for folder_child in folder_children:
-                        if delete_chats:
-                            Chats.delete_chats_by_user_id_and_folder_id(
-                                user_id, folder_child.id
-                            )
 
                         delete_children(folder_child)
+                        folder_ids.append(folder_child.id)
 
                         folder = db.query(Folder).filter_by(id=folder_child.id).first()
                         db.delete(folder)
@@ -269,10 +278,62 @@ class FolderTable:
                 delete_children(folder)
                 db.delete(folder)
                 db.commit()
-                return True
+                return folder_ids
         except Exception as e:
             log.error(f"delete_folder: {e}")
-            return False
+            return []
+
+    def normalize_folder_name(self, name: str) -> str:
+        # Replace _ and space with a single space, lower case, collapse multiple spaces
+        name = re.sub(r"[\s_]+", " ", name)
+        return name.strip().lower()
+
+    def search_folders_by_names(
+        self, user_id: str, queries: list[str]
+    ) -> list[FolderModel]:
+        """
+        Search for folders for a user where the name matches any of the queries, treating _ and space as equivalent, case-insensitive.
+        """
+        normalized_queries = [self.normalize_folder_name(q) for q in queries]
+        if not normalized_queries:
+            return []
+
+        results = {}
+        with get_db() as db:
+            folders = db.query(Folder).filter_by(user_id=user_id).all()
+            for folder in folders:
+                if self.normalize_folder_name(folder.name) in normalized_queries:
+                    results[folder.id] = FolderModel.model_validate(folder)
+
+                    # get children folders
+                    children = self.get_children_folders_by_id_and_user_id(
+                        folder.id, user_id
+                    )
+                    for child in children:
+                        results[child.id] = child
+
+        # Return the results as a list
+        if not results:
+            return []
+        else:
+            results = list(results.values())
+            return results
+
+    def search_folders_by_name_contains(
+        self, user_id: str, query: str
+    ) -> list[FolderModel]:
+        """
+        Partial match: normalized name contains (as substring) the normalized query.
+        """
+        normalized_query = self.normalize_folder_name(query)
+        results = []
+        with get_db() as db:
+            folders = db.query(Folder).filter_by(user_id=user_id).all()
+            for folder in folders:
+                norm_name = self.normalize_folder_name(folder.name)
+                if normalized_query in norm_name:
+                    results.append(FolderModel.model_validate(folder))
+        return results
 
 
 Folders = FolderTable()

backend/open_webui/models/groups.py

@@ -83,10 +83,14 @@ class GroupForm(BaseModel):
     permissions: Optional[dict] = None
 
 
-class GroupUpdateForm(GroupForm):
+class UserIdsForm(BaseModel):
     user_ids: Optional[list[str]] = None
 
 
+class GroupUpdateForm(GroupForm, UserIdsForm):
+    pass
+
+
 class GroupTable:
     def insert_new_group(
         self, user_id: str, form_data: GroupForm
@@ -207,9 +211,39 @@ class GroupTable:
             except Exception:
                 return False
 
-    def sync_user_groups_by_group_names(
+    def create_groups_by_group_names(
         self, user_id: str, group_names: list[str]
-    ) -> bool:
+    ) -> list[GroupModel]:
+
+        # check for existing groups
+        existing_groups = self.get_groups()
+        existing_group_names = {group.name for group in existing_groups}
+
+        new_groups = []
+
+        with get_db() as db:
+            for group_name in group_names:
+                if group_name not in existing_group_names:
+                    new_group = GroupModel(
+                        id=str(uuid.uuid4()),
+                        user_id=user_id,
+                        name=group_name,
+                        description="",
+                        created_at=int(time.time()),
+                        updated_at=int(time.time()),
+                    )
+                    try:
+                        result = Group(**new_group.model_dump())
+                        db.add(result)
+                        db.commit()
+                        db.refresh(result)
+                        new_groups.append(GroupModel.model_validate(result))
+                    except Exception as e:
+                        log.exception(e)
+                        continue
+            return new_groups
+
+    def sync_groups_by_group_names(self, user_id: str, group_names: list[str]) -> bool:
         with get_db() as db:
             try:
                 groups = db.query(Group).filter(Group.name.in_(group_names)).all()
@@ -245,5 +279,53 @@ class GroupTable:
                 log.exception(e)
                 return False
 
+    def add_users_to_group(
+        self, id: str, user_ids: Optional[list[str]] = None
+    ) -> Optional[GroupModel]:
+        try:
+            with get_db() as db:
+                group = db.query(Group).filter_by(id=id).first()
+                if not group:
+                    return None
+
+                if not group.user_ids:
+                    group.user_ids = []
+
+                for user_id in user_ids:
+                    if user_id not in group.user_ids:
+                        group.user_ids.append(user_id)
+
+                group.updated_at = int(time.time())
+                db.commit()
+                db.refresh(group)
+                return GroupModel.model_validate(group)
+        except Exception as e:
+            log.exception(e)
+            return None
+
+    def remove_users_from_group(
+        self, id: str, user_ids: Optional[list[str]] = None
+    ) -> Optional[GroupModel]:
+        try:
+            with get_db() as db:
+                group = db.query(Group).filter_by(id=id).first()
+                if not group:
+                    return None
+
+                if not group.user_ids:
+                    return GroupModel.model_validate(group)
+
+                for user_id in user_ids:
+                    if user_id in group.user_ids:
+                        group.user_ids.remove(user_id)
+
+                group.updated_at = int(time.time())
+                db.commit()
+                db.refresh(group)
+                return GroupModel.model_validate(group)
+        except Exception as e:
+            log.exception(e)
+            return None
+
 
 Groups = GroupTable()

backend/open_webui/models/memories.py

@@ -71,9 +71,13 @@ class MemoriesTable:
     ) -> Optional[MemoryModel]:
         with get_db() as db:
             try:
-                db.query(Memory).filter_by(id=id, user_id=user_id).update(
-                    {"content": content, "updated_at": int(time.time())}
-                )
+                memory = db.get(Memory, id)
+                if not memory or memory.user_id != user_id:
+                    return None
+
+                memory.content = content
+                memory.updated_at = int(time.time())
+
                 db.commit()
                 return self.get_memory_by_id(id)
             except Exception:
@@ -127,7 +131,12 @@ class MemoriesTable:
     def delete_memory_by_id_and_user_id(self, id: str, user_id: str) -> bool:
         with get_db() as db:
             try:
-                db.query(Memory).filter_by(id=id, user_id=user_id).delete()
+                memory = db.get(Memory, id)
+                if not memory or memory.user_id != user_id:
+                    return None
+
+                # Delete the memory
+                db.delete(memory)
                 db.commit()
 
                 return True

backend/open_webui/models/models.py

@@ -269,5 +269,49 @@ class ModelsTable:
         except Exception:
             return False
 
+    def sync_models(self, user_id: str, models: list[ModelModel]) -> list[ModelModel]:
+        try:
+            with get_db() as db:
+                # Get existing models
+                existing_models = db.query(Model).all()
+                existing_ids = {model.id for model in existing_models}
+
+                # Prepare a set of new model IDs
+                new_model_ids = {model.id for model in models}
+
+                # Update or insert models
+                for model in models:
+                    if model.id in existing_ids:
+                        db.query(Model).filter_by(id=model.id).update(
+                            {
+                                **model.model_dump(),
+                                "user_id": user_id,
+                                "updated_at": int(time.time()),
+                            }
+                        )
+                    else:
+                        new_model = Model(
+                            **{
+                                **model.model_dump(),
+                                "user_id": user_id,
+                                "updated_at": int(time.time()),
+                            }
+                        )
+                        db.add(new_model)
+
+                # Remove models that are no longer present
+                for model in existing_models:
+                    if model.id not in new_model_ids:
+                        db.delete(model)
+
+                db.commit()
+
+                return [
+                    ModelModel.model_validate(model) for model in db.query(Model).all()
+                ]
+        except Exception as e:
+            log.exception(f"Error syncing models for user {user_id}: {e}")
+            return []
+
 
 Models = ModelsTable()

backend/open_webui/models/notes.py

@@ -62,6 +62,13 @@ class NoteForm(BaseModel):
     access_control: Optional[dict] = None
 
 
+class NoteUpdateForm(BaseModel):
+    title: Optional[str] = None
+    data: Optional[dict] = None
+    meta: Optional[dict] = None
+    access_control: Optional[dict] = None
+
+
 class NoteUserResponse(NoteModel):
     user: Optional[UserResponse] = None
 
@@ -110,16 +117,26 @@ class NoteTable:
             note = db.query(Note).filter(Note.id == id).first()
             return NoteModel.model_validate(note) if note else None
 
-    def update_note_by_id(self, id: str, form_data: NoteForm) -> Optional[NoteModel]:
+    def update_note_by_id(
+        self, id: str, form_data: NoteUpdateForm
+    ) -> Optional[NoteModel]:
         with get_db() as db:
             note = db.query(Note).filter(Note.id == id).first()
             if not note:
                 return None
 
-            note.title = form_data.title
-            note.data = form_data.data
-            note.meta = form_data.meta
-            note.access_control = form_data.access_control
+            form_data = form_data.model_dump(exclude_unset=True)
+
+            if "title" in form_data:
+                note.title = form_data["title"]
+            if "data" in form_data:
+                note.data = {**note.data, **form_data["data"]}
+            if "meta" in form_data:
+                note.meta = {**note.meta, **form_data["meta"]}
+
+            if "access_control" in form_data:
+                note.access_control = form_data["access_control"]
+
             note.updated_at = int(time.time_ns())
 
             db.commit()

backend/open_webui/models/users.py

@@ -74,6 +74,18 @@ class UserListResponse(BaseModel):
     total: int
 
 
+class UserInfoResponse(BaseModel):
+    id: str
+    name: str
+    email: str
+    role: str
+
+
+class UserInfoListResponse(BaseModel):
+    users: list[UserInfoResponse]
+    total: int
+
+
 class UserResponse(BaseModel):
     id: str
     name: str
@@ -95,6 +107,7 @@ class UserRoleUpdateForm(BaseModel):
 
 
 class UserUpdateForm(BaseModel):
+    role: str
     name: str
     email: str
     profile_image_url: str
@@ -245,6 +258,10 @@ class UsersTable:
         with get_db() as db:
             return db.query(User).count()
 
+    def has_users(self) -> bool:
+        with get_db() as db:
+            return db.query(db.query(User).exists()).scalar()
+
     def get_first_user(self) -> UserModel:
         try:
             with get_db() as db:
@@ -369,7 +386,7 @@ class UsersTable:
         except Exception:
             return False
 
-    def update_user_api_key_by_id(self, id: str, api_key: str) -> str:
+    def update_user_api_key_by_id(self, id: str, api_key: str) -> bool:
         try:
             with get_db() as db:
                 result = db.query(User).filter_by(id=id).update({"api_key": api_key})

backend/open_webui/retrieval/loaders/datalab_marker.py

@@ -15,24 +15,28 @@ class DatalabMarkerLoader:
         self,
         file_path: str,
         api_key: str,
-        langs: Optional[str] = None,
+        api_base_url: str,
+        additional_config: Optional[str] = None,
         use_llm: bool = False,
         skip_cache: bool = False,
         force_ocr: bool = False,
         paginate: bool = False,
         strip_existing_ocr: bool = False,
         disable_image_extraction: bool = False,
+        format_lines: bool = False,
         output_format: str = None,
     ):
         self.file_path = file_path
         self.api_key = api_key
-        self.langs = langs
+        self.api_base_url = api_base_url
+        self.additional_config = additional_config
         self.use_llm = use_llm
         self.skip_cache = skip_cache
         self.force_ocr = force_ocr
         self.paginate = paginate
         self.strip_existing_ocr = strip_existing_ocr
         self.disable_image_extraction = disable_image_extraction
+        self.format_lines = format_lines
         self.output_format = output_format
 
     def _get_mime_type(self, filename: str) -> str:
@@ -60,7 +64,7 @@ class DatalabMarkerLoader:
         return mime_map.get(ext, "application/octet-stream")
 
     def check_marker_request_status(self, request_id: str) -> dict:
-        url = f"https://www.datalab.to/api/v1/marker/{request_id}"
+        url = f"{self.api_base_url}/marker/{request_id}"
         headers = {"X-Api-Key": self.api_key}
         try:
             response = requests.get(url, headers=headers)
@@ -81,22 +85,24 @@ class DatalabMarkerLoader:
             )
 
     def load(self) -> List[Document]:
-        url = "https://www.datalab.to/api/v1/marker"
         filename = os.path.basename(self.file_path)
         mime_type = self._get_mime_type(filename)
         headers = {"X-Api-Key": self.api_key}
 
         form_data = {
-            "langs": self.langs,
             "use_llm": str(self.use_llm).lower(),
             "skip_cache": str(self.skip_cache).lower(),
             "force_ocr": str(self.force_ocr).lower(),
             "paginate": str(self.paginate).lower(),
             "strip_existing_ocr": str(self.strip_existing_ocr).lower(),
             "disable_image_extraction": str(self.disable_image_extraction).lower(),
+            "format_lines": str(self.format_lines).lower(),
             "output_format": self.output_format,
         }
 
+        if self.additional_config and self.additional_config.strip():
+            form_data["additional_config"] = self.additional_config
+
         log.info(
             f"Datalab Marker POST request parameters: {{'filename': '{filename}', 'mime_type': '{mime_type}', **{form_data}}}"
         )
@@ -105,7 +111,10 @@ class DatalabMarkerLoader:
             with open(self.file_path, "rb") as f:
                 files = {"file": (filename, f, mime_type)}
                 response = requests.post(
-                    url, data=form_data, files=files, headers=headers
+                    f"{self.api_base_url}/marker",
+                    data=form_data,
+                    files=files,
+                    headers=headers,
                 )
                 response.raise_for_status()
                 result = response.json()
@@ -133,74 +142,92 @@ class DatalabMarkerLoader:
 
         check_url = result.get("request_check_url")
         request_id = result.get("request_id")
-        if not check_url:
-            raise HTTPException(
-                status.HTTP_502_BAD_GATEWAY, detail="No request_check_url returned."
-            )
 
-        for _ in range(300):  # Up to 10 minutes
-            time.sleep(2)
-            try:
-                poll_response = requests.get(check_url, headers=headers)
-                poll_response.raise_for_status()
-                poll_result = poll_response.json()
-            except (requests.HTTPError, ValueError) as e:
-                raw_body = poll_response.text
-                log.error(f"Polling error: {e}, response body: {raw_body}")
-                raise HTTPException(
-                    status.HTTP_502_BAD_GATEWAY, detail=f"Polling failed: {e}"
-                )
-
-            status_val = poll_result.get("status")
-            success_val = poll_result.get("success")
-
-            if status_val == "complete":
-                summary = {
-                    k: poll_result.get(k)
-                    for k in (
-                        "status",
-                        "output_format",
-                        "success",
-                        "error",
-                        "page_count",
-                        "total_cost",
+        # Check if this is a direct response (self-hosted) or polling response (DataLab)
+        if check_url:
+            # DataLab polling pattern
+            for _ in range(300):  # Up to 10 minutes
+                time.sleep(2)
+                try:
+                    poll_response = requests.get(check_url, headers=headers)
+                    poll_response.raise_for_status()
+                    poll_result = poll_response.json()
+                except (requests.HTTPError, ValueError) as e:
+                    raw_body = poll_response.text
+                    log.error(f"Polling error: {e}, response body: {raw_body}")
+                    raise HTTPException(
+                        status.HTTP_502_BAD_GATEWAY, detail=f"Polling failed: {e}"
                     )
-                }
-                log.info(
-                    f"Marker processing completed successfully: {json.dumps(summary, indent=2)}"
-                )
-                break
 
-            if status_val == "failed" or success_val is False:
-                log.error(
-                    f"Marker poll failed full response: {json.dumps(poll_result, indent=2)}"
-                )
-                error_msg = (
-                    poll_result.get("error")
-                    or "Marker returned failure without error message"
+                status_val = poll_result.get("status")
+                success_val = poll_result.get("success")
+
+                if status_val == "complete":
+                    summary = {
+                        k: poll_result.get(k)
+                        for k in (
+                            "status",
+                            "output_format",
+                            "success",
+                            "error",
+                            "page_count",
+                            "total_cost",
+                        )
+                    }
+                    log.info(
+                        f"Marker processing completed successfully: {json.dumps(summary, indent=2)}"
+                    )
+                    break
+
+                if status_val == "failed" or success_val is False:
+                    log.error(
+                        f"Marker poll failed full response: {json.dumps(poll_result, indent=2)}"
+                    )
+                    error_msg = (
+                        poll_result.get("error")
+                        or "Marker returned failure without error message"
+                    )
+                    raise HTTPException(
+                        status.HTTP_400_BAD_REQUEST,
+                        detail=f"Marker processing failed: {error_msg}",
+                    )
+            else:
+                raise HTTPException(
+                    status.HTTP_504_GATEWAY_TIMEOUT,
+                    detail="Marker processing timed out",
                 )
+
+            if not poll_result.get("success", False):
+                error_msg = poll_result.get("error") or "Unknown processing error"
                 raise HTTPException(
                     status.HTTP_400_BAD_REQUEST,
-                    detail=f"Marker processing failed: {error_msg}",
+                    detail=f"Final processing failed: {error_msg}",
                 )
+
+            # DataLab format - content in format-specific fields
+            content_key = self.output_format.lower()
+            raw_content = poll_result.get(content_key)
+            final_result = poll_result
         else:
-            raise HTTPException(
-                status.HTTP_504_GATEWAY_TIMEOUT, detail="Marker processing timed out"
-            )
+            # Self-hosted direct response - content in "output" field
+            if "output" in result:
+                log.info("Self-hosted Marker returned direct response without polling")
+                raw_content = result.get("output")
+                final_result = result
+            else:
+                available_fields = (
+                    list(result.keys())
+                    if isinstance(result, dict)
+                    else "non-dict response"
+                )
+                raise HTTPException(
+                    status.HTTP_502_BAD_GATEWAY,
+                    detail=f"Custom Marker endpoint returned success but no 'output' field found. Available fields: {available_fields}. Expected either 'request_check_url' for polling or 'output' field for direct response.",
+                )
 
-        if not poll_result.get("success", False):
-            error_msg = poll_result.get("error") or "Unknown processing error"
-            raise HTTPException(
-                status.HTTP_400_BAD_REQUEST,
-                detail=f"Final processing failed: {error_msg}",
-            )
-
-        content_key = self.output_format.lower()
-        raw_content = poll_result.get(content_key)
-
-        if content_key == "json":
+        if self.output_format.lower() == "json":
             full_text = json.dumps(raw_content, indent=2)
-        elif content_key in {"markdown", "html"}:
+        elif self.output_format.lower() in {"markdown", "html"}:
             full_text = str(raw_content).strip()
         else:
             raise HTTPException(
@@ -211,14 +238,14 @@ class DatalabMarkerLoader:
         if not full_text:
             raise HTTPException(
                 status.HTTP_400_BAD_REQUEST,
-                detail="Datalab Marker returned empty content",
+                detail="Marker returned empty content",
             )
 
         marker_output_dir = os.path.join("/app/backend/data/uploads", "marker_output")
         os.makedirs(marker_output_dir, exist_ok=True)
 
         file_ext_map = {"markdown": "md", "json": "json", "html": "html"}
-        file_ext = file_ext_map.get(content_key, "txt")
+        file_ext = file_ext_map.get(self.output_format.lower(), "txt")
         output_filename = f"{os.path.splitext(filename)[0]}.{file_ext}"
         output_path = os.path.join(marker_output_dir, output_filename)
 
@@ -231,13 +258,13 @@ class DatalabMarkerLoader:
 
         metadata = {
             "source": filename,
-            "output_format": poll_result.get("output_format", self.output_format),
-            "page_count": poll_result.get("page_count", 0),
+            "output_format": final_result.get("output_format", self.output_format),
+            "page_count": final_result.get("page_count", 0),
             "processed_with_llm": self.use_llm,
             "request_id": request_id or "",
         }
 
-        images = poll_result.get("images", {})
+        images = final_result.get("images", {})
         if images:
             metadata["image_count"] = len(images)
             metadata["images"] = json.dumps(list(images.keys()))

backend/open_webui/retrieval/loaders/external_document.py

@@ -1,5 +1,5 @@
 import requests
-import logging
+import logging, os
 from typing import Iterator, List, Union
 
 from langchain_core.document_loaders import BaseLoader
@@ -25,7 +25,7 @@ class ExternalDocumentLoader(BaseLoader):
         self.file_path = file_path
         self.mime_type = mime_type
 
-    def load(self) -> list[Document]:
+    def load(self) -> List[Document]:
         with open(self.file_path, "rb") as f:
             data = f.read()
 
@@ -36,23 +36,48 @@ class ExternalDocumentLoader(BaseLoader):
         if self.api_key is not None:
             headers["Authorization"] = f"Bearer {self.api_key}"
 
+        try:
+            headers["X-Filename"] = os.path.basename(self.file_path)
+        except:
+            pass
+
         url = self.url
         if url.endswith("/"):
             url = url[:-1]
 
-        r = requests.put(f"{url}/process", data=data, headers=headers)
+        try:
+            response = requests.put(f"{url}/process", data=data, headers=headers)
+        except Exception as e:
+            log.error(f"Error connecting to endpoint: {e}")
+            raise Exception(f"Error connecting to endpoint: {e}")
 
-        if r.ok:
-            res = r.json()
+        if response.ok:
+
+            response_data = response.json()
+            if response_data:
+                if isinstance(response_data, dict):
+                    return [
+                        Document(
+                            page_content=response_data.get("page_content"),
+                            metadata=response_data.get("metadata"),
+                        )
+                    ]
+                elif isinstance(response_data, list):
+                    documents = []
+                    for document in response_data:
+                        documents.append(
+                            Document(
+                                page_content=document.get("page_content"),
+                                metadata=document.get("metadata"),
+                            )
+                        )
+                    return documents
+                else:
+                    raise Exception("Error loading document: Unable to parse content")
 
-            if res:
-                return [
-                    Document(
-                        page_content=res.get("page_content"),
-                        metadata=res.get("metadata"),
-                    )
-                ]
             else:
                 raise Exception("Error loading document: No content returned")
         else:
-            raise Exception(f"Error loading document: {r.status_code} {r.text}")
+            raise Exception(
+                f"Error loading document: {response.status_code} {response.text}"
+            )

backend/open_webui/retrieval/loaders/main.py

@@ -2,6 +2,7 @@ import requests
 import logging
 import ftfy
 import sys
+import json
 
 from langchain_community.document_loaders import (
     AzureAIDocumentIntelligenceLoader,
@@ -13,7 +14,7 @@ from langchain_community.document_loaders import (
     TextLoader,
     UnstructuredEPubLoader,
     UnstructuredExcelLoader,
-    UnstructuredMarkdownLoader,
+    UnstructuredODTLoader,
     UnstructuredPowerPointLoader,
     UnstructuredRSTLoader,
     UnstructuredXMLLoader,
@@ -76,7 +77,6 @@ known_source_ext = [
     "swift",
     "vue",
     "svelte",
-    "msg",
     "ex",
     "exs",
     "erl",
@@ -147,17 +147,32 @@ class DoclingLoader:
                 )
             }
 
-            params = {
-                "image_export_mode": "placeholder",
-                "table_mode": "accurate",
-            }
+            params = {"image_export_mode": "placeholder", "table_mode": "accurate"}
 
             if self.params:
-                if self.params.get("do_picture_classification"):
-                    params["do_picture_classification"] = self.params.get(
-                        "do_picture_classification"
+                if self.params.get("do_picture_description"):
+                    params["do_picture_description"] = self.params.get(
+                        "do_picture_description"
                     )
 
+                    picture_description_mode = self.params.get(
+                        "picture_description_mode", ""
+                    ).lower()
+
+                    if picture_description_mode == "local" and self.params.get(
+                        "picture_description_local", {}
+                    ):
+                        params["picture_description_local"] = json.dumps(
+                            self.params.get("picture_description_local", {})
+                        )
+
+                    elif picture_description_mode == "api" and self.params.get(
+                        "picture_description_api", {}
+                    ):
+                        params["picture_description_api"] = json.dumps(
+                            self.params.get("picture_description_api", {})
+                        )
+
                 if self.params.get("ocr_engine") and self.params.get("ocr_lang"):
                     params["ocr_engine"] = self.params.get("ocr_engine")
                     params["ocr_lang"] = [
@@ -166,7 +181,7 @@ class DoclingLoader:
                         if lang.strip()
                     ]
 
-            endpoint = f"{self.url}/v1alpha/convert/file"
+            endpoint = f"{self.url}/v1/convert/file"
             r = requests.post(endpoint, files=files, data=params)
 
         if r.ok:
@@ -211,7 +226,10 @@ class Loader:
 
     def _is_text_file(self, file_ext: str, file_content_type: str) -> bool:
         return file_ext in known_source_ext or (
-            file_content_type and file_content_type.find("text/") >= 0
+            file_content_type
+            and file_content_type.find("text/") >= 0
+            # Avoid text/html files being detected as text
+            and not file_content_type.find("html") >= 0
         )
 
     def _get_loader(self, filename: str, file_content_type: str, file_path: str):
@@ -263,10 +281,15 @@ class Loader:
                 "tiff",
             ]
         ):
+            api_base_url = self.kwargs.get("DATALAB_MARKER_API_BASE_URL", "")
+            if not api_base_url or api_base_url.strip() == "":
+                api_base_url = "https://www.datalab.to/api/v1"
+
             loader = DatalabMarkerLoader(
                 file_path=file_path,
                 api_key=self.kwargs["DATALAB_MARKER_API_KEY"],
-                langs=self.kwargs.get("DATALAB_MARKER_LANGS"),
+                api_base_url=api_base_url,
+                additional_config=self.kwargs.get("DATALAB_MARKER_ADDITIONAL_CONFIG"),
                 use_llm=self.kwargs.get("DATALAB_MARKER_USE_LLM", False),
                 skip_cache=self.kwargs.get("DATALAB_MARKER_SKIP_CACHE", False),
                 force_ocr=self.kwargs.get("DATALAB_MARKER_FORCE_OCR", False),
@@ -277,6 +300,7 @@ class Loader:
                 disable_image_extraction=self.kwargs.get(
                     "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION", False
                 ),
+                format_lines=self.kwargs.get("DATALAB_MARKER_FORMAT_LINES", False),
                 output_format=self.kwargs.get(
                     "DATALAB_MARKER_OUTPUT_FORMAT", "markdown"
                 ),
@@ -285,17 +309,20 @@ class Loader:
             if self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:
+                # Build params for DoclingLoader
+                params = self.kwargs.get("DOCLING_PARAMS", {})
+                if not isinstance(params, dict):
+                    try:
+                        params = json.loads(params)
+                    except json.JSONDecodeError:
+                        log.error("Invalid DOCLING_PARAMS format, expected JSON object")
+                        params = {}
+
                 loader = DoclingLoader(
                     url=self.kwargs.get("DOCLING_SERVER_URL"),
                     file_path=file_path,
                     mime_type=file_content_type,
-                    params={
-                        "ocr_engine": self.kwargs.get("DOCLING_OCR_ENGINE"),
-                        "ocr_lang": self.kwargs.get("DOCLING_OCR_LANG"),
-                        "do_picture_classification": self.kwargs.get(
-                            "DOCLING_DO_PICTURE_DESCRIPTION"
-                        ),
-                    },
+                    params=params,
                 )
         elif (
             self.engine == "document_intelligence"
@@ -371,6 +398,8 @@ class Loader:
                 loader = UnstructuredPowerPointLoader(file_path)
             elif file_ext == "msg":
                 loader = OutlookMessageLoader(file_path)
+            elif file_ext == "odt":
+                loader = UnstructuredODTLoader(file_path)
             elif self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:

backend/open_webui/retrieval/loaders/mistral.py

@@ -20,6 +20,14 @@ class MistralLoader:
     """
     Enhanced Mistral OCR loader with both sync and async support.
     Loads documents by processing them through the Mistral OCR API.
+
+    Performance Optimizations:
+    - Differentiated timeouts for different operations
+    - Intelligent retry logic with exponential backoff
+    - Memory-efficient file streaming for large files
+    - Connection pooling and keepalive optimization
+    - Semaphore-based concurrency control for batch processing
+    - Enhanced error handling with retryable error classification
     """
 
     BASE_API_URL = "https://api.mistral.ai/v1"
@@ -53,17 +61,40 @@ class MistralLoader:
         self.max_retries = max_retries
         self.debug = enable_debug_logging
 
-        # Pre-compute file info for performance
+        # PERFORMANCE OPTIMIZATION: Differentiated timeouts for different operations
+        # This prevents long-running OCR operations from affecting quick operations
+        # and improves user experience by failing fast on operations that should be quick
+        self.upload_timeout = min(
+            timeout, 120
+        )  # Cap upload at 2 minutes - prevents hanging on large files
+        self.url_timeout = (
+            30  # URL requests should be fast - fail quickly if API is slow
+        )
+        self.ocr_timeout = (
+            timeout  # OCR can take the full timeout - this is the heavy operation
+        )
+        self.cleanup_timeout = (
+            30  # Cleanup should be quick - don't hang on file deletion
+        )
+
+        # PERFORMANCE OPTIMIZATION: Pre-compute file info to avoid repeated filesystem calls
+        # This avoids multiple os.path.basename() and os.path.getsize() calls during processing
         self.file_name = os.path.basename(file_path)
         self.file_size = os.path.getsize(file_path)
 
+        # ENHANCEMENT: Added User-Agent for better API tracking and debugging
         self.headers = {
             "Authorization": f"Bearer {self.api_key}",
-            "User-Agent": "OpenWebUI-MistralLoader/2.0",
+            "User-Agent": "OpenWebUI-MistralLoader/2.0",  # Helps API provider track usage
         }
 
     def _debug_log(self, message: str, *args) -> None:
-        """Conditional debug logging for performance."""
+        """
+        PERFORMANCE OPTIMIZATION: Conditional debug logging for performance.
+
+        Only processes debug messages when debug mode is enabled, avoiding
+        string formatting overhead in production environments.
+        """
         if self.debug:
             log.debug(message, *args)
 
@@ -115,53 +146,118 @@ class MistralLoader:
             log.error(f"Unexpected error processing response: {e}")
             raise
 
+    def _is_retryable_error(self, error: Exception) -> bool:
+        """
+        ENHANCEMENT: Intelligent error classification for retry logic.
+
+        Determines if an error is retryable based on its type and status code.
+        This prevents wasting time retrying errors that will never succeed
+        (like authentication errors) while ensuring transient errors are retried.
+
+        Retryable errors:
+        - Network connection errors (temporary network issues)
+        - Timeouts (server might be temporarily overloaded)
+        - Server errors (5xx status codes - server-side issues)
+        - Rate limiting (429 status - temporary throttling)
+
+        Non-retryable errors:
+        - Authentication errors (401, 403 - won't fix with retry)
+        - Bad request errors (400 - malformed request)
+        - Not found errors (404 - resource doesn't exist)
+        """
+        if isinstance(error, requests.exceptions.ConnectionError):
+            return True  # Network issues are usually temporary
+        if isinstance(error, requests.exceptions.Timeout):
+            return True  # Timeouts might resolve on retry
+        if isinstance(error, requests.exceptions.HTTPError):
+            # Only retry on server errors (5xx) or rate limits (429)
+            if hasattr(error, "response") and error.response is not None:
+                status_code = error.response.status_code
+                return status_code >= 500 or status_code == 429
+            return False
+        if isinstance(
+            error, (aiohttp.ClientConnectionError, aiohttp.ServerTimeoutError)
+        ):
+            return True  # Async network/timeout errors are retryable
+        if isinstance(error, aiohttp.ClientResponseError):
+            return error.status >= 500 or error.status == 429
+        return False  # All other errors are non-retryable
+
     def _retry_request_sync(self, request_func, *args, **kwargs):
-        """Synchronous retry logic with exponential backoff."""
+        """
+        ENHANCEMENT: Synchronous retry logic with intelligent error classification.
+
+        Uses exponential backoff with jitter to avoid thundering herd problems.
+        The wait time increases exponentially but is capped at 30 seconds to
+        prevent excessive delays. Only retries errors that are likely to succeed
+        on subsequent attempts.
+        """
         for attempt in range(self.max_retries):
             try:
                 return request_func(*args, **kwargs)
-            except (requests.exceptions.RequestException, Exception) as e:
-                if attempt == self.max_retries - 1:
+            except Exception as e:
+                if attempt == self.max_retries - 1 or not self._is_retryable_error(e):
                     raise
 
-                wait_time = (2**attempt) + 0.5
+                # PERFORMANCE OPTIMIZATION: Exponential backoff with cap
+                # Prevents overwhelming the server while ensuring reasonable retry delays
+                wait_time = min((2**attempt) + 0.5, 30)  # Cap at 30 seconds
                 log.warning(
-                    f"Request failed (attempt {attempt + 1}/{self.max_retries}): {e}. Retrying in {wait_time}s..."
+                    f"Retryable error (attempt {attempt + 1}/{self.max_retries}): {e}. "
+                    f"Retrying in {wait_time}s..."
                 )
                 time.sleep(wait_time)
 
     async def _retry_request_async(self, request_func, *args, **kwargs):
-        """Async retry logic with exponential backoff."""
+        """
+        ENHANCEMENT: Async retry logic with intelligent error classification.
+
+        Async version of retry logic that doesn't block the event loop during
+        wait periods. Uses the same exponential backoff strategy as sync version.
+        """
         for attempt in range(self.max_retries):
             try:
                 return await request_func(*args, **kwargs)
-            except (aiohttp.ClientError, asyncio.TimeoutError) as e:
-                if attempt == self.max_retries - 1:
+            except Exception as e:
+                if attempt == self.max_retries - 1 or not self._is_retryable_error(e):
                     raise
 
-                wait_time = (2**attempt) + 0.5
+                # PERFORMANCE OPTIMIZATION: Non-blocking exponential backoff
+                wait_time = min((2**attempt) + 0.5, 30)  # Cap at 30 seconds
                 log.warning(
-                    f"Request failed (attempt {attempt + 1}/{self.max_retries}): {e}. Retrying in {wait_time}s..."
+                    f"Retryable error (attempt {attempt + 1}/{self.max_retries}): {e}. "
+                    f"Retrying in {wait_time}s..."
                 )
-                await asyncio.sleep(wait_time)
+                await asyncio.sleep(wait_time)  # Non-blocking wait
 
     def _upload_file(self) -> str:
-        """Uploads the file to Mistral for OCR processing (sync version)."""
+        """
+        PERFORMANCE OPTIMIZATION: Enhanced file upload with streaming consideration.
+
+        Uploads the file to Mistral for OCR processing (sync version).
+        Uses context manager for file handling to ensure proper resource cleanup.
+        Although streaming is not enabled for this endpoint, the file is opened
+        in a context manager to minimize memory usage duration.
+        """
         log.info("Uploading file to Mistral API")
         url = f"{self.BASE_API_URL}/files"
-        file_name = os.path.basename(self.file_path)
 
         def upload_request():
+            # MEMORY OPTIMIZATION: Use context manager to minimize file handle lifetime
+            # This ensures the file is closed immediately after reading, reducing memory usage
             with open(self.file_path, "rb") as f:
-                files = {"file": (file_name, f, "application/pdf")}
+                files = {"file": (self.file_name, f, "application/pdf")}
                 data = {"purpose": "ocr"}
 
+                # NOTE: stream=False is required for this endpoint
+                # The Mistral API doesn't support chunked uploads for this endpoint
                 response = requests.post(
                     url,
                     headers=self.headers,
                     files=files,
                     data=data,
-                    timeout=self.timeout,
+                    timeout=self.upload_timeout,  # Use specialized upload timeout
+                    stream=False,  # Keep as False for this endpoint
                 )
 
             return self._handle_response(response)
@@ -209,7 +305,7 @@ class MistralLoader:
                 url,
                 data=writer,
                 headers=self.headers,
-                timeout=aiohttp.ClientTimeout(total=self.timeout),
+                timeout=aiohttp.ClientTimeout(total=self.upload_timeout),
             ) as response:
                 return await self._handle_response_async(response)
 
@@ -231,7 +327,7 @@ class MistralLoader:
 
         def url_request():
             response = requests.get(
-                url, headers=signed_url_headers, params=params, timeout=self.timeout
+                url, headers=signed_url_headers, params=params, timeout=self.url_timeout
             )
             return self._handle_response(response)
 
@@ -261,7 +357,7 @@ class MistralLoader:
                 url,
                 headers=headers,
                 params=params,
-                timeout=aiohttp.ClientTimeout(total=self.timeout),
+                timeout=aiohttp.ClientTimeout(total=self.url_timeout),
             ) as response:
                 return await self._handle_response_async(response)
 
@@ -294,7 +390,7 @@ class MistralLoader:
 
         def ocr_request():
             response = requests.post(
-                url, headers=ocr_headers, json=payload, timeout=self.timeout
+                url, headers=ocr_headers, json=payload, timeout=self.ocr_timeout
             )
             return self._handle_response(response)
 
@@ -336,7 +432,7 @@ class MistralLoader:
                 url,
                 json=payload,
                 headers=headers,
-                timeout=aiohttp.ClientTimeout(total=self.timeout),
+                timeout=aiohttp.ClientTimeout(total=self.ocr_timeout),
             ) as response:
                 ocr_response = await self._handle_response_async(response)
 
@@ -353,7 +449,9 @@ class MistralLoader:
         url = f"{self.BASE_API_URL}/files/{file_id}"
 
         try:
-            response = requests.delete(url, headers=self.headers, timeout=30)
+            response = requests.delete(
+                url, headers=self.headers, timeout=self.cleanup_timeout
+            )
             delete_response = self._handle_response(response)
             log.info(f"File deleted successfully: {delete_response}")
         except Exception as e:
@@ -372,7 +470,7 @@ class MistralLoader:
                     url=f"{self.BASE_API_URL}/files/{file_id}",
                     headers=self.headers,
                     timeout=aiohttp.ClientTimeout(
-                        total=30
+                        total=self.cleanup_timeout
                     ),  # Shorter timeout for cleanup
                 ) as response:
                     return await self._handle_response_async(response)
@@ -388,29 +486,40 @@ class MistralLoader:
     async def _get_session(self):
         """Context manager for HTTP session with optimized settings."""
         connector = aiohttp.TCPConnector(
-            limit=10,  # Total connection limit
-            limit_per_host=5,  # Per-host connection limit
-            ttl_dns_cache=300,  # DNS cache TTL
+            limit=20,  # Increased total connection limit for better throughput
+            limit_per_host=10,  # Increased per-host limit for API endpoints
+            ttl_dns_cache=600,  # Longer DNS cache TTL (10 minutes)
             use_dns_cache=True,
-            keepalive_timeout=30,
+            keepalive_timeout=60,  # Increased keepalive for connection reuse
             enable_cleanup_closed=True,
+            force_close=False,  # Allow connection reuse
+            resolver=aiohttp.AsyncResolver(),  # Use async DNS resolver
+        )
+
+        timeout = aiohttp.ClientTimeout(
+            total=self.timeout,
+            connect=30,  # Connection timeout
+            sock_read=60,  # Socket read timeout
         )
 
         async with aiohttp.ClientSession(
             connector=connector,
-            timeout=aiohttp.ClientTimeout(total=self.timeout),
+            timeout=timeout,
             headers={"User-Agent": "OpenWebUI-MistralLoader/2.0"},
+            raise_for_status=False,  # We handle status codes manually
+            trust_env=True,
         ) as session:
             yield session
 
     def _process_results(self, ocr_response: Dict[str, Any]) -> List[Document]:
-        """Process OCR results into Document objects with enhanced metadata."""
+        """Process OCR results into Document objects with enhanced metadata and memory efficiency."""
         pages_data = ocr_response.get("pages")
         if not pages_data:
             log.warning("No pages found in OCR response.")
             return [
                 Document(
-                    page_content="No text content found", metadata={"error": "no_pages"}
+                    page_content="No text content found",
+                    metadata={"error": "no_pages", "file_name": self.file_name},
                 )
             ]
 
@@ -418,41 +527,44 @@ class MistralLoader:
         total_pages = len(pages_data)
         skipped_pages = 0
 
+        # Process pages in a memory-efficient way
         for page_data in pages_data:
             page_content = page_data.get("markdown")
             page_index = page_data.get("index")  # API uses 0-based index
 
-            if page_content is not None and page_index is not None:
-                # Clean up content efficiently
-                cleaned_content = (
-                    page_content.strip()
-                    if isinstance(page_content, str)
-                    else str(page_content)
-                )
-
-                if cleaned_content:  # Only add non-empty pages
-                    documents.append(
-                        Document(
-                            page_content=cleaned_content,
-                            metadata={
-                                "page": page_index,  # 0-based index from API
-                                "page_label": page_index
-                                + 1,  # 1-based label for convenience
-                                "total_pages": total_pages,
-                                "file_name": self.file_name,
-                                "file_size": self.file_size,
-                                "processing_engine": "mistral-ocr",
-                            },
-                        )
-                    )
-                else:
-                    skipped_pages += 1
-                    self._debug_log(f"Skipping empty page {page_index}")
-            else:
+            if page_content is None or page_index is None:
                 skipped_pages += 1
                 self._debug_log(
-                    f"Skipping page due to missing 'markdown' or 'index'. Data: {page_data}"
+                    f"Skipping page due to missing 'markdown' or 'index'. Data keys: {list(page_data.keys())}"
                 )
+                continue
+
+            # Clean up content efficiently with early exit for empty content
+            if isinstance(page_content, str):
+                cleaned_content = page_content.strip()
+            else:
+                cleaned_content = str(page_content).strip()
+
+            if not cleaned_content:
+                skipped_pages += 1
+                self._debug_log(f"Skipping empty page {page_index}")
+                continue
+
+            # Create document with optimized metadata
+            documents.append(
+                Document(
+                    page_content=cleaned_content,
+                    metadata={
+                        "page": page_index,  # 0-based index from API
+                        "page_label": page_index + 1,  # 1-based label for convenience
+                        "total_pages": total_pages,
+                        "file_name": self.file_name,
+                        "file_size": self.file_size,
+                        "processing_engine": "mistral-ocr",
+                        "content_length": len(cleaned_content),
+                    },
+                )
+            )
 
         if skipped_pages > 0:
             log.info(
@@ -467,7 +579,11 @@ class MistralLoader:
             return [
                 Document(
                     page_content="No valid text content found in document",
-                    metadata={"error": "no_valid_pages", "total_pages": total_pages},
+                    metadata={
+                        "error": "no_valid_pages",
+                        "total_pages": total_pages,
+                        "file_name": self.file_name,
+                    },
                 )
             ]
 
@@ -585,12 +701,14 @@ class MistralLoader:
     @staticmethod
     async def load_multiple_async(
         loaders: List["MistralLoader"],
+        max_concurrent: int = 5,  # Limit concurrent requests
     ) -> List[List[Document]]:
         """
-        Process multiple files concurrently for maximum performance.
+        Process multiple files concurrently with controlled concurrency.
 
         Args:
             loaders: List of MistralLoader instances
+            max_concurrent: Maximum number of concurrent requests
 
         Returns:
             List of document lists, one for each loader
@@ -598,11 +716,20 @@ class MistralLoader:
         if not loaders:
             return []
 
-        log.info(f"Starting concurrent processing of {len(loaders)} files")
+        log.info(
+            f"Starting concurrent processing of {len(loaders)} files with max {max_concurrent} concurrent"
+        )
         start_time = time.time()
 
-        # Process all files concurrently
-        tasks = [loader.load_async() for loader in loaders]
+        # Use semaphore to control concurrency
+        semaphore = asyncio.Semaphore(max_concurrent)
+
+        async def process_with_semaphore(loader: "MistralLoader") -> List[Document]:
+            async with semaphore:
+                return await loader.load_async()
+
+        # Process all files with controlled concurrency
+        tasks = [process_with_semaphore(loader) for loader in loaders]
         results = await asyncio.gather(*tasks, return_exceptions=True)
 
         # Handle any exceptions in results
@@ -624,10 +751,18 @@ class MistralLoader:
             else:
                 processed_results.append(result)
 
+        # MONITORING: Log comprehensive batch processing statistics
         total_time = time.time() - start_time
         total_docs = sum(len(docs) for docs in processed_results)
+        success_count = sum(
+            1 for result in results if not isinstance(result, Exception)
+        )
+        failure_count = len(results) - success_count
+
         log.info(
-            f"Batch processing completed in {total_time:.2f}s, produced {total_docs} total documents"
+            f"Batch processing completed in {total_time:.2f}s: "
+            f"{success_count} files succeeded, {failure_count} files failed, "
+            f"produced {total_docs} total documents"
         )
 
         return processed_results

backend/open_webui/retrieval/loaders/youtube.py

@@ -1,4 +1,5 @@
 import logging
+from xml.etree.ElementTree import ParseError
 
 from typing import Any, Dict, Generator, List, Optional, Sequence, Union
 from urllib.parse import parse_qs, urlparse
@@ -93,7 +94,6 @@ class YoutubeLoader:
                 "http": self.proxy_url,
                 "https": self.proxy_url,
             }
-            # Don't log complete URL because it might contain secrets
             log.debug(f"Using proxy URL: {self.proxy_url[:14]}...")
         else:
             youtube_proxies = None
@@ -110,11 +110,37 @@ class YoutubeLoader:
         for lang in self.language:
             try:
                 transcript = transcript_list.find_transcript([lang])
+                if transcript.is_generated:
+                    log.debug(f"Found generated transcript for language '{lang}'")
+                    try:
+                        transcript = transcript_list.find_manually_created_transcript(
+                            [lang]
+                        )
+                        log.debug(f"Found manual transcript for language '{lang}'")
+                    except NoTranscriptFound:
+                        log.debug(
+                            f"No manual transcript found for language '{lang}', using generated"
+                        )
+                        pass
+
                 log.debug(f"Found transcript for language '{lang}'")
-                transcript_pieces: List[Dict[str, Any]] = transcript.fetch()
+                try:
+                    transcript_pieces: List[Dict[str, Any]] = transcript.fetch()
+                except ParseError:
+                    log.debug(f"Empty or invalid transcript for language '{lang}'")
+                    continue
+
+                if not transcript_pieces:
+                    log.debug(f"Empty transcript for language '{lang}'")
+                    continue
+
                 transcript_text = " ".join(
                     map(
-                        lambda transcript_piece: transcript_piece.text.strip(" "),
+                        lambda transcript_piece: (
+                            transcript_piece.text.strip(" ")
+                            if hasattr(transcript_piece, "text")
+                            else ""
+                        ),
                         transcript_pieces,
                     )
                 )
@@ -131,6 +157,4 @@ class YoutubeLoader:
         log.warning(
             f"No transcript found for any of the specified languages: {languages_tried}. Verify if the video has transcripts, add more languages if needed."
         )
-        raise NoTranscriptFound(
-            f"No transcript found for any supported language. Verify if the video has transcripts, add more languages if needed."
-        )
+        raise NoTranscriptFound(self.video_id, self.language, list(transcript_list))

backend/open_webui/retrieval/models/external.py

@@ -1,8 +1,10 @@
 import logging
 import requests
 from typing import Optional, List, Tuple
+from urllib.parse import quote
 
-from open_webui.env import SRC_LOG_LEVELS
+
+from open_webui.env import ENABLE_FORWARD_USER_INFO_HEADERS, SRC_LOG_LEVELS
 from open_webui.retrieval.models.base_reranker import BaseReranker
 
 
@@ -21,7 +23,9 @@ class ExternalReranker(BaseReranker):
         self.url = url
         self.model = model
 
-    def predict(self, sentences: List[Tuple[str, str]]) -> Optional[List[float]]:
+    def predict(
+        self, sentences: List[Tuple[str, str]], user=None
+    ) -> Optional[List[float]]:
         query = sentences[0][0]
         docs = [i[1] for i in sentences]
 
@@ -41,6 +45,16 @@ class ExternalReranker(BaseReranker):
                 headers={
                     "Content-Type": "application/json",
                     "Authorization": f"Bearer {self.api_key}",
+                    **(
+                        {
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
+                            "X-OpenWebUI-User-Id": user.id,
+                            "X-OpenWebUI-User-Email": user.email,
+                            "X-OpenWebUI-User-Role": user.role,
+                        }
+                        if ENABLE_FORWARD_USER_INFO_HEADERS and user
+                        else {}
+                    ),
                 },
                 json=payload,
             )

backend/open_webui/retrieval/utils.py

@@ -5,7 +5,9 @@ from typing import Optional, Union
 import requests
 import hashlib
 from concurrent.futures import ThreadPoolExecutor
+import time
 
+from urllib.parse import quote
 from huggingface_hub import snapshot_download
 from langchain.retrievers import ContextualCompressionRetriever, EnsembleRetriever
 from langchain_community.retrievers import BM25Retriever
@@ -16,8 +18,11 @@ from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT
 
 from open_webui.models.users import UserModel
 from open_webui.models.files import Files
+from open_webui.models.knowledge import Knowledges
+from open_webui.models.notes import Notes
 
 from open_webui.retrieval.vector.main import GetResult
+from open_webui.utils.access_control import has_access
 
 
 from open_webui.env import (
@@ -400,12 +405,13 @@ def get_embedding_function(
     url,
     key,
     embedding_batch_size,
+    azure_api_version=None,
 ):
     if embedding_engine == "":
         return lambda query, prefix=None, user=None: embedding_function.encode(
             query, **({"prompt": prefix} if prefix else {})
         ).tolist()
-    elif embedding_engine in ["ollama", "openai"]:
+    elif embedding_engine in ["ollama", "openai", "azure_openai"]:
         func = lambda query, prefix=None, user=None: generate_embeddings(
             engine=embedding_engine,
             model=embedding_model,
@@ -414,6 +420,7 @@ def get_embedding_function(
             url=url,
             key=key,
             user=user,
+            azure_api_version=azure_api_version,
         )
 
         def generate_multiple(query, prefix, user, func):
@@ -438,9 +445,20 @@ def get_embedding_function(
         raise ValueError(f"Unknown embedding engine: {embedding_engine}")
 
 
-def get_sources_from_files(
+def get_reranking_function(reranking_engine, reranking_model, reranking_function):
+    if reranking_function is None:
+        return None
+    if reranking_engine == "external":
+        return lambda sentences, user=None: reranking_function.predict(
+            sentences, user=user
+        )
+    else:
+        return lambda sentences, user=None: reranking_function.predict(sentences)
+
+
+def get_sources_from_items(
     request,
-    files,
+    items,
     queries,
     embedding_function,
     k,
@@ -450,159 +468,217 @@ def get_sources_from_files(
     hybrid_bm25_weight,
     hybrid_search,
     full_context=False,
+    user: Optional[UserModel] = None,
 ):
     log.debug(
-        f"files: {files} {queries} {embedding_function} {reranking_function} {full_context}"
+        f"items: {items} {queries} {embedding_function} {reranking_function} {full_context}"
     )
 
     extracted_collections = []
-    relevant_contexts = []
+    query_results = []
 
-    for file in files:
+    for item in items:
+        query_result = None
+        collection_names = []
 
-        context = None
-        if file.get("docs"):
-            # BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL
-            context = {
-                "documents": [[doc.get("content") for doc in file.get("docs")]],
-                "metadatas": [[doc.get("metadata") for doc in file.get("docs")]],
-            }
-        elif file.get("context") == "full":
-            # Manual Full Mode Toggle
-            context = {
-                "documents": [[file.get("file").get("data", {}).get("content")]],
-                "metadatas": [[{"file_id": file.get("id"), "name": file.get("name")}]],
-            }
-        elif (
-            file.get("type") != "web_search"
-            and request.app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL
-        ):
-            # BYPASS_EMBEDDING_AND_RETRIEVAL
-            if file.get("type") == "collection":
-                file_ids = file.get("data", {}).get("file_ids", [])
+        if item.get("type") == "text":
+            # Raw Text
+            # Used during temporary chat file uploads
 
-                documents = []
-                metadatas = []
-                for file_id in file_ids:
-                    file_object = Files.get_file_by_id(file_id)
-
-                    if file_object:
-                        documents.append(file_object.data.get("content", ""))
-                        metadatas.append(
-                            {
-                                "file_id": file_id,
-                                "name": file_object.filename,
-                                "source": file_object.filename,
-                            }
-                        )
-
-                context = {
-                    "documents": [documents],
-                    "metadatas": [metadatas],
+            if item.get("file"):
+                # if item has file data, use it
+                query_result = {
+                    "documents": [
+                        [item.get("file", {}).get("data", {}).get("content")]
+                    ],
+                    "metadatas": [
+                        [item.get("file", {}).get("data", {}).get("meta", {})]
+                    ],
+                }
+            else:
+                # Fallback to item content
+                query_result = {
+                    "documents": [[item.get("content")]],
+                    "metadatas": [
+                        [{"file_id": item.get("id"), "name": item.get("name")}]
+                    ],
                 }
 
-            elif file.get("id"):
-                file_object = Files.get_file_by_id(file.get("id"))
-                if file_object:
-                    context = {
-                        "documents": [[file_object.data.get("content", "")]],
+        elif item.get("type") == "note":
+            # Note Attached
+            note = Notes.get_note_by_id(item.get("id"))
+
+            if note and (
+                user.role == "admin"
+                or note.user_id == user.id
+                or has_access(user.id, "read", note.access_control)
+            ):
+                # User has access to the note
+                query_result = {
+                    "documents": [[note.data.get("content", {}).get("md", "")]],
+                    "metadatas": [[{"file_id": note.id, "name": note.title}]],
+                }
+
+        elif item.get("type") == "file":
+            if (
+                item.get("context") == "full"
+                or request.app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL
+            ):
+                if item.get("file", {}).get("data", {}).get("content", ""):
+                    # Manual Full Mode Toggle
+                    # Used from chat file modal, we can assume that the file content will be available from item.get("file").get("data", {}).get("content")
+                    query_result = {
+                        "documents": [
+                            [item.get("file", {}).get("data", {}).get("content", "")]
+                        ],
                         "metadatas": [
                             [
                                 {
-                                    "file_id": file.get("id"),
-                                    "name": file_object.filename,
-                                    "source": file_object.filename,
+                                    "file_id": item.get("id"),
+                                    "name": item.get("name"),
+                                    **item.get("file")
+                                    .get("data", {})
+                                    .get("metadata", {}),
                                 }
                             ]
                         ],
                     }
-            elif file.get("file").get("data"):
-                context = {
-                    "documents": [[file.get("file").get("data", {}).get("content")]],
-                    "metadatas": [
-                        [file.get("file").get("data", {}).get("metadata", {})]
-                    ],
-                }
-        else:
-            collection_names = []
-            if file.get("type") == "collection":
-                if file.get("legacy"):
-                    collection_names = file.get("collection_names", [])
+                elif item.get("id"):
+                    file_object = Files.get_file_by_id(item.get("id"))
+                    if file_object:
+                        query_result = {
+                            "documents": [[file_object.data.get("content", "")]],
+                            "metadatas": [
+                                [
+                                    {
+                                        "file_id": item.get("id"),
+                                        "name": file_object.filename,
+                                        "source": file_object.filename,
+                                    }
+                                ]
+                            ],
+                        }
+            else:
+                # Fallback to collection names
+                if item.get("legacy"):
+                    collection_names.append(f"{item['id']}")
                 else:
-                    collection_names.append(file["id"])
-            elif file.get("collection_name"):
-                collection_names.append(file["collection_name"])
-            elif file.get("id"):
-                if file.get("legacy"):
-                    collection_names.append(f"{file['id']}")
-                else:
-                    collection_names.append(f"file-{file['id']}")
+                    collection_names.append(f"file-{item['id']}")
 
+        elif item.get("type") == "collection":
+            if (
+                item.get("context") == "full"
+                or request.app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL
+            ):
+                # Manual Full Mode Toggle for Collection
+                knowledge_base = Knowledges.get_knowledge_by_id(item.get("id"))
+
+                if knowledge_base and (
+                    user.role == "admin"
+                    or has_access(user.id, "read", knowledge_base.access_control)
+                ):
+
+                    file_ids = knowledge_base.data.get("file_ids", [])
+
+                    documents = []
+                    metadatas = []
+                    for file_id in file_ids:
+                        file_object = Files.get_file_by_id(file_id)
+
+                        if file_object:
+                            documents.append(file_object.data.get("content", ""))
+                            metadatas.append(
+                                {
+                                    "file_id": file_id,
+                                    "name": file_object.filename,
+                                    "source": file_object.filename,
+                                }
+                            )
+
+                    query_result = {
+                        "documents": [documents],
+                        "metadatas": [metadatas],
+                    }
+            else:
+                # Fallback to collection names
+                if item.get("legacy"):
+                    collection_names = item.get("collection_names", [])
+                else:
+                    collection_names.append(item["id"])
+
+        elif item.get("docs"):
+            # BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL
+            query_result = {
+                "documents": [[doc.get("content") for doc in item.get("docs")]],
+                "metadatas": [[doc.get("metadata") for doc in item.get("docs")]],
+            }
+        elif item.get("collection_name"):
+            # Direct Collection Name
+            collection_names.append(item["collection_name"])
+        elif item.get("collection_names"):
+            # Collection Names List
+            collection_names.extend(item["collection_names"])
+
+        # If query_result is None
+        # Fallback to collection names and vector search the collections
+        if query_result is None and collection_names:
             collection_names = set(collection_names).difference(extracted_collections)
             if not collection_names:
-                log.debug(f"skipping {file} as it has already been extracted")
+                log.debug(f"skipping {item} as it has already been extracted")
                 continue
 
-            if full_context:
-                try:
-                    context = get_all_items_from_collections(collection_names)
-                except Exception as e:
-                    log.exception(e)
-
-            else:
-                try:
-                    context = None
-                    if file.get("type") == "text":
-                        context = file["content"]
-                    else:
-                        if hybrid_search:
-                            try:
-                                context = query_collection_with_hybrid_search(
-                                    collection_names=collection_names,
-                                    queries=queries,
-                                    embedding_function=embedding_function,
-                                    k=k,
-                                    reranking_function=reranking_function,
-                                    k_reranker=k_reranker,
-                                    r=r,
-                                    hybrid_bm25_weight=hybrid_bm25_weight,
-                                )
-                            except Exception as e:
-                                log.debug(
-                                    "Error when using hybrid search, using"
-                                    " non hybrid search as fallback."
-                                )
-
-                        if (not hybrid_search) or (context is None):
-                            context = query_collection(
+            try:
+                if full_context:
+                    query_result = get_all_items_from_collections(collection_names)
+                else:
+                    query_result = None  # Initialize to None
+                    if hybrid_search:
+                        try:
+                            query_result = query_collection_with_hybrid_search(
                                 collection_names=collection_names,
                                 queries=queries,
                                 embedding_function=embedding_function,
                                 k=k,
+                                reranking_function=reranking_function,
+                                k_reranker=k_reranker,
+                                r=r,
+                                hybrid_bm25_weight=hybrid_bm25_weight,
                             )
-                except Exception as e:
-                    log.exception(e)
+                        except Exception as e:
+                            log.debug(
+                                "Error when using hybrid search, using non hybrid search as fallback."
+                            )
+
+                    # fallback to non-hybrid search
+                    if not hybrid_search and query_result is None:
+                        query_result = query_collection(
+                            collection_names=collection_names,
+                            queries=queries,
+                            embedding_function=embedding_function,
+                            k=k,
+                        )
+            except Exception as e:
+                log.exception(e)
 
             extracted_collections.extend(collection_names)
 
-        if context:
-            if "data" in file:
-                del file["data"]
-
-            relevant_contexts.append({**context, "file": file})
+        if query_result:
+            if "data" in item:
+                del item["data"]
+            query_results.append({**query_result, "file": item})
 
     sources = []
-    for context in relevant_contexts:
+    for query_result in query_results:
         try:
-            if "documents" in context:
-                if "metadatas" in context:
+            if "documents" in query_result:
+                if "metadatas" in query_result:
                     source = {
-                        "source": context["file"],
-                        "document": context["documents"][0],
-                        "metadata": context["metadatas"][0],
+                        "source": query_result["file"],
+                        "document": query_result["documents"][0],
+                        "metadata": query_result["metadatas"][0],
                     }
-                    if "distances" in context and context["distances"]:
-                        source["distances"] = context["distances"][0]
+                    if "distances" in query_result and query_result["distances"]:
+                        source["distances"] = query_result["distances"][0]
 
                     sources.append(source)
         except Exception as e:
@@ -675,7 +751,7 @@ def generate_openai_batch_embeddings(
                 "Authorization": f"Bearer {key}",
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -697,6 +773,60 @@ def generate_openai_batch_embeddings(
         return None
 
 
+def generate_azure_openai_batch_embeddings(
+    model: str,
+    texts: list[str],
+    url: str,
+    key: str = "",
+    version: str = "",
+    prefix: str = None,
+    user: UserModel = None,
+) -> Optional[list[list[float]]]:
+    try:
+        log.debug(
+            f"generate_azure_openai_batch_embeddings:deployment {model} batch size: {len(texts)}"
+        )
+        json_data = {"input": texts}
+        if isinstance(RAG_EMBEDDING_PREFIX_FIELD_NAME, str) and isinstance(prefix, str):
+            json_data[RAG_EMBEDDING_PREFIX_FIELD_NAME] = prefix
+
+        url = f"{url}/openai/deployments/{model}/embeddings?api-version={version}"
+
+        for _ in range(5):
+            r = requests.post(
+                url,
+                headers={
+                    "Content-Type": "application/json",
+                    "api-key": key,
+                    **(
+                        {
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
+                            "X-OpenWebUI-User-Id": user.id,
+                            "X-OpenWebUI-User-Email": user.email,
+                            "X-OpenWebUI-User-Role": user.role,
+                        }
+                        if ENABLE_FORWARD_USER_INFO_HEADERS and user
+                        else {}
+                    ),
+                },
+                json=json_data,
+            )
+            if r.status_code == 429:
+                retry = float(r.headers.get("Retry-After", "1"))
+                time.sleep(retry)
+                continue
+            r.raise_for_status()
+            data = r.json()
+            if "data" in data:
+                return [elem["embedding"] for elem in data["data"]]
+            else:
+                raise Exception("Something went wrong :/")
+        return None
+    except Exception as e:
+        log.exception(f"Error generating azure openai batch embeddings: {e}")
+        return None
+
+
 def generate_ollama_batch_embeddings(
     model: str,
     texts: list[str],
@@ -720,7 +850,7 @@ def generate_ollama_batch_embeddings(
                 "Authorization": f"Bearer {key}",
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -761,38 +891,33 @@ def generate_embeddings(
             text = f"{prefix}{text}"
 
     if engine == "ollama":
-        if isinstance(text, list):
-            embeddings = generate_ollama_batch_embeddings(
-                **{
-                    "model": model,
-                    "texts": text,
-                    "url": url,
-                    "key": key,
-                    "prefix": prefix,
-                    "user": user,
-                }
-            )
-        else:
-            embeddings = generate_ollama_batch_embeddings(
-                **{
-                    "model": model,
-                    "texts": [text],
-                    "url": url,
-                    "key": key,
-                    "prefix": prefix,
-                    "user": user,
-                }
-            )
+        embeddings = generate_ollama_batch_embeddings(
+            **{
+                "model": model,
+                "texts": text if isinstance(text, list) else [text],
+                "url": url,
+                "key": key,
+                "prefix": prefix,
+                "user": user,
+            }
+        )
         return embeddings[0] if isinstance(text, str) else embeddings
     elif engine == "openai":
-        if isinstance(text, list):
-            embeddings = generate_openai_batch_embeddings(
-                model, text, url, key, prefix, user
-            )
-        else:
-            embeddings = generate_openai_batch_embeddings(
-                model, [text], url, key, prefix, user
-            )
+        embeddings = generate_openai_batch_embeddings(
+            model, text if isinstance(text, list) else [text], url, key, prefix, user
+        )
+        return embeddings[0] if isinstance(text, str) else embeddings
+    elif engine == "azure_openai":
+        azure_api_version = kwargs.get("azure_api_version", "")
+        embeddings = generate_azure_openai_batch_embeddings(
+            model,
+            text if isinstance(text, list) else [text],
+            url,
+            key,
+            azure_api_version,
+            prefix,
+            user,
+        )
         return embeddings[0] if isinstance(text, str) else embeddings
 
 
@@ -822,7 +947,7 @@ class RerankCompressor(BaseDocumentCompressor):
         reranking = self.reranking_function is not None
 
         if reranking:
-            scores = self.reranking_function.predict(
+            scores = self.reranking_function(
                 [(query, doc.page_content) for doc in documents]
             )
         else:

backend/open_webui/retrieval/vector/dbs/chroma.py

@@ -11,6 +11,8 @@ from open_webui.retrieval.vector.main import (
     SearchResult,
     GetResult,
 )
+from open_webui.retrieval.vector.utils import stringify_metadata
+
 from open_webui.config import (
     CHROMA_DATA_PATH,
     CHROMA_HTTP_HOST,
@@ -144,7 +146,7 @@ class ChromaClient(VectorDBBase):
         ids = [item["id"] for item in items]
         documents = [item["text"] for item in items]
         embeddings = [item["vector"] for item in items]
-        metadatas = [item["metadata"] for item in items]
+        metadatas = [stringify_metadata(item["metadata"]) for item in items]
 
         for batch in create_batches(
             api=self.client,
@@ -164,7 +166,7 @@ class ChromaClient(VectorDBBase):
         ids = [item["id"] for item in items]
         documents = [item["text"] for item in items]
         embeddings = [item["vector"] for item in items]
-        metadatas = [item["metadata"] for item in items]
+        metadatas = [stringify_metadata(item["metadata"]) for item in items]
 
         collection.upsert(
             ids=ids, documents=documents, embeddings=embeddings, metadatas=metadatas

backend/open_webui/retrieval/vector/dbs/elasticsearch.py

@@ -2,6 +2,8 @@ from elasticsearch import Elasticsearch, BadRequestError
 from typing import Optional
 import ssl
 from elasticsearch.helpers import bulk, scan
+
+from open_webui.retrieval.vector.utils import stringify_metadata
 from open_webui.retrieval.vector.main import (
     VectorDBBase,
     VectorItem,
@@ -243,7 +245,7 @@ class ElasticsearchClient(VectorDBBase):
                         "collection": collection_name,
                         "vector": item["vector"],
                         "text": item["text"],
-                        "metadata": item["metadata"],
+                        "metadata": stringify_metadata(item["metadata"]),
                     },
                 }
                 for item in batch
@@ -264,7 +266,7 @@ class ElasticsearchClient(VectorDBBase):
                         "collection": collection_name,
                         "vector": item["vector"],
                         "text": item["text"],
-                        "metadata": item["metadata"],
+                        "metadata": stringify_metadata(item["metadata"]),
                     },
                     "doc_as_upsert": True,
                 }

backend/open_webui/retrieval/vector/dbs/milvus.py

@@ -3,6 +3,8 @@ from pymilvus import FieldSchema, DataType
 import json
 import logging
 from typing import Optional
+
+from open_webui.retrieval.vector.utils import stringify_metadata
 from open_webui.retrieval.vector.main import (
     VectorDBBase,
     VectorItem,
@@ -311,7 +313,7 @@ class MilvusClient(VectorDBBase):
                     "id": item["id"],
                     "vector": item["vector"],
                     "data": {"text": item["text"]},
-                    "metadata": item["metadata"],
+                    "metadata": stringify_metadata(item["metadata"]),
                 }
                 for item in items
             ],
@@ -347,7 +349,7 @@ class MilvusClient(VectorDBBase):
                     "id": item["id"],
                     "vector": item["vector"],
                     "data": {"text": item["text"]},
-                    "metadata": item["metadata"],
+                    "metadata": stringify_metadata(item["metadata"]),
                 }
                 for item in items
             ],

backend/open_webui/retrieval/vector/dbs/opensearch.py

@@ -2,6 +2,7 @@ from opensearchpy import OpenSearch
 from opensearchpy.helpers import bulk
 from typing import Optional
 
+from open_webui.retrieval.vector.utils import stringify_metadata
 from open_webui.retrieval.vector.main import (
     VectorDBBase,
     VectorItem,
@@ -157,10 +158,10 @@ class OpenSearchClient(VectorDBBase):
 
         for field, value in filter.items():
             query_body["query"]["bool"]["filter"].append(
-                {"match": {"metadata." + str(field): value}}
+                {"term": {"metadata." + str(field) + ".keyword": value}}
             )
 
-        size = limit if limit else 10
+        size = limit if limit else 10000
 
         try:
             result = self.client.search(
@@ -200,12 +201,13 @@ class OpenSearchClient(VectorDBBase):
                     "_source": {
                         "vector": item["vector"],
                         "text": item["text"],
-                        "metadata": item["metadata"],
+                        "metadata": stringify_metadata(item["metadata"]),
                     },
                 }
                 for item in batch
             ]
             bulk(self.client, actions)
+        self.client.indices.refresh(self._get_index_name(collection_name))
 
     def upsert(self, collection_name: str, items: list[VectorItem]):
         self._create_index_if_not_exists(
@@ -221,13 +223,14 @@ class OpenSearchClient(VectorDBBase):
                     "doc": {
                         "vector": item["vector"],
                         "text": item["text"],
-                        "metadata": item["metadata"],
+                        "metadata": stringify_metadata(item["metadata"]),
                     },
                     "doc_as_upsert": True,
                 }
                 for item in batch
             ]
             bulk(self.client, actions)
+        self.client.indices.refresh(self._get_index_name(collection_name))
 
     def delete(
         self,
@@ -251,11 +254,12 @@ class OpenSearchClient(VectorDBBase):
             }
             for field, value in filter.items():
                 query_body["query"]["bool"]["filter"].append(
-                    {"match": {"metadata." + str(field): value}}
+                    {"term": {"metadata." + str(field) + ".keyword": value}}
                 )
             self.client.delete_by_query(
                 index=self._get_index_name(collection_name), body=query_body
             )
+        self.client.indices.refresh(self._get_index_name(collection_name))
 
     def reset(self):
         indices = self.client.indices.get(index=f"{self.index_prefix}_*")

backend/open_webui/retrieval/vector/dbs/oracle23ai.py

@@ -0,0 +1,943 @@
+"""
+Oracle 23ai Vector Database Client - Fixed Version
+
+# .env
+VECTOR_DB = "oracle23ai"
+
+## DBCS or oracle 23ai free
+ORACLE_DB_USE_WALLET =  false
+ORACLE_DB_USER = "DEMOUSER"
+ORACLE_DB_PASSWORD = "Welcome123456"
+ORACLE_DB_DSN = "localhost:1521/FREEPDB1"
+
+## ADW or ATP
+# ORACLE_DB_USE_WALLET =  true
+# ORACLE_DB_USER = "DEMOUSER"
+# ORACLE_DB_PASSWORD = "Welcome123456"
+# ORACLE_DB_DSN = "medium"
+# ORACLE_DB_DSN = "(description=  (retry_count=3)(retry_delay=3)(address=(protocol=tcps)(port=1522)(host=xx.oraclecloud.com))(connect_data=(service_name=yy.adb.oraclecloud.com))(security=(ssl_server_dn_match=no)))"
+# ORACLE_WALLET_DIR = "/home/opc/adb_wallet"
+# ORACLE_WALLET_PASSWORD = "Welcome1"
+
+ORACLE_VECTOR_LENGTH = 768
+
+ORACLE_DB_POOL_MIN = 2
+ORACLE_DB_POOL_MAX = 10
+ORACLE_DB_POOL_INCREMENT = 1
+"""
+
+from typing import Optional, List, Dict, Any, Union
+from decimal import Decimal
+import logging
+import os
+import threading
+import time
+import json
+import array
+import oracledb
+
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
+
+from open_webui.config import (
+    ORACLE_DB_USE_WALLET,
+    ORACLE_DB_USER,
+    ORACLE_DB_PASSWORD,
+    ORACLE_DB_DSN,
+    ORACLE_WALLET_DIR,
+    ORACLE_WALLET_PASSWORD,
+    ORACLE_VECTOR_LENGTH,
+    ORACLE_DB_POOL_MIN,
+    ORACLE_DB_POOL_MAX,
+    ORACLE_DB_POOL_INCREMENT,
+)
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class Oracle23aiClient(VectorDBBase):
+    """
+    Oracle Vector Database Client for vector similarity search using Oracle Database 23ai.
+
+    This client provides an interface to store, retrieve, and search vector embeddings
+    in an Oracle database. It uses connection pooling for efficient database access
+    and supports vector similarity search operations.
+
+    Attributes:
+        pool: Connection pool for Oracle database connections
+    """
+
+    def __init__(self) -> None:
+        """
+        Initialize the Oracle23aiClient with a connection pool.
+
+        Creates a connection pool with configurable min/max connections, initializes
+        the database schema if needed, and sets up necessary tables and indexes.
+
+        Raises:
+            ValueError: If required configuration parameters are missing
+            Exception: If database initialization fails
+        """
+        self.pool = None
+
+        try:
+            # Create the appropriate connection pool based on DB type
+            if ORACLE_DB_USE_WALLET:
+                self._create_adb_pool()
+            else:  # DBCS
+                self._create_dbcs_pool()
+
+            dsn = ORACLE_DB_DSN
+            log.info(f"Creating Connection Pool [{ORACLE_DB_USER}:**@{dsn}]")
+
+            with self.get_connection() as connection:
+                log.info(f"Connection version: {connection.version}")
+                self._initialize_database(connection)
+
+            log.info("Oracle Vector Search initialization complete.")
+        except Exception as e:
+            log.exception(f"Error during Oracle Vector Search initialization: {e}")
+            raise
+
+    def _create_adb_pool(self) -> None:
+        """
+        Create connection pool for Oracle Autonomous Database.
+
+        Uses wallet-based authentication.
+        """
+        self.pool = oracledb.create_pool(
+            user=ORACLE_DB_USER,
+            password=ORACLE_DB_PASSWORD,
+            dsn=ORACLE_DB_DSN,
+            min=ORACLE_DB_POOL_MIN,
+            max=ORACLE_DB_POOL_MAX,
+            increment=ORACLE_DB_POOL_INCREMENT,
+            config_dir=ORACLE_WALLET_DIR,
+            wallet_location=ORACLE_WALLET_DIR,
+            wallet_password=ORACLE_WALLET_PASSWORD,
+        )
+        log.info("Created ADB connection pool with wallet authentication.")
+
+    def _create_dbcs_pool(self) -> None:
+        """
+        Create connection pool for Oracle Database Cloud Service.
+
+        Uses basic authentication without wallet.
+        """
+        self.pool = oracledb.create_pool(
+            user=ORACLE_DB_USER,
+            password=ORACLE_DB_PASSWORD,
+            dsn=ORACLE_DB_DSN,
+            min=ORACLE_DB_POOL_MIN,
+            max=ORACLE_DB_POOL_MAX,
+            increment=ORACLE_DB_POOL_INCREMENT,
+        )
+        log.info("Created DB connection pool with basic authentication.")
+
+    def get_connection(self):
+        """
+        Acquire a connection from the connection pool with retry logic.
+
+        Returns:
+            connection: A database connection with output type handler configured
+        """
+        max_retries = 3
+        for attempt in range(max_retries):
+            try:
+                connection = self.pool.acquire()
+                connection.outputtypehandler = self._output_type_handler
+                return connection
+            except oracledb.DatabaseError as e:
+                (error_obj,) = e.args
+                log.exception(
+                    f"Connection attempt {attempt + 1} failed: {error_obj.message}"
+                )
+
+                if attempt < max_retries - 1:
+                    wait_time = 2**attempt
+                    log.info(f"Retrying in {wait_time} seconds...")
+                    time.sleep(wait_time)
+                else:
+                    raise
+
+    def start_health_monitor(self, interval_seconds: int = 60):
+        """
+        Start a background thread to periodically check the health of the connection pool.
+
+        Args:
+            interval_seconds (int): Number of seconds between health checks
+        """
+
+        def _monitor():
+            while True:
+                try:
+                    log.info("[HealthCheck] Running periodic DB health check...")
+                    self.ensure_connection()
+                    log.info("[HealthCheck] Connection is healthy.")
+                except Exception as e:
+                    log.exception(f"[HealthCheck] Connection health check failed: {e}")
+                time.sleep(interval_seconds)
+
+        thread = threading.Thread(target=_monitor, daemon=True)
+        thread.start()
+        log.info(f"Started DB health monitor every {interval_seconds} seconds.")
+
+    def _reconnect_pool(self):
+        """
+        Attempt to reinitialize the connection pool if it's been closed or broken.
+        """
+        try:
+            log.info("Attempting to reinitialize the Oracle connection pool...")
+
+            # Close existing pool if it exists
+            if self.pool:
+                try:
+                    self.pool.close()
+                except Exception as close_error:
+                    log.warning(f"Error closing existing pool: {close_error}")
+
+            # Re-create the appropriate connection pool based on DB type
+            if ORACLE_DB_USE_WALLET:
+                self._create_adb_pool()
+            else:  # DBCS
+                self._create_dbcs_pool()
+
+            log.info("Connection pool reinitialized.")
+        except Exception as e:
+            log.exception(f"Failed to reinitialize the connection pool: {e}")
+            raise
+
+    def ensure_connection(self):
+        """
+        Ensure the database connection is alive, reconnecting pool if needed.
+        """
+        try:
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute("SELECT 1 FROM dual")
+        except Exception as e:
+            log.exception(
+                f"Connection check failed: {e}, attempting to reconnect pool..."
+            )
+            self._reconnect_pool()
+
+    def _output_type_handler(self, cursor, metadata):
+        """
+        Handle Oracle vector type conversion.
+
+        Args:
+            cursor: Oracle database cursor
+            metadata: Metadata for the column
+
+        Returns:
+            A variable with appropriate conversion for vector types
+        """
+        if metadata.type_code is oracledb.DB_TYPE_VECTOR:
+            return cursor.var(
+                metadata.type_code, arraysize=cursor.arraysize, outconverter=list
+            )
+
+    def _initialize_database(self, connection) -> None:
+        """
+        Initialize database schema, tables and indexes.
+
+        Creates the document_chunk table and necessary indexes if they don't exist.
+
+        Args:
+            connection: Oracle database connection
+
+        Raises:
+            Exception: If schema initialization fails
+        """
+        with connection.cursor() as cursor:
+            try:
+                log.info("Creating Table document_chunk")
+                cursor.execute(
+                    """
+                    BEGIN
+                        EXECUTE IMMEDIATE '
+                            CREATE TABLE IF NOT EXISTS document_chunk (
+                                id VARCHAR2(255) PRIMARY KEY,
+                                collection_name VARCHAR2(255) NOT NULL,
+                                text CLOB,
+                                vmetadata JSON,
+                                vector vector(*, float32)
+                            )
+                        ';
+                    EXCEPTION
+                        WHEN OTHERS THEN
+                            IF SQLCODE != -955 THEN
+                                RAISE;
+                            END IF;
+                    END;
+                """
+                )
+
+                log.info("Creating Index document_chunk_collection_name_idx")
+                cursor.execute(
+                    """
+                    BEGIN
+                        EXECUTE IMMEDIATE '
+                            CREATE INDEX IF NOT EXISTS document_chunk_collection_name_idx
+                            ON document_chunk (collection_name)
+                        ';
+                    EXCEPTION
+                        WHEN OTHERS THEN
+                            IF SQLCODE != -955 THEN
+                                RAISE;
+                            END IF;
+                    END;
+                """
+                )
+
+                log.info("Creating VECTOR INDEX document_chunk_vector_ivf_idx")
+                cursor.execute(
+                    """
+                    BEGIN
+                        EXECUTE IMMEDIATE '
+                            CREATE VECTOR INDEX IF NOT EXISTS document_chunk_vector_ivf_idx 
+                            ON document_chunk(vector) 
+                            ORGANIZATION NEIGHBOR PARTITIONS
+                            DISTANCE COSINE
+                            WITH TARGET ACCURACY 95
+                            PARAMETERS (TYPE IVF, NEIGHBOR PARTITIONS 100)
+                        ';
+                    EXCEPTION
+                        WHEN OTHERS THEN
+                            IF SQLCODE != -955 THEN
+                                RAISE;
+                            END IF;
+                    END;
+                """
+                )
+
+                connection.commit()
+                log.info("Database initialization completed successfully.")
+
+            except Exception as e:
+                connection.rollback()
+                log.exception(f"Error during database initialization: {e}")
+                raise
+
+    def check_vector_length(self) -> None:
+        """
+        Check vector length compatibility (placeholder).
+
+        This method would check if the configured vector length matches the database schema.
+        Currently implemented as a placeholder.
+        """
+        pass
+
+    def _vector_to_blob(self, vector: List[float]) -> bytes:
+        """
+        Convert a vector to Oracle BLOB format.
+
+        Args:
+            vector (List[float]): The vector to convert
+
+        Returns:
+            bytes: The vector in Oracle BLOB format
+        """
+        return array.array("f", vector)
+
+    def adjust_vector_length(self, vector: List[float]) -> List[float]:
+        """
+        Adjust vector to the expected length if needed.
+
+        Args:
+            vector (List[float]): The vector to adjust
+
+        Returns:
+            List[float]: The adjusted vector
+        """
+        return vector
+
+    def _decimal_handler(self, obj):
+        """
+        Handle Decimal objects for JSON serialization.
+
+        Args:
+            obj: Object to serialize
+
+        Returns:
+            float: Converted decimal value
+
+        Raises:
+            TypeError: If object is not JSON serializable
+        """
+        if isinstance(obj, Decimal):
+            return float(obj)
+        raise TypeError(f"{obj} is not JSON serializable")
+
+    def _metadata_to_json(self, metadata: Dict) -> str:
+        """
+        Convert metadata dictionary to JSON string.
+
+        Args:
+            metadata (Dict): Metadata dictionary
+
+        Returns:
+            str: JSON representation of metadata
+        """
+        return json.dumps(metadata, default=self._decimal_handler) if metadata else "{}"
+
+    def _json_to_metadata(self, json_str: str) -> Dict:
+        """
+        Convert JSON string to metadata dictionary.
+
+        Args:
+            json_str (str): JSON string
+
+        Returns:
+            Dict: Metadata dictionary
+        """
+        return json.loads(json_str) if json_str else {}
+
+    def insert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """
+        Insert vector items into the database.
+
+        Args:
+            collection_name (str): Name of the collection
+            items (List[VectorItem]): List of vector items to insert
+
+        Raises:
+            Exception: If insertion fails
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> items = [
+            ...     {"id": "1", "text": "Sample text", "vector": [0.1, 0.2, ...], "metadata": {"source": "doc1"}},
+            ...     {"id": "2", "text": "Another text", "vector": [0.3, 0.4, ...], "metadata": {"source": "doc2"}}
+            ... ]
+            >>> client.insert("my_collection", items)
+        """
+        log.info(f"Inserting {len(items)} items into collection '{collection_name}'.")
+
+        with self.get_connection() as connection:
+            try:
+                with connection.cursor() as cursor:
+                    for item in items:
+                        vector_blob = self._vector_to_blob(item["vector"])
+                        metadata_json = self._metadata_to_json(item["metadata"])
+
+                        cursor.execute(
+                            """
+                            INSERT INTO document_chunk 
+                            (id, collection_name, text, vmetadata, vector) 
+                            VALUES (:id, :collection_name, :text, :metadata, :vector)
+                        """,
+                            {
+                                "id": item["id"],
+                                "collection_name": collection_name,
+                                "text": item["text"],
+                                "metadata": metadata_json,
+                                "vector": vector_blob,
+                            },
+                        )
+
+                connection.commit()
+                log.info(
+                    f"Successfully inserted {len(items)} items into collection '{collection_name}'."
+                )
+
+            except Exception as e:
+                connection.rollback()
+                log.exception(f"Error during insert: {e}")
+                raise
+
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """
+        Update or insert vector items into the database.
+
+        If an item with the same ID exists, it will be updated;
+        otherwise, it will be inserted.
+
+        Args:
+            collection_name (str): Name of the collection
+            items (List[VectorItem]): List of vector items to upsert
+
+        Raises:
+            Exception: If upsert operation fails
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> items = [
+            ...     {"id": "1", "text": "Updated text", "vector": [0.1, 0.2, ...], "metadata": {"source": "doc1"}},
+            ...     {"id": "3", "text": "New item", "vector": [0.5, 0.6, ...], "metadata": {"source": "doc3"}}
+            ... ]
+            >>> client.upsert("my_collection", items)
+        """
+        log.info(f"Upserting {len(items)} items into collection '{collection_name}'.")
+
+        with self.get_connection() as connection:
+            try:
+                with connection.cursor() as cursor:
+                    for item in items:
+                        vector_blob = self._vector_to_blob(item["vector"])
+                        metadata_json = self._metadata_to_json(item["metadata"])
+
+                        cursor.execute(
+                            """
+                            MERGE INTO document_chunk d
+                            USING (SELECT :merge_id as id FROM dual) s
+                            ON (d.id = s.id)
+                            WHEN MATCHED THEN
+                                UPDATE SET 
+                                    collection_name = :upd_collection_name,
+                                    text = :upd_text,
+                                    vmetadata = :upd_metadata,
+                                    vector = :upd_vector
+                            WHEN NOT MATCHED THEN
+                                INSERT (id, collection_name, text, vmetadata, vector)
+                                VALUES (:ins_id, :ins_collection_name, :ins_text, :ins_metadata, :ins_vector)
+                        """,
+                            {
+                                "merge_id": item["id"],
+                                "upd_collection_name": collection_name,
+                                "upd_text": item["text"],
+                                "upd_metadata": metadata_json,
+                                "upd_vector": vector_blob,
+                                "ins_id": item["id"],
+                                "ins_collection_name": collection_name,
+                                "ins_text": item["text"],
+                                "ins_metadata": metadata_json,
+                                "ins_vector": vector_blob,
+                            },
+                        )
+
+                connection.commit()
+                log.info(
+                    f"Successfully upserted {len(items)} items into collection '{collection_name}'."
+                )
+
+            except Exception as e:
+                connection.rollback()
+                log.exception(f"Error during upsert: {e}")
+                raise
+
+    def search(
+        self, collection_name: str, vectors: List[List[Union[float, int]]], limit: int
+    ) -> Optional[SearchResult]:
+        """
+        Search for similar vectors in the database.
+
+        Performs vector similarity search using cosine distance.
+
+        Args:
+            collection_name (str): Name of the collection to search
+            vectors (List[List[Union[float, int]]]): Query vectors to find similar items for
+            limit (int): Maximum number of results to return per query
+
+        Returns:
+            Optional[SearchResult]: Search results containing ids, distances, documents, and metadata
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> query_vector = [0.1, 0.2, 0.3, ...]  # Must match VECTOR_LENGTH
+            >>> results = client.search("my_collection", [query_vector], limit=5)
+            >>> if results:
+            ...     log.info(f"Found {len(results.ids[0])} matches")
+            ...     for i, (id, dist) in enumerate(zip(results.ids[0], results.distances[0])):
+            ...         log.info(f"Match {i+1}: id={id}, distance={dist}")
+        """
+        log.info(
+            f"Searching items from collection '{collection_name}' with limit {limit}."
+        )
+
+        try:
+            if not vectors:
+                log.warning("No vectors provided for search.")
+                return None
+
+            num_queries = len(vectors)
+
+            ids = [[] for _ in range(num_queries)]
+            distances = [[] for _ in range(num_queries)]
+            documents = [[] for _ in range(num_queries)]
+            metadatas = [[] for _ in range(num_queries)]
+
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    for qid, vector in enumerate(vectors):
+                        vector_blob = self._vector_to_blob(vector)
+
+                        cursor.execute(
+                            """
+                            SELECT dc.id, dc.text, 
+                                JSON_SERIALIZE(dc.vmetadata RETURNING VARCHAR2(4096)) as vmetadata,
+                                VECTOR_DISTANCE(dc.vector, :query_vector, COSINE) as distance
+                            FROM document_chunk dc
+                            WHERE dc.collection_name = :collection_name
+                            ORDER BY VECTOR_DISTANCE(dc.vector, :query_vector, COSINE)
+                            FETCH APPROX FIRST :limit ROWS ONLY
+                        """,
+                            {
+                                "query_vector": vector_blob,
+                                "collection_name": collection_name,
+                                "limit": limit,
+                            },
+                        )
+
+                        results = cursor.fetchall()
+
+                        for row in results:
+                            ids[qid].append(row[0])
+                            documents[qid].append(
+                                row[1].read()
+                                if isinstance(row[1], oracledb.LOB)
+                                else str(row[1])
+                            )
+                            # 🔧 FIXED: Parse JSON metadata properly
+                            metadata_str = (
+                                row[2].read()
+                                if isinstance(row[2], oracledb.LOB)
+                                else row[2]
+                            )
+                            metadatas[qid].append(self._json_to_metadata(metadata_str))
+                            distances[qid].append(float(row[3]))
+
+            log.info(
+                f"Search completed. Found {sum(len(ids[i]) for i in range(num_queries))} total results."
+            )
+
+            return SearchResult(
+                ids=ids, distances=distances, documents=documents, metadatas=metadatas
+            )
+
+        except Exception as e:
+            log.exception(f"Error during search: {e}")
+            return None
+
+    def query(
+        self, collection_name: str, filter: Dict, limit: Optional[int] = None
+    ) -> Optional[GetResult]:
+        """
+        Query items based on metadata filters.
+
+        Retrieves items that match specified metadata criteria.
+
+        Args:
+            collection_name (str): Name of the collection to query
+            filter (Dict[str, Any]): Metadata filters to apply
+            limit (Optional[int]): Maximum number of results to return
+
+        Returns:
+            Optional[GetResult]: Query results containing ids, documents, and metadata
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> filter = {"source": "doc1", "category": "finance"}
+            >>> results = client.query("my_collection", filter, limit=20)
+            >>> if results:
+            ...     print(f"Found {len(results.ids[0])} matching documents")
+        """
+        log.info(f"Querying items from collection '{collection_name}' with filters.")
+
+        try:
+            limit = limit or 100
+
+            query = """
+                SELECT id, text, JSON_SERIALIZE(vmetadata RETURNING VARCHAR2(4096)) as vmetadata 
+                FROM document_chunk
+                WHERE collection_name = :collection_name
+            """
+
+            params = {"collection_name": collection_name}
+
+            for i, (key, value) in enumerate(filter.items()):
+                param_name = f"value_{i}"
+                query += f" AND JSON_VALUE(vmetadata, '$.{key}' RETURNING VARCHAR2(4096)) = :{param_name}"
+                params[param_name] = str(value)
+
+            query += " FETCH FIRST :limit ROWS ONLY"
+            params["limit"] = limit
+
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute(query, params)
+                    results = cursor.fetchall()
+
+            if not results:
+                log.info("No results found for query.")
+                return None
+
+            ids = [[row[0] for row in results]]
+            documents = [
+                [
+                    row[1].read() if isinstance(row[1], oracledb.LOB) else str(row[1])
+                    for row in results
+                ]
+            ]
+            # 🔧 FIXED: Parse JSON metadata properly
+            metadatas = [
+                [
+                    self._json_to_metadata(
+                        row[2].read() if isinstance(row[2], oracledb.LOB) else row[2]
+                    )
+                    for row in results
+                ]
+            ]
+
+            log.info(f"Query completed. Found {len(results)} results.")
+
+            return GetResult(ids=ids, documents=documents, metadatas=metadatas)
+
+        except Exception as e:
+            log.exception(f"Error during query: {e}")
+            return None
+
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        """
+        Get all items in a collection.
+
+        Retrieves items from a specified collection up to the limit.
+
+        Args:
+            collection_name (str): Name of the collection to retrieve
+            limit (Optional[int]): Maximum number of items to retrieve
+
+        Returns:
+            Optional[GetResult]: Result containing ids, documents, and metadata
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> results = client.get("my_collection", limit=50)
+            >>> if results:
+            ...     print(f"Retrieved {len(results.ids[0])} documents from collection")
+        """
+        log.info(
+            f"Getting items from collection '{collection_name}' with limit {limit}."
+        )
+
+        try:
+            limit = limit or 1000
+
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute(
+                        """
+                        SELECT /*+ MONITOR */ id, text, JSON_SERIALIZE(vmetadata RETURNING VARCHAR2(4096)) as vmetadata
+                        FROM document_chunk
+                        WHERE collection_name = :collection_name
+                        FETCH FIRST :limit ROWS ONLY
+                    """,
+                        {"collection_name": collection_name, "limit": limit},
+                    )
+
+                    results = cursor.fetchall()
+
+            if not results:
+                log.info("No results found.")
+                return None
+
+            ids = [[row[0] for row in results]]
+            documents = [
+                [
+                    row[1].read() if isinstance(row[1], oracledb.LOB) else str(row[1])
+                    for row in results
+                ]
+            ]
+            # 🔧 FIXED: Parse JSON metadata properly
+            metadatas = [
+                [
+                    self._json_to_metadata(
+                        row[2].read() if isinstance(row[2], oracledb.LOB) else row[2]
+                    )
+                    for row in results
+                ]
+            ]
+
+            return GetResult(ids=ids, documents=documents, metadatas=metadatas)
+
+        except Exception as e:
+            log.exception(f"Error during get: {e}")
+            return None
+
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[List[str]] = None,
+        filter: Optional[Dict[str, Any]] = None,
+    ) -> None:
+        """
+        Delete items from the database.
+
+        Deletes items from a collection based on IDs or metadata filters.
+
+        Args:
+            collection_name (str): Name of the collection to delete from
+            ids (Optional[List[str]]): Specific item IDs to delete
+            filter (Optional[Dict[str, Any]]): Metadata filters for deletion
+
+        Raises:
+            Exception: If deletion fails
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> # Delete specific items by ID
+            >>> client.delete("my_collection", ids=["1", "3", "5"])
+            >>> # Or delete by metadata filter
+            >>> client.delete("my_collection", filter={"source": "deprecated_source"})
+        """
+        log.info(f"Deleting items from collection '{collection_name}'.")
+
+        try:
+            query = (
+                "DELETE FROM document_chunk WHERE collection_name = :collection_name"
+            )
+            params = {"collection_name": collection_name}
+
+            if ids:
+                # 🔧 FIXED: Use proper parameterized query to prevent SQL injection
+                placeholders = ",".join([f":id_{i}" for i in range(len(ids))])
+                query += f" AND id IN ({placeholders})"
+                for i, id_val in enumerate(ids):
+                    params[f"id_{i}"] = id_val
+
+            if filter:
+                for i, (key, value) in enumerate(filter.items()):
+                    param_name = f"value_{i}"
+                    query += f" AND JSON_VALUE(vmetadata, '$.{key}' RETURNING VARCHAR2(4096)) = :{param_name}"
+                    params[param_name] = str(value)
+
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute(query, params)
+                    deleted = cursor.rowcount
+                connection.commit()
+
+            log.info(f"Deleted {deleted} items from collection '{collection_name}'.")
+
+        except Exception as e:
+            log.exception(f"Error during delete: {e}")
+            raise
+
+    def reset(self) -> None:
+        """
+        Reset the database by deleting all items.
+
+        Deletes all items from the document_chunk table.
+
+        Raises:
+            Exception: If reset fails
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> client.reset()  # Warning: Removes all data!
+        """
+        log.info("Resetting database - deleting all items.")
+
+        try:
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute("DELETE FROM document_chunk")
+                    deleted = cursor.rowcount
+                connection.commit()
+
+            log.info(
+                f"Reset complete. Deleted {deleted} items from 'document_chunk' table."
+            )
+
+        except Exception as e:
+            log.exception(f"Error during reset: {e}")
+            raise
+
+    def close(self) -> None:
+        """
+        Close the database connection pool.
+
+        Properly closes the connection pool and releases all resources.
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> # After finishing all operations
+            >>> client.close()
+        """
+        try:
+            if hasattr(self, "pool") and self.pool:
+                self.pool.close()
+                log.info("Oracle Vector Search connection pool closed.")
+        except Exception as e:
+            log.exception(f"Error closing connection pool: {e}")
+
+    def has_collection(self, collection_name: str) -> bool:
+        """
+        Check if a collection exists.
+
+        Args:
+            collection_name (str): Name of the collection to check
+
+        Returns:
+            bool: True if the collection exists, False otherwise
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> if client.has_collection("my_collection"):
+            ...     print("Collection exists!")
+            ... else:
+            ...     print("Collection does not exist.")
+        """
+        try:
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute(
+                        """
+                        SELECT COUNT(*)
+                        FROM document_chunk
+                        WHERE collection_name = :collection_name
+                        FETCH FIRST 1 ROWS ONLY
+                    """,
+                        {"collection_name": collection_name},
+                    )
+
+                    count = cursor.fetchone()[0]
+
+            return count > 0
+
+        except Exception as e:
+            log.exception(f"Error checking collection existence: {e}")
+            return False
+
+    def delete_collection(self, collection_name: str) -> None:
+        """
+        Delete an entire collection.
+
+        Removes all items belonging to the specified collection.
+
+        Args:
+            collection_name (str): Name of the collection to delete
+
+        Example:
+            >>> client = Oracle23aiClient()
+            >>> client.delete_collection("obsolete_collection")
+        """
+        log.info(f"Deleting collection '{collection_name}'.")
+
+        try:
+            with self.get_connection() as connection:
+                with connection.cursor() as cursor:
+                    cursor.execute(
+                        """
+                        DELETE FROM document_chunk 
+                        WHERE collection_name = :collection_name
+                    """,
+                        {"collection_name": collection_name},
+                    )
+
+                    deleted = cursor.rowcount
+                connection.commit()
+
+            log.info(
+                f"Collection '{collection_name}' deleted. Removed {deleted} items."
+            )
+
+        except Exception as e:
+            log.exception(f"Error deleting collection '{collection_name}': {e}")
+            raise

backend/open_webui/retrieval/vector/dbs/pgvector.py

@@ -1,12 +1,16 @@
 from typing import Optional, List, Dict, Any
 import logging
+import json
 from sqlalchemy import (
+    func,
+    literal,
     cast,
     column,
     create_engine,
     Column,
     Integer,
     MetaData,
+    LargeBinary,
     select,
     text,
     Text,
@@ -14,7 +18,7 @@ from sqlalchemy import (
     values,
 )
 from sqlalchemy.sql import true
-from sqlalchemy.pool import NullPool
+from sqlalchemy.pool import NullPool, QueuePool
 
 from sqlalchemy.orm import declarative_base, scoped_session, sessionmaker
 from sqlalchemy.dialects.postgresql import JSONB, array
@@ -22,13 +26,24 @@ from pgvector.sqlalchemy import Vector
 from sqlalchemy.ext.mutable import MutableDict
 from sqlalchemy.exc import NoSuchTableError
 
+
+from open_webui.retrieval.vector.utils import stringify_metadata
 from open_webui.retrieval.vector.main import (
     VectorDBBase,
     VectorItem,
     SearchResult,
     GetResult,
 )
-from open_webui.config import PGVECTOR_DB_URL, PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH
+from open_webui.config import (
+    PGVECTOR_DB_URL,
+    PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH,
+    PGVECTOR_PGCRYPTO,
+    PGVECTOR_PGCRYPTO_KEY,
+    PGVECTOR_POOL_SIZE,
+    PGVECTOR_POOL_MAX_OVERFLOW,
+    PGVECTOR_POOL_TIMEOUT,
+    PGVECTOR_POOL_RECYCLE,
+)
 
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -39,14 +54,27 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
+def pgcrypto_encrypt(val, key):
+    return func.pgp_sym_encrypt(val, literal(key))
+
+
+def pgcrypto_decrypt(col, key, outtype="text"):
+    return func.cast(func.pgp_sym_decrypt(col, literal(key)), outtype)
+
+
 class DocumentChunk(Base):
     __tablename__ = "document_chunk"
 
     id = Column(Text, primary_key=True)
     vector = Column(Vector(dim=VECTOR_LENGTH), nullable=True)
     collection_name = Column(Text, nullable=False)
-    text = Column(Text, nullable=True)
-    vmetadata = Column(MutableDict.as_mutable(JSONB), nullable=True)
+
+    if PGVECTOR_PGCRYPTO:
+        text = Column(LargeBinary, nullable=True)
+        vmetadata = Column(LargeBinary, nullable=True)
+    else:
+        text = Column(Text, nullable=True)
+        vmetadata = Column(MutableDict.as_mutable(JSONB), nullable=True)
 
 
 class PgvectorClient(VectorDBBase):
@@ -58,9 +86,24 @@ class PgvectorClient(VectorDBBase):
 
             self.session = Session
         else:
-            engine = create_engine(
-                PGVECTOR_DB_URL, pool_pre_ping=True, poolclass=NullPool
-            )
+            if isinstance(PGVECTOR_POOL_SIZE, int):
+                if PGVECTOR_POOL_SIZE > 0:
+                    engine = create_engine(
+                        PGVECTOR_DB_URL,
+                        pool_size=PGVECTOR_POOL_SIZE,
+                        max_overflow=PGVECTOR_POOL_MAX_OVERFLOW,
+                        pool_timeout=PGVECTOR_POOL_TIMEOUT,
+                        pool_recycle=PGVECTOR_POOL_RECYCLE,
+                        pool_pre_ping=True,
+                        poolclass=QueuePool,
+                    )
+                else:
+                    engine = create_engine(
+                        PGVECTOR_DB_URL, pool_pre_ping=True, poolclass=NullPool
+                    )
+            else:
+                engine = create_engine(PGVECTOR_DB_URL, pool_pre_ping=True)
+
             SessionLocal = sessionmaker(
                 autocommit=False, autoflush=False, bind=engine, expire_on_commit=False
             )
@@ -70,6 +113,15 @@ class PgvectorClient(VectorDBBase):
             # Ensure the pgvector extension is available
             self.session.execute(text("CREATE EXTENSION IF NOT EXISTS vector;"))
 
+            if PGVECTOR_PGCRYPTO:
+                # Ensure the pgcrypto extension is available for encryption
+                self.session.execute(text("CREATE EXTENSION IF NOT EXISTS pgcrypto;"))
+
+                if not PGVECTOR_PGCRYPTO_KEY:
+                    raise ValueError(
+                        "PGVECTOR_PGCRYPTO_KEY must be set when PGVECTOR_PGCRYPTO is enabled."
+                    )
+
             # Check vector length consistency
             self.check_vector_length()
 
@@ -147,22 +199,54 @@ class PgvectorClient(VectorDBBase):
 
     def insert(self, collection_name: str, items: List[VectorItem]) -> None:
         try:
-            new_items = []
-            for item in items:
-                vector = self.adjust_vector_length(item["vector"])
-                new_chunk = DocumentChunk(
-                    id=item["id"],
-                    vector=vector,
-                    collection_name=collection_name,
-                    text=item["text"],
-                    vmetadata=item["metadata"],
+            if PGVECTOR_PGCRYPTO:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    # Use raw SQL for BYTEA/pgcrypto
+                    # Ensure metadata is converted to its JSON text representation
+                    json_metadata = json.dumps(item["metadata"])
+                    self.session.execute(
+                        text(
+                            """
+                            INSERT INTO document_chunk
+                            (id, vector, collection_name, text, vmetadata)
+                            VALUES (
+                                :id, :vector, :collection_name,
+                                pgp_sym_encrypt(:text, :key),
+                                pgp_sym_encrypt(:metadata_text, :key)
+                            )
+                            ON CONFLICT (id) DO NOTHING
+                        """
+                        ),
+                        {
+                            "id": item["id"],
+                            "vector": vector,
+                            "collection_name": collection_name,
+                            "text": item["text"],
+                            "metadata_text": json_metadata,
+                            "key": PGVECTOR_PGCRYPTO_KEY,
+                        },
+                    )
+                self.session.commit()
+                log.info(f"Encrypted & inserted {len(items)} into '{collection_name}'")
+
+            else:
+                new_items = []
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    new_chunk = DocumentChunk(
+                        id=item["id"],
+                        vector=vector,
+                        collection_name=collection_name,
+                        text=item["text"],
+                        vmetadata=stringify_metadata(item["metadata"]),
+                    )
+                    new_items.append(new_chunk)
+                self.session.bulk_save_objects(new_items)
+                self.session.commit()
+                log.info(
+                    f"Inserted {len(new_items)} items into collection '{collection_name}'."
                 )
-                new_items.append(new_chunk)
-            self.session.bulk_save_objects(new_items)
-            self.session.commit()
-            log.info(
-                f"Inserted {len(new_items)} items into collection '{collection_name}'."
-            )
         except Exception as e:
             self.session.rollback()
             log.exception(f"Error during insert: {e}")
@@ -170,33 +254,66 @@ class PgvectorClient(VectorDBBase):
 
     def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
         try:
-            for item in items:
-                vector = self.adjust_vector_length(item["vector"])
-                existing = (
-                    self.session.query(DocumentChunk)
-                    .filter(DocumentChunk.id == item["id"])
-                    .first()
+            if PGVECTOR_PGCRYPTO:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    json_metadata = json.dumps(item["metadata"])
+                    self.session.execute(
+                        text(
+                            """
+                            INSERT INTO document_chunk
+                            (id, vector, collection_name, text, vmetadata)
+                            VALUES (
+                                :id, :vector, :collection_name,
+                                pgp_sym_encrypt(:text, :key),
+                                pgp_sym_encrypt(:metadata_text, :key)
+                            )
+                            ON CONFLICT (id) DO UPDATE SET
+                              vector = EXCLUDED.vector,
+                              collection_name = EXCLUDED.collection_name,
+                              text = EXCLUDED.text,
+                              vmetadata = EXCLUDED.vmetadata
+                        """
+                        ),
+                        {
+                            "id": item["id"],
+                            "vector": vector,
+                            "collection_name": collection_name,
+                            "text": item["text"],
+                            "metadata_text": json_metadata,
+                            "key": PGVECTOR_PGCRYPTO_KEY,
+                        },
+                    )
+                self.session.commit()
+                log.info(f"Encrypted & upserted {len(items)} into '{collection_name}'")
+            else:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    existing = (
+                        self.session.query(DocumentChunk)
+                        .filter(DocumentChunk.id == item["id"])
+                        .first()
+                    )
+                    if existing:
+                        existing.vector = vector
+                        existing.text = item["text"]
+                        existing.vmetadata = stringify_metadata(item["metadata"])
+                        existing.collection_name = (
+                            collection_name  # Update collection_name if necessary
+                        )
+                    else:
+                        new_chunk = DocumentChunk(
+                            id=item["id"],
+                            vector=vector,
+                            collection_name=collection_name,
+                            text=item["text"],
+                            vmetadata=stringify_metadata(item["metadata"]),
+                        )
+                        self.session.add(new_chunk)
+                self.session.commit()
+                log.info(
+                    f"Upserted {len(items)} items into collection '{collection_name}'."
                 )
-                if existing:
-                    existing.vector = vector
-                    existing.text = item["text"]
-                    existing.vmetadata = item["metadata"]
-                    existing.collection_name = (
-                        collection_name  # Update collection_name if necessary
-                    )
-                else:
-                    new_chunk = DocumentChunk(
-                        id=item["id"],
-                        vector=vector,
-                        collection_name=collection_name,
-                        text=item["text"],
-                        vmetadata=item["metadata"],
-                    )
-                    self.session.add(new_chunk)
-            self.session.commit()
-            log.info(
-                f"Upserted {len(items)} items into collection '{collection_name}'."
-            )
         except Exception as e:
             self.session.rollback()
             log.exception(f"Error during upsert: {e}")
@@ -230,16 +347,32 @@ class PgvectorClient(VectorDBBase):
                 .alias("query_vectors")
             )
 
+            result_fields = [
+                DocumentChunk.id,
+            ]
+            if PGVECTOR_PGCRYPTO:
+                result_fields.append(
+                    pgcrypto_decrypt(
+                        DocumentChunk.text, PGVECTOR_PGCRYPTO_KEY, Text
+                    ).label("text")
+                )
+                result_fields.append(
+                    pgcrypto_decrypt(
+                        DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                    ).label("vmetadata")
+                )
+            else:
+                result_fields.append(DocumentChunk.text)
+                result_fields.append(DocumentChunk.vmetadata)
+            result_fields.append(
+                (DocumentChunk.vector.cosine_distance(query_vectors.c.q_vector)).label(
+                    "distance"
+                )
+            )
+
             # Build the lateral subquery for each query vector
             subq = (
-                select(
-                    DocumentChunk.id,
-                    DocumentChunk.text,
-                    DocumentChunk.vmetadata,
-                    (
-                        DocumentChunk.vector.cosine_distance(query_vectors.c.q_vector)
-                    ).label("distance"),
-                )
+                select(*result_fields)
                 .where(DocumentChunk.collection_name == collection_name)
                 .order_by(
                     (DocumentChunk.vector.cosine_distance(query_vectors.c.q_vector))
@@ -288,10 +421,12 @@ class PgvectorClient(VectorDBBase):
                 documents[qid].append(row.text)
                 metadatas[qid].append(row.vmetadata)
 
+            self.session.rollback()  # read-only transaction
             return SearchResult(
                 ids=ids, distances=distances, documents=documents, metadatas=metadatas
             )
         except Exception as e:
+            self.session.rollback()
             log.exception(f"Error during search: {e}")
             return None
 
@@ -299,17 +434,43 @@ class PgvectorClient(VectorDBBase):
         self, collection_name: str, filter: Dict[str, Any], limit: Optional[int] = None
     ) -> Optional[GetResult]:
         try:
-            query = self.session.query(DocumentChunk).filter(
-                DocumentChunk.collection_name == collection_name
-            )
+            if PGVECTOR_PGCRYPTO:
+                # Build where clause for vmetadata filter
+                where_clauses = [DocumentChunk.collection_name == collection_name]
+                for key, value in filter.items():
+                    # decrypt then check key: JSON filter after decryption
+                    where_clauses.append(
+                        pgcrypto_decrypt(
+                            DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                        )[key].astext
+                        == str(value)
+                    )
+                stmt = select(
+                    DocumentChunk.id,
+                    pgcrypto_decrypt(
+                        DocumentChunk.text, PGVECTOR_PGCRYPTO_KEY, Text
+                    ).label("text"),
+                    pgcrypto_decrypt(
+                        DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                    ).label("vmetadata"),
+                ).where(*where_clauses)
+                if limit is not None:
+                    stmt = stmt.limit(limit)
+                results = self.session.execute(stmt).all()
+            else:
+                query = self.session.query(DocumentChunk).filter(
+                    DocumentChunk.collection_name == collection_name
+                )
 
-            for key, value in filter.items():
-                query = query.filter(DocumentChunk.vmetadata[key].astext == str(value))
+                for key, value in filter.items():
+                    query = query.filter(
+                        DocumentChunk.vmetadata[key].astext == str(value)
+                    )
 
-            if limit is not None:
-                query = query.limit(limit)
+                if limit is not None:
+                    query = query.limit(limit)
 
-            results = query.all()
+                results = query.all()
 
             if not results:
                 return None
@@ -318,12 +479,14 @@ class PgvectorClient(VectorDBBase):
             documents = [[result.text for result in results]]
             metadatas = [[result.vmetadata for result in results]]
 
+            self.session.rollback()  # read-only transaction
             return GetResult(
                 ids=ids,
                 documents=documents,
                 metadatas=metadatas,
             )
         except Exception as e:
+            self.session.rollback()
             log.exception(f"Error during query: {e}")
             return None
 
@@ -331,23 +494,43 @@ class PgvectorClient(VectorDBBase):
         self, collection_name: str, limit: Optional[int] = None
     ) -> Optional[GetResult]:
         try:
-            query = self.session.query(DocumentChunk).filter(
-                DocumentChunk.collection_name == collection_name
-            )
-            if limit is not None:
-                query = query.limit(limit)
+            if PGVECTOR_PGCRYPTO:
+                stmt = select(
+                    DocumentChunk.id,
+                    pgcrypto_decrypt(
+                        DocumentChunk.text, PGVECTOR_PGCRYPTO_KEY, Text
+                    ).label("text"),
+                    pgcrypto_decrypt(
+                        DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                    ).label("vmetadata"),
+                ).where(DocumentChunk.collection_name == collection_name)
+                if limit is not None:
+                    stmt = stmt.limit(limit)
+                results = self.session.execute(stmt).all()
+                ids = [[row.id for row in results]]
+                documents = [[row.text for row in results]]
+                metadatas = [[row.vmetadata for row in results]]
+            else:
 
-            results = query.all()
+                query = self.session.query(DocumentChunk).filter(
+                    DocumentChunk.collection_name == collection_name
+                )
+                if limit is not None:
+                    query = query.limit(limit)
 
-            if not results:
-                return None
+                results = query.all()
 
-            ids = [[result.id for result in results]]
-            documents = [[result.text for result in results]]
-            metadatas = [[result.vmetadata for result in results]]
+                if not results:
+                    return None
 
+                ids = [[result.id for result in results]]
+                documents = [[result.text for result in results]]
+                metadatas = [[result.vmetadata for result in results]]
+
+            self.session.rollback()  # read-only transaction
             return GetResult(ids=ids, documents=documents, metadatas=metadatas)
         except Exception as e:
+            self.session.rollback()
             log.exception(f"Error during get: {e}")
             return None
 
@@ -358,17 +541,33 @@ class PgvectorClient(VectorDBBase):
         filter: Optional[Dict[str, Any]] = None,
     ) -> None:
         try:
-            query = self.session.query(DocumentChunk).filter(
-                DocumentChunk.collection_name == collection_name
-            )
-            if ids:
-                query = query.filter(DocumentChunk.id.in_(ids))
-            if filter:
-                for key, value in filter.items():
-                    query = query.filter(
-                        DocumentChunk.vmetadata[key].astext == str(value)
-                    )
-            deleted = query.delete(synchronize_session=False)
+            if PGVECTOR_PGCRYPTO:
+                wheres = [DocumentChunk.collection_name == collection_name]
+                if ids:
+                    wheres.append(DocumentChunk.id.in_(ids))
+                if filter:
+                    for key, value in filter.items():
+                        wheres.append(
+                            pgcrypto_decrypt(
+                                DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                            )[key].astext
+                            == str(value)
+                        )
+                stmt = DocumentChunk.__table__.delete().where(*wheres)
+                result = self.session.execute(stmt)
+                deleted = result.rowcount
+            else:
+                query = self.session.query(DocumentChunk).filter(
+                    DocumentChunk.collection_name == collection_name
+                )
+                if ids:
+                    query = query.filter(DocumentChunk.id.in_(ids))
+                if filter:
+                    for key, value in filter.items():
+                        query = query.filter(
+                            DocumentChunk.vmetadata[key].astext == str(value)
+                        )
+                deleted = query.delete(synchronize_session=False)
             self.session.commit()
             log.info(f"Deleted {deleted} items from collection '{collection_name}'.")
         except Exception as e:
@@ -399,8 +598,10 @@ class PgvectorClient(VectorDBBase):
                 .first()
                 is not None
             )
+            self.session.rollback()  # read-only transaction
             return exists
         except Exception as e:
+            self.session.rollback()
             log.exception(f"Error checking collection existence: {e}")
             return False
 

backend/open_webui/retrieval/vector/dbs/pinecone.py

@@ -3,10 +3,19 @@ import logging
 import time  # for measuring elapsed time
 from pinecone import Pinecone, ServerlessSpec
 
+# Add gRPC support for better performance (Pinecone best practice)
+try:
+    from pinecone.grpc import PineconeGRPC
+
+    GRPC_AVAILABLE = True
+except ImportError:
+    GRPC_AVAILABLE = False
+
 import asyncio  # for async upserts
 import functools  # for partial binding in async tasks
 
 import concurrent.futures  # for parallel batch upserts
+import random  # for jitter in retry backoff
 
 from open_webui.retrieval.vector.main import (
     VectorDBBase,
@@ -47,7 +56,24 @@ class PineconeClient(VectorDBBase):
         self.cloud = PINECONE_CLOUD
 
         # Initialize Pinecone client for improved performance
-        self.client = Pinecone(api_key=self.api_key)
+        if GRPC_AVAILABLE:
+            # Use gRPC client for better performance (Pinecone recommendation)
+            self.client = PineconeGRPC(
+                api_key=self.api_key,
+                pool_threads=20,  # Improved connection pool size
+                timeout=30,  # Reasonable timeout for operations
+            )
+            self.using_grpc = True
+            log.info("Using Pinecone gRPC client for optimal performance")
+        else:
+            # Fallback to HTTP client with enhanced connection pooling
+            self.client = Pinecone(
+                api_key=self.api_key,
+                pool_threads=20,  # Improved connection pool size
+                timeout=30,  # Reasonable timeout for operations
+            )
+            self.using_grpc = False
+            log.info("Using Pinecone HTTP client (gRPC not available)")
 
         # Persistent executor for batch operations
         self._executor = concurrent.futures.ThreadPoolExecutor(max_workers=5)
@@ -91,12 +117,53 @@ class PineconeClient(VectorDBBase):
                 log.info(f"Using existing Pinecone index '{self.index_name}'")
 
             # Connect to the index
-            self.index = self.client.Index(self.index_name)
+            self.index = self.client.Index(
+                self.index_name,
+                pool_threads=20,  # Enhanced connection pool for index operations
+            )
 
         except Exception as e:
             log.error(f"Failed to initialize Pinecone index: {e}")
             raise RuntimeError(f"Failed to initialize Pinecone index: {e}")
 
+    def _retry_pinecone_operation(self, operation_func, max_retries=3):
+        """Retry Pinecone operations with exponential backoff for rate limits and network issues."""
+        for attempt in range(max_retries):
+            try:
+                return operation_func()
+            except Exception as e:
+                error_str = str(e).lower()
+                # Check if it's a retryable error (rate limits, network issues, timeouts)
+                is_retryable = any(
+                    keyword in error_str
+                    for keyword in [
+                        "rate limit",
+                        "quota",
+                        "timeout",
+                        "network",
+                        "connection",
+                        "unavailable",
+                        "internal error",
+                        "429",
+                        "500",
+                        "502",
+                        "503",
+                        "504",
+                    ]
+                )
+
+                if not is_retryable or attempt == max_retries - 1:
+                    # Don't retry for non-retryable errors or on final attempt
+                    raise
+
+                # Exponential backoff with jitter
+                delay = (2**attempt) + random.uniform(0, 1)
+                log.warning(
+                    f"Pinecone operation failed (attempt {attempt + 1}/{max_retries}), "
+                    f"retrying in {delay:.2f}s: {e}"
+                )
+                time.sleep(delay)
+
     def _create_points(
         self, items: List[VectorItem], collection_name_with_prefix: str
     ) -> List[Dict[str, Any]]:
@@ -223,7 +290,8 @@ class PineconeClient(VectorDBBase):
         elapsed = time.time() - start_time
         log.debug(f"Insert of {len(points)} vectors took {elapsed:.2f} seconds")
         log.info(
-            f"Successfully inserted {len(points)} vectors in parallel batches into '{collection_name_with_prefix}'"
+            f"Successfully inserted {len(points)} vectors in parallel batches "
+            f"into '{collection_name_with_prefix}'"
         )
 
     def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
@@ -254,7 +322,8 @@ class PineconeClient(VectorDBBase):
         elapsed = time.time() - start_time
         log.debug(f"Upsert of {len(points)} vectors took {elapsed:.2f} seconds")
         log.info(
-            f"Successfully upserted {len(points)} vectors in parallel batches into '{collection_name_with_prefix}'"
+            f"Successfully upserted {len(points)} vectors in parallel batches "
+            f"into '{collection_name_with_prefix}'"
         )
 
     async def insert_async(self, collection_name: str, items: List[VectorItem]) -> None:
@@ -285,7 +354,8 @@ class PineconeClient(VectorDBBase):
                 log.error(f"Error in async insert batch: {result}")
                 raise result
         log.info(
-            f"Successfully async inserted {len(points)} vectors in batches into '{collection_name_with_prefix}'"
+            f"Successfully async inserted {len(points)} vectors in batches "
+            f"into '{collection_name_with_prefix}'"
         )
 
     async def upsert_async(self, collection_name: str, items: List[VectorItem]) -> None:
@@ -316,7 +386,8 @@ class PineconeClient(VectorDBBase):
                 log.error(f"Error in async upsert batch: {result}")
                 raise result
         log.info(
-            f"Successfully async upserted {len(points)} vectors in batches into '{collection_name_with_prefix}'"
+            f"Successfully async upserted {len(points)} vectors in batches "
+            f"into '{collection_name_with_prefix}'"
         )
 
     def search(
@@ -457,10 +528,12 @@ class PineconeClient(VectorDBBase):
                     # This is a limitation of Pinecone - be careful with ID uniqueness
                     self.index.delete(ids=batch_ids)
                     log.debug(
-                        f"Deleted batch of {len(batch_ids)} vectors by ID from '{collection_name_with_prefix}'"
+                        f"Deleted batch of {len(batch_ids)} vectors by ID "
+                        f"from '{collection_name_with_prefix}'"
                     )
                 log.info(
-                    f"Successfully deleted {len(ids)} vectors by ID from '{collection_name_with_prefix}'"
+                    f"Successfully deleted {len(ids)} vectors by ID "
+                    f"from '{collection_name_with_prefix}'"
                 )
 
             elif filter:

backend/open_webui/retrieval/vector/dbs/qdrant.py

@@ -18,6 +18,9 @@ from open_webui.config import (
     QDRANT_ON_DISK,
     QDRANT_GRPC_PORT,
     QDRANT_PREFER_GRPC,
+    QDRANT_COLLECTION_PREFIX,
+    QDRANT_TIMEOUT,
+    QDRANT_HNSW_M,
 )
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -29,12 +32,14 @@ log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 class QdrantClient(VectorDBBase):
     def __init__(self):
-        self.collection_prefix = "open-webui"
+        self.collection_prefix = QDRANT_COLLECTION_PREFIX
         self.QDRANT_URI = QDRANT_URI
         self.QDRANT_API_KEY = QDRANT_API_KEY
         self.QDRANT_ON_DISK = QDRANT_ON_DISK
         self.PREFER_GRPC = QDRANT_PREFER_GRPC
         self.GRPC_PORT = QDRANT_GRPC_PORT
+        self.QDRANT_TIMEOUT = QDRANT_TIMEOUT
+        self.QDRANT_HNSW_M = QDRANT_HNSW_M
 
         if not self.QDRANT_URI:
             self.client = None
@@ -52,9 +57,14 @@ class QdrantClient(VectorDBBase):
                 grpc_port=self.GRPC_PORT,
                 prefer_grpc=self.PREFER_GRPC,
                 api_key=self.QDRANT_API_KEY,
+                timeout=self.QDRANT_TIMEOUT,
             )
         else:
-            self.client = Qclient(url=self.QDRANT_URI, api_key=self.QDRANT_API_KEY)
+            self.client = Qclient(
+                url=self.QDRANT_URI,
+                api_key=self.QDRANT_API_KEY,
+                timeout=QDRANT_TIMEOUT,
+            )
 
     def _result_to_get_result(self, points) -> GetResult:
         ids = []
@@ -84,8 +94,30 @@ class QdrantClient(VectorDBBase):
                 distance=models.Distance.COSINE,
                 on_disk=self.QDRANT_ON_DISK,
             ),
+            hnsw_config=models.HnswConfigDiff(
+                m=self.QDRANT_HNSW_M,
+            ),
         )
 
+        # Create payload indexes for efficient filtering
+        self.client.create_payload_index(
+            collection_name=collection_name_with_prefix,
+            field_name="metadata.hash",
+            field_schema=models.KeywordIndexParams(
+                type=models.KeywordIndexType.KEYWORD,
+                is_tenant=False,
+                on_disk=self.QDRANT_ON_DISK,
+            ),
+        )
+        self.client.create_payload_index(
+            collection_name=collection_name_with_prefix,
+            field_name="metadata.file_id",
+            field_schema=models.KeywordIndexParams(
+                type=models.KeywordIndexType.KEYWORD,
+                is_tenant=False,
+                on_disk=self.QDRANT_ON_DISK,
+            ),
+        )
         log.info(f"collection {collection_name_with_prefix} successfully created!")
 
     def _create_collection_if_not_exists(self, collection_name, dimension):
@@ -151,23 +183,23 @@ class QdrantClient(VectorDBBase):
                     )
                 )
 
-            points = self.client.query_points(
+            points = self.client.scroll(
                 collection_name=f"{self.collection_prefix}_{collection_name}",
-                query_filter=models.Filter(should=field_conditions),
+                scroll_filter=models.Filter(should=field_conditions),
                 limit=limit,
             )
-            return self._result_to_get_result(points.points)
+            return self._result_to_get_result(points[0])
         except Exception as e:
             log.exception(f"Error querying a collection '{collection_name}': {e}")
             return None
 
     def get(self, collection_name: str) -> Optional[GetResult]:
         # Get all the items in the collection.
-        points = self.client.query_points(
+        points = self.client.scroll(
             collection_name=f"{self.collection_prefix}_{collection_name}",
             limit=NO_LIMIT,  # otherwise qdrant would set limit to 10!
         )
-        return self._result_to_get_result(points.points)
+        return self._result_to_get_result(points[0])
 
     def insert(self, collection_name: str, items: list[VectorItem]):
         # Insert the items into the collection, if the collection does not exist, it will be created.

backend/open_webui/retrieval/vector/dbs/qdrant_multitenancy.py

@@ -1,5 +1,5 @@
 import logging
-from typing import Optional, Tuple
+from typing import Optional, Tuple, List, Dict, Any
 from urllib.parse import urlparse
 
 import grpc
@@ -9,6 +9,9 @@ from open_webui.config import (
     QDRANT_ON_DISK,
     QDRANT_PREFER_GRPC,
     QDRANT_URI,
+    QDRANT_COLLECTION_PREFIX,
+    QDRANT_TIMEOUT,
+    QDRANT_HNSW_M,
 )
 from open_webui.env import SRC_LOG_LEVELS
 from open_webui.retrieval.vector.main import (
@@ -23,39 +26,62 @@ from qdrant_client.http.models import PointStruct
 from qdrant_client.models import models
 
 NO_LIMIT = 999999999
+TENANT_ID_FIELD = "tenant_id"
+DEFAULT_DIMENSION = 384
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
+def _tenant_filter(tenant_id: str) -> models.FieldCondition:
+    return models.FieldCondition(
+        key=TENANT_ID_FIELD, match=models.MatchValue(value=tenant_id)
+    )
+
+
+def _metadata_filter(key: str, value: Any) -> models.FieldCondition:
+    return models.FieldCondition(
+        key=f"metadata.{key}", match=models.MatchValue(value=value)
+    )
+
+
 class QdrantClient(VectorDBBase):
     def __init__(self):
-        self.collection_prefix = "open-webui"
+        self.collection_prefix = QDRANT_COLLECTION_PREFIX
         self.QDRANT_URI = QDRANT_URI
         self.QDRANT_API_KEY = QDRANT_API_KEY
         self.QDRANT_ON_DISK = QDRANT_ON_DISK
         self.PREFER_GRPC = QDRANT_PREFER_GRPC
         self.GRPC_PORT = QDRANT_GRPC_PORT
+        self.QDRANT_TIMEOUT = QDRANT_TIMEOUT
+        self.QDRANT_HNSW_M = QDRANT_HNSW_M
 
         if not self.QDRANT_URI:
-            self.client = None
-            return
+            raise ValueError(
+                "QDRANT_URI is not set. Please configure it in the environment variables."
+            )
 
         # Unified handling for either scheme
         parsed = urlparse(self.QDRANT_URI)
         host = parsed.hostname or self.QDRANT_URI
         http_port = parsed.port or 6333  # default REST port
 
-        if self.PREFER_GRPC:
-            self.client = Qclient(
+        self.client = (
+            Qclient(
                 host=host,
                 port=http_port,
                 grpc_port=self.GRPC_PORT,
                 prefer_grpc=self.PREFER_GRPC,
                 api_key=self.QDRANT_API_KEY,
+                timeout=self.QDRANT_TIMEOUT,
             )
-        else:
-            self.client = Qclient(url=self.QDRANT_URI, api_key=self.QDRANT_API_KEY)
+            if self.PREFER_GRPC
+            else Qclient(
+                url=self.QDRANT_URI,
+                api_key=self.QDRANT_API_KEY,
+                timeout=self.QDRANT_TIMEOUT,
+            )
+        )
 
         # Main collection types for multi-tenancy
         self.MEMORY_COLLECTION = f"{self.collection_prefix}_memories"
@@ -65,23 +91,13 @@ class QdrantClient(VectorDBBase):
         self.HASH_BASED_COLLECTION = f"{self.collection_prefix}_hash-based"
 
     def _result_to_get_result(self, points) -> GetResult:
-        ids = []
-        documents = []
-        metadatas = []
-
+        ids, documents, metadatas = [], [], []
         for point in points:
             payload = point.payload
             ids.append(point.id)
             documents.append(payload["text"])
             metadatas.append(payload["metadata"])
-
-        return GetResult(
-            **{
-                "ids": [ids],
-                "documents": [documents],
-                "metadatas": [metadatas],
-            }
-        )
+        return GetResult(ids=[ids], documents=[documents], metadatas=[metadatas])
 
     def _get_collection_and_tenant_id(self, collection_name: str) -> Tuple[str, str]:
         """
@@ -113,143 +129,53 @@ class QdrantClient(VectorDBBase):
         else:
             return self.KNOWLEDGE_COLLECTION, tenant_id
 
-    def _extract_error_message(self, exception):
-        """
-        Extract error message from either HTTP or gRPC exceptions
-
-        Returns:
-            tuple: (status_code, error_message)
-        """
-        # Check if it's an HTTP exception
-        if isinstance(exception, UnexpectedResponse):
-            try:
-                error_data = exception.structured()
-                error_msg = error_data.get("status", {}).get("error", "")
-                return exception.status_code, error_msg
-            except Exception as inner_e:
-                log.error(f"Failed to parse HTTP error: {inner_e}")
-                return exception.status_code, str(exception)
-
-        # Check if it's a gRPC exception
-        elif isinstance(exception, grpc.RpcError):
-            # Extract status code from gRPC error
-            status_code = None
-            if hasattr(exception, "code") and callable(exception.code):
-                status_code = exception.code().value[0]
-
-            # Extract error message
-            error_msg = str(exception)
-            if "details =" in error_msg:
-                # Parse the details line which contains the actual error message
-                try:
-                    details_line = [
-                        line.strip()
-                        for line in error_msg.split("\n")
-                        if "details =" in line
-                    ][0]
-                    error_msg = details_line.split("details =")[1].strip(' "')
-                except (IndexError, AttributeError):
-                    # Fall back to full message if parsing fails
-                    pass
-
-            return status_code, error_msg
-
-        # For any other type of exception
-        return None, str(exception)
-
-    def _is_collection_not_found_error(self, exception):
-        """
-        Check if the exception is due to collection not found, supporting both HTTP and gRPC
-        """
-        status_code, error_msg = self._extract_error_message(exception)
-
-        # HTTP error (404)
-        if (
-            status_code == 404
-            and "Collection" in error_msg
-            and "doesn't exist" in error_msg
-        ):
-            return True
-
-        # gRPC error (NOT_FOUND status)
-        if (
-            isinstance(exception, grpc.RpcError)
-            and exception.code() == grpc.StatusCode.NOT_FOUND
-        ):
-            return True
-
-        return False
-
-    def _is_dimension_mismatch_error(self, exception):
-        """
-        Check if the exception is due to dimension mismatch, supporting both HTTP and gRPC
-        """
-        status_code, error_msg = self._extract_error_message(exception)
-
-        # Common patterns in both HTTP and gRPC
-        return (
-            "Vector dimension error" in error_msg
-            or "dimensions mismatch" in error_msg
-            or "invalid vector size" in error_msg
-        )
-
-    def _create_multi_tenant_collection_if_not_exists(
-        self, mt_collection_name: str, dimension: int = 384
+    def _create_multi_tenant_collection(
+        self, mt_collection_name: str, dimension: int = DEFAULT_DIMENSION
     ):
         """
-        Creates a collection with multi-tenancy configuration if it doesn't exist.
-        Default dimension is set to 384 which corresponds to 'sentence-transformers/all-MiniLM-L6-v2'.
-        When creating collections dynamically (insert/upsert), the actual vector dimensions will be used.
+        Creates a collection with multi-tenancy configuration and payload indexes for tenant_id and metadata fields.
         """
-        try:
-            # Try to create the collection directly - will fail if it already exists
-            self.client.create_collection(
-                collection_name=mt_collection_name,
-                vectors_config=models.VectorParams(
-                    size=dimension,
-                    distance=models.Distance.COSINE,
-                    on_disk=self.QDRANT_ON_DISK,
-                ),
-                hnsw_config=models.HnswConfigDiff(
-                    payload_m=16,  # Enable per-tenant indexing
-                    m=0,
-                    on_disk=self.QDRANT_ON_DISK,
-                ),
-            )
+        self.client.create_collection(
+            collection_name=mt_collection_name,
+            vectors_config=models.VectorParams(
+                size=dimension,
+                distance=models.Distance.COSINE,
+                on_disk=self.QDRANT_ON_DISK,
+            ),
+            # Disable global index building due to multitenancy
+            # For more details https://qdrant.tech/documentation/guides/multiple-partitions/#calibrate-performance
+            hnsw_config=models.HnswConfigDiff(
+                payload_m=self.QDRANT_HNSW_M,
+                m=0,
+            ),
+        )
+        log.info(
+            f"Multi-tenant collection {mt_collection_name} created with dimension {dimension}!"
+        )
 
-            # Create tenant ID payload index
+        self.client.create_payload_index(
+            collection_name=mt_collection_name,
+            field_name=TENANT_ID_FIELD,
+            field_schema=models.KeywordIndexParams(
+                type=models.KeywordIndexType.KEYWORD,
+                is_tenant=True,
+                on_disk=self.QDRANT_ON_DISK,
+            ),
+        )
+
+        for field in ("metadata.hash", "metadata.file_id"):
             self.client.create_payload_index(
                 collection_name=mt_collection_name,
-                field_name="tenant_id",
+                field_name=field,
                 field_schema=models.KeywordIndexParams(
                     type=models.KeywordIndexType.KEYWORD,
-                    is_tenant=True,
                     on_disk=self.QDRANT_ON_DISK,
                 ),
-                wait=True,
             )
 
-            log.info(
-                f"Multi-tenant collection {mt_collection_name} created with dimension {dimension}!"
-            )
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            # Check for the specific error indicating collection already exists
-            status_code, error_msg = self._extract_error_message(e)
-
-            # HTTP status code 409 or gRPC ALREADY_EXISTS
-            if (isinstance(e, UnexpectedResponse) and status_code == 409) or (
-                isinstance(e, grpc.RpcError)
-                and e.code() == grpc.StatusCode.ALREADY_EXISTS
-            ):
-                if "already exists" in error_msg:
-                    log.debug(f"Collection {mt_collection_name} already exists")
-                    return
-            # If it's not an already exists error, re-raise
-            raise e
-        except Exception as e:
-            raise e
-
-    def _create_points(self, items: list[VectorItem], tenant_id: str):
+    def _create_points(
+        self, items: List[VectorItem], tenant_id: str
+    ) -> List[PointStruct]:
         """
         Create point structs from vector items with tenant ID.
         """
@@ -260,56 +186,42 @@ class QdrantClient(VectorDBBase):
                 payload={
                     "text": item["text"],
                     "metadata": item["metadata"],
-                    "tenant_id": tenant_id,
+                    TENANT_ID_FIELD: tenant_id,
                 },
             )
             for item in items
         ]
 
+    def _ensure_collection(
+        self, mt_collection_name: str, dimension: int = DEFAULT_DIMENSION
+    ):
+        """
+        Ensure the collection exists and payload indexes are created for tenant_id and metadata fields.
+        """
+        if not self.client.collection_exists(collection_name=mt_collection_name):
+            self._create_multi_tenant_collection(mt_collection_name, dimension)
+
     def has_collection(self, collection_name: str) -> bool:
         """
         Check if a logical collection exists by checking for any points with the tenant ID.
         """
         if not self.client:
             return False
-
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        # Create tenant filter
-        tenant_filter = models.FieldCondition(
-            key="tenant_id", match=models.MatchValue(value=tenant_id)
-        )
-
-        try:
-            # Try directly querying - most of the time collection should exist
-            response = self.client.query_points(
-                collection_name=mt_collection,
-                query_filter=models.Filter(must=[tenant_filter]),
-                limit=1,
-            )
-
-            # Collection exists with this tenant ID if there are points
-            return len(response.points) > 0
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            if self._is_collection_not_found_error(e):
-                log.debug(f"Collection {mt_collection} doesn't exist")
-                return False
-            else:
-                # For other API errors, log and return False
-                _, error_msg = self._extract_error_message(e)
-                log.warning(f"Unexpected Qdrant error: {error_msg}")
-                return False
-        except Exception as e:
-            # For any other errors, log and return False
-            log.debug(f"Error checking collection {mt_collection}: {e}")
+        if not self.client.collection_exists(collection_name=mt_collection):
             return False
+        tenant_filter = _tenant_filter(tenant_id)
+        count_result = self.client.count(
+            collection_name=mt_collection,
+            count_filter=models.Filter(must=[tenant_filter]),
+        )
+        return count_result.count > 0
 
     def delete(
         self,
         collection_name: str,
-        ids: Optional[list[str]] = None,
-        filter: Optional[dict] = None,
+        ids: Optional[List[str]] = None,
+        filter: Optional[Dict[str, Any]] = None,
     ):
         """
         Delete vectors by ID or filter from a collection with tenant isolation.
@@ -317,189 +229,76 @@ class QdrantClient(VectorDBBase):
         if not self.client:
             return None
 
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+        if not self.client.collection_exists(collection_name=mt_collection):
+            log.debug(f"Collection {mt_collection} doesn't exist, nothing to delete")
+            return None
 
-        # Create tenant filter
-        tenant_filter = models.FieldCondition(
-            key="tenant_id", match=models.MatchValue(value=tenant_id)
+        must_conditions = [_tenant_filter(tenant_id)]
+        should_conditions = []
+        if ids:
+            should_conditions = [_metadata_filter("id", id_value) for id_value in ids]
+        elif filter:
+            must_conditions += [_metadata_filter(k, v) for k, v in filter.items()]
+
+        return self.client.delete(
+            collection_name=mt_collection,
+            points_selector=models.FilterSelector(
+                filter=models.Filter(must=must_conditions, should=should_conditions)
+            ),
         )
 
-        must_conditions = [tenant_filter]
-        should_conditions = []
-
-        if ids:
-            for id_value in ids:
-                should_conditions.append(
-                    models.FieldCondition(
-                        key="metadata.id",
-                        match=models.MatchValue(value=id_value),
-                    ),
-                )
-        elif filter:
-            for key, value in filter.items():
-                must_conditions.append(
-                    models.FieldCondition(
-                        key=f"metadata.{key}",
-                        match=models.MatchValue(value=value),
-                    ),
-                )
-
-        try:
-            # Try to delete directly - most of the time collection should exist
-            update_result = self.client.delete(
-                collection_name=mt_collection,
-                points_selector=models.FilterSelector(
-                    filter=models.Filter(must=must_conditions, should=should_conditions)
-                ),
-            )
-
-            return update_result
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            if self._is_collection_not_found_error(e):
-                log.debug(
-                    f"Collection {mt_collection} doesn't exist, nothing to delete"
-                )
-                return None
-            else:
-                # For other API errors, log and re-raise
-                _, error_msg = self._extract_error_message(e)
-                log.warning(f"Unexpected Qdrant error: {error_msg}")
-                raise
-        except Exception as e:
-            # For non-Qdrant exceptions, re-raise
-            raise
-
     def search(
-        self, collection_name: str, vectors: list[list[float | int]], limit: int
+        self, collection_name: str, vectors: List[List[float | int]], limit: int
     ) -> Optional[SearchResult]:
         """
         Search for the nearest neighbor items based on the vectors with tenant isolation.
         """
-        if not self.client:
+        if not self.client or not vectors:
             return None
-
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        # Get the vector dimension from the query vector
-        dimension = len(vectors[0]) if vectors and len(vectors) > 0 else None
-
-        try:
-            # Try the search operation directly - most of the time collection should exist
-
-            # Create tenant filter
-            tenant_filter = models.FieldCondition(
-                key="tenant_id", match=models.MatchValue(value=tenant_id)
-            )
-
-            # Ensure vector dimensions match the collection
-            collection_dim = self.client.get_collection(
-                mt_collection
-            ).config.params.vectors.size
-
-            if collection_dim != dimension:
-                if collection_dim < dimension:
-                    vectors = [vector[:collection_dim] for vector in vectors]
-                else:
-                    vectors = [
-                        vector + [0] * (collection_dim - dimension)
-                        for vector in vectors
-                    ]
-
-            # Search with tenant filter
-            prefetch_query = models.Prefetch(
-                filter=models.Filter(must=[tenant_filter]),
-                limit=NO_LIMIT,
-            )
-            query_response = self.client.query_points(
-                collection_name=mt_collection,
-                query=vectors[0],
-                prefetch=prefetch_query,
-                limit=limit,
-            )
-
-            get_result = self._result_to_get_result(query_response.points)
-            return SearchResult(
-                ids=get_result.ids,
-                documents=get_result.documents,
-                metadatas=get_result.metadatas,
-                # qdrant distance is [-1, 1], normalize to [0, 1]
-                distances=[
-                    [(point.score + 1.0) / 2.0 for point in query_response.points]
-                ],
-            )
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            if self._is_collection_not_found_error(e):
-                log.debug(
-                    f"Collection {mt_collection} doesn't exist, search returns None"
-                )
-                return None
-            else:
-                # For other API errors, log and re-raise
-                _, error_msg = self._extract_error_message(e)
-                log.warning(f"Unexpected Qdrant error during search: {error_msg}")
-                raise
-        except Exception as e:
-            # For non-Qdrant exceptions, log and return None
-            log.exception(f"Error searching collection '{collection_name}': {e}")
+        if not self.client.collection_exists(collection_name=mt_collection):
+            log.debug(f"Collection {mt_collection} doesn't exist, search returns None")
             return None
 
-    def query(self, collection_name: str, filter: dict, limit: Optional[int] = None):
+        tenant_filter = _tenant_filter(tenant_id)
+        query_response = self.client.query_points(
+            collection_name=mt_collection,
+            query=vectors[0],
+            limit=limit,
+            query_filter=models.Filter(must=[tenant_filter]),
+        )
+        get_result = self._result_to_get_result(query_response.points)
+        return SearchResult(
+            ids=get_result.ids,
+            documents=get_result.documents,
+            metadatas=get_result.metadatas,
+            distances=[[(point.score + 1.0) / 2.0 for point in query_response.points]],
+        )
+
+    def query(
+        self, collection_name: str, filter: Dict[str, Any], limit: Optional[int] = None
+    ):
         """
         Query points with filters and tenant isolation.
         """
         if not self.client:
             return None
-
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        # Set default limit if not provided
+        if not self.client.collection_exists(collection_name=mt_collection):
+            log.debug(f"Collection {mt_collection} doesn't exist, query returns None")
+            return None
         if limit is None:
             limit = NO_LIMIT
-
-        # Create tenant filter
-        tenant_filter = models.FieldCondition(
-            key="tenant_id", match=models.MatchValue(value=tenant_id)
-        )
-
-        # Create metadata filters
-        field_conditions = []
-        for key, value in filter.items():
-            field_conditions.append(
-                models.FieldCondition(
-                    key=f"metadata.{key}", match=models.MatchValue(value=value)
-                )
-            )
-
-        # Combine tenant filter with metadata filters
+        tenant_filter = _tenant_filter(tenant_id)
+        field_conditions = [_metadata_filter(k, v) for k, v in filter.items()]
         combined_filter = models.Filter(must=[tenant_filter, *field_conditions])
-
-        try:
-            # Try the query directly - most of the time collection should exist
-            points = self.client.query_points(
-                collection_name=mt_collection,
-                query_filter=combined_filter,
-                limit=limit,
-            )
-
-            return self._result_to_get_result(points.points)
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            if self._is_collection_not_found_error(e):
-                log.debug(
-                    f"Collection {mt_collection} doesn't exist, query returns None"
-                )
-                return None
-            else:
-                # For other API errors, log and re-raise
-                _, error_msg = self._extract_error_message(e)
-                log.warning(f"Unexpected Qdrant error during query: {error_msg}")
-                raise
-        except Exception as e:
-            # For non-Qdrant exceptions, log and re-raise
-            log.exception(f"Error querying collection '{collection_name}': {e}")
-            return None
+        points = self.client.scroll(
+            collection_name=mt_collection,
+            scroll_filter=combined_filter,
+            limit=limit,
+        )
+        return self._result_to_get_result(points[0])
 
     def get(self, collection_name: str) -> Optional[GetResult]:
         """
@@ -507,169 +306,36 @@ class QdrantClient(VectorDBBase):
         """
         if not self.client:
             return None
-
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        # Create tenant filter
-        tenant_filter = models.FieldCondition(
-            key="tenant_id", match=models.MatchValue(value=tenant_id)
-        )
-
-        try:
-            # Try to get points directly - most of the time collection should exist
-            points = self.client.query_points(
-                collection_name=mt_collection,
-                query_filter=models.Filter(must=[tenant_filter]),
-                limit=NO_LIMIT,
-            )
-
-            return self._result_to_get_result(points.points)
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            if self._is_collection_not_found_error(e):
-                log.debug(f"Collection {mt_collection} doesn't exist, get returns None")
-                return None
-            else:
-                # For other API errors, log and re-raise
-                _, error_msg = self._extract_error_message(e)
-                log.warning(f"Unexpected Qdrant error during get: {error_msg}")
-                raise
-        except Exception as e:
-            # For non-Qdrant exceptions, log and return None
-            log.exception(f"Error getting collection '{collection_name}': {e}")
+        if not self.client.collection_exists(collection_name=mt_collection):
+            log.debug(f"Collection {mt_collection} doesn't exist, get returns None")
             return None
-
-    def _handle_operation_with_error_retry(
-        self, operation_name, mt_collection, points, dimension
-    ):
-        """
-        Private helper to handle common error cases for insert and upsert operations.
-
-        Args:
-            operation_name: 'insert' or 'upsert'
-            mt_collection: The multi-tenant collection name
-            points: The vector points to insert/upsert
-            dimension: The dimension of the vectors
-
-        Returns:
-            The operation result (for upsert) or None (for insert)
-        """
-        try:
-            if operation_name == "insert":
-                self.client.upload_points(mt_collection, points)
-                return None
-            else:  # upsert
-                return self.client.upsert(mt_collection, points)
-        except (UnexpectedResponse, grpc.RpcError) as e:
-            # Handle collection not found
-            if self._is_collection_not_found_error(e):
-                log.info(
-                    f"Collection {mt_collection} doesn't exist. Creating it with dimension {dimension}."
-                )
-                # Create collection with correct dimensions from our vectors
-                self._create_multi_tenant_collection_if_not_exists(
-                    mt_collection_name=mt_collection, dimension=dimension
-                )
-                # Try operation again - no need for dimension adjustment since we just created with correct dimensions
-                if operation_name == "insert":
-                    self.client.upload_points(mt_collection, points)
-                    return None
-                else:  # upsert
-                    return self.client.upsert(mt_collection, points)
-
-            # Handle dimension mismatch
-            elif self._is_dimension_mismatch_error(e):
-                # For dimension errors, the collection must exist, so get its configuration
-                mt_collection_info = self.client.get_collection(mt_collection)
-                existing_size = mt_collection_info.config.params.vectors.size
-
-                log.info(
-                    f"Dimension mismatch: Collection {mt_collection} expects {existing_size}, got {dimension}"
-                )
-
-                if existing_size < dimension:
-                    # Truncate vectors to fit
-                    log.info(
-                        f"Truncating vectors from {dimension} to {existing_size} dimensions"
-                    )
-                    points = [
-                        PointStruct(
-                            id=point.id,
-                            vector=point.vector[:existing_size],
-                            payload=point.payload,
-                        )
-                        for point in points
-                    ]
-                elif existing_size > dimension:
-                    # Pad vectors with zeros
-                    log.info(
-                        f"Padding vectors from {dimension} to {existing_size} dimensions with zeros"
-                    )
-                    points = [
-                        PointStruct(
-                            id=point.id,
-                            vector=point.vector
-                            + [0] * (existing_size - len(point.vector)),
-                            payload=point.payload,
-                        )
-                        for point in points
-                    ]
-                # Try operation again with adjusted dimensions
-                if operation_name == "insert":
-                    self.client.upload_points(mt_collection, points)
-                    return None
-                else:  # upsert
-                    return self.client.upsert(mt_collection, points)
-            else:
-                # Not a known error we can handle, log and re-raise
-                _, error_msg = self._extract_error_message(e)
-                log.warning(f"Unhandled Qdrant error: {error_msg}")
-                raise
-        except Exception as e:
-            # For non-Qdrant exceptions, re-raise
-            raise
-
-    def insert(self, collection_name: str, items: list[VectorItem]):
-        """
-        Insert items with tenant ID.
-        """
-        if not self.client or not items:
-            return None
-
-        # Map to multi-tenant collection and tenant ID
-        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        # Get dimensions from the actual vectors
-        dimension = len(items[0]["vector"]) if items else None
-
-        # Create points with tenant ID
-        points = self._create_points(items, tenant_id)
-
-        # Handle the operation with error retry
-        return self._handle_operation_with_error_retry(
-            "insert", mt_collection, points, dimension
+        tenant_filter = _tenant_filter(tenant_id)
+        points = self.client.scroll(
+            collection_name=mt_collection,
+            scroll_filter=models.Filter(must=[tenant_filter]),
+            limit=NO_LIMIT,
         )
+        return self._result_to_get_result(points[0])
 
-    def upsert(self, collection_name: str, items: list[VectorItem]):
+    def upsert(self, collection_name: str, items: List[VectorItem]):
         """
         Upsert items with tenant ID.
         """
         if not self.client or not items:
             return None
-
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        # Get dimensions from the actual vectors
-        dimension = len(items[0]["vector"]) if items else None
-
-        # Create points with tenant ID
+        dimension = len(items[0]["vector"])
+        self._ensure_collection(mt_collection, dimension)
         points = self._create_points(items, tenant_id)
+        self.client.upload_points(mt_collection, points)
+        return None
 
-        # Handle the operation with error retry
-        return self._handle_operation_with_error_retry(
-            "upsert", mt_collection, points, dimension
-        )
+    def insert(self, collection_name: str, items: List[VectorItem]):
+        """
+        Insert items with tenant ID.
+        """
+        return self.upsert(collection_name, items)
 
     def reset(self):
         """
@@ -677,11 +343,9 @@ class QdrantClient(VectorDBBase):
         """
         if not self.client:
             return None
-
-        collection_names = self.client.get_collections().collections
-        for collection_name in collection_names:
-            if collection_name.name.startswith(self.collection_prefix):
-                self.client.delete_collection(collection_name=collection_name.name)
+        for collection in self.client.get_collections().collections:
+            if collection.name.startswith(self.collection_prefix):
+                self.client.delete_collection(collection_name=collection.name)
 
     def delete_collection(self, collection_name: str):
         """
@@ -689,24 +353,13 @@ class QdrantClient(VectorDBBase):
         """
         if not self.client:
             return None
-
-        # Map to multi-tenant collection and tenant ID
         mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
-
-        tenant_filter = models.FieldCondition(
-            key="tenant_id", match=models.MatchValue(value=tenant_id)
-        )
-
-        field_conditions = [tenant_filter]
-
-        update_result = self.client.delete(
+        if not self.client.collection_exists(collection_name=mt_collection):
+            log.debug(f"Collection {mt_collection} doesn't exist, nothing to delete")
+            return None
+        self.client.delete(
             collection_name=mt_collection,
             points_selector=models.FilterSelector(
-                filter=models.Filter(must=field_conditions)
+                filter=models.Filter(must=[_tenant_filter(tenant_id)])
             ),
         )
-
-        if self.client.get_collection(mt_collection).points_count == 0:
-            self.client.delete_collection(mt_collection)
-
-        return update_result

backend/open_webui/retrieval/vector/dbs/s3vector.py

@@ -0,0 +1,752 @@
+from backend.open_webui.retrieval.vector.utils import stringify_metadata
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    GetResult,
+    SearchResult,
+)
+from open_webui.config import S3_VECTOR_BUCKET_NAME, S3_VECTOR_REGION
+from open_webui.env import SRC_LOG_LEVELS
+from typing import List, Optional, Dict, Any, Union
+import logging
+import boto3
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class S3VectorClient(VectorDBBase):
+    """
+    AWS S3 Vector integration for Open WebUI Knowledge.
+    """
+
+    def __init__(self):
+        self.bucket_name = S3_VECTOR_BUCKET_NAME
+        self.region = S3_VECTOR_REGION
+
+        # Simple validation - log warnings instead of raising exceptions
+        if not self.bucket_name:
+            log.warning("S3_VECTOR_BUCKET_NAME not set - S3Vector will not work")
+        if not self.region:
+            log.warning("S3_VECTOR_REGION not set - S3Vector will not work")
+
+        if self.bucket_name and self.region:
+            try:
+                self.client = boto3.client("s3vectors", region_name=self.region)
+                log.info(
+                    f"S3Vector client initialized for bucket '{self.bucket_name}' in region '{self.region}'"
+                )
+            except Exception as e:
+                log.error(f"Failed to initialize S3Vector client: {e}")
+                self.client = None
+        else:
+            self.client = None
+
+    def _create_index(
+        self,
+        index_name: str,
+        dimension: int,
+        data_type: str = "float32",
+        distance_metric: str = "cosine",
+    ) -> None:
+        """
+        Create a new index in the S3 vector bucket for the given collection if it does not exist.
+        """
+        if self.has_collection(index_name):
+            log.debug(f"Index '{index_name}' already exists, skipping creation")
+            return
+
+        try:
+            self.client.create_index(
+                vectorBucketName=self.bucket_name,
+                indexName=index_name,
+                dataType=data_type,
+                dimension=dimension,
+                distanceMetric=distance_metric,
+            )
+            log.info(
+                f"Created S3 index: {index_name} (dim={dimension}, type={data_type}, metric={distance_metric})"
+            )
+        except Exception as e:
+            log.error(f"Error creating S3 index '{index_name}': {e}")
+            raise
+
+    def _filter_metadata(
+        self, metadata: Dict[str, Any], item_id: str
+    ) -> Dict[str, Any]:
+        """
+        Filter vector metadata keys to comply with S3 Vector API limit of 10 keys maximum.
+        """
+        if not isinstance(metadata, dict) or len(metadata) <= 10:
+            return metadata
+
+        # Keep only the first 10 keys, prioritizing important ones based on actual Open WebUI metadata
+        important_keys = [
+            "text",  # The actual document content
+            "file_id",  # File ID
+            "source",  # Document source file
+            "title",  # Document title
+            "page",  # Page number
+            "total_pages",  # Total pages in document
+            "embedding_config",  # Embedding configuration
+            "created_by",  # User who created it
+            "name",  # Document name
+            "hash",  # Content hash
+        ]
+        filtered_metadata = {}
+
+        # First, add important keys if they exist
+        for key in important_keys:
+            if key in metadata:
+                filtered_metadata[key] = metadata[key]
+            if len(filtered_metadata) >= 10:
+                break
+
+        # If we still have room, add other keys
+        if len(filtered_metadata) < 10:
+            for key, value in metadata.items():
+                if key not in filtered_metadata:
+                    filtered_metadata[key] = value
+                    if len(filtered_metadata) >= 10:
+                        break
+
+        log.warning(
+            f"Metadata for key '{item_id}' had {len(metadata)} keys, limited to 10 keys"
+        )
+        return filtered_metadata
+
+    def has_collection(self, collection_name: str) -> bool:
+        """
+        Check if a vector index (collection) exists in the S3 vector bucket.
+        """
+
+        try:
+            response = self.client.list_indexes(vectorBucketName=self.bucket_name)
+            indexes = response.get("indexes", [])
+            return any(idx.get("indexName") == collection_name for idx in indexes)
+        except Exception as e:
+            log.error(f"Error listing indexes: {e}")
+            return False
+
+    def delete_collection(self, collection_name: str) -> None:
+        """
+        Delete an entire S3 Vector index/collection.
+        """
+
+        if not self.has_collection(collection_name):
+            log.warning(
+                f"Collection '{collection_name}' does not exist, nothing to delete"
+            )
+            return
+
+        try:
+            log.info(f"Deleting collection '{collection_name}'")
+            self.client.delete_index(
+                vectorBucketName=self.bucket_name, indexName=collection_name
+            )
+            log.info(f"Successfully deleted collection '{collection_name}'")
+        except Exception as e:
+            log.error(f"Error deleting collection '{collection_name}': {e}")
+            raise
+
+    def insert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """
+        Insert vector items into the S3 Vector index. Create index if it does not exist.
+        """
+        if not items:
+            log.warning("No items to insert")
+            return
+
+        dimension = len(items[0]["vector"])
+
+        try:
+            if not self.has_collection(collection_name):
+                log.info(f"Index '{collection_name}' does not exist. Creating index.")
+                self._create_index(
+                    index_name=collection_name,
+                    dimension=dimension,
+                    data_type="float32",
+                    distance_metric="cosine",
+                )
+
+            # Prepare vectors for insertion
+            vectors = []
+            for item in items:
+                # Ensure vector data is in the correct format for S3 Vector API
+                vector_data = item["vector"]
+                if isinstance(vector_data, list):
+                    # Convert list to float32 values as required by S3 Vector API
+                    vector_data = [float(x) for x in vector_data]
+
+                # Prepare metadata, ensuring the text field is preserved
+                metadata = item.get("metadata", {}).copy()
+
+                # Add the text field to metadata so it's available for retrieval
+                metadata["text"] = item["text"]
+
+                # Convert metadata to string format for consistency
+                metadata = stringify_metadata(metadata)
+
+                # Filter metadata to comply with S3 Vector API limit of 10 keys
+                metadata = self._filter_metadata(metadata, item["id"])
+
+                vectors.append(
+                    {
+                        "key": item["id"],
+                        "data": {"float32": vector_data},
+                        "metadata": metadata,
+                    }
+                )
+            # Insert vectors
+            self.client.put_vectors(
+                vectorBucketName=self.bucket_name,
+                indexName=collection_name,
+                vectors=vectors,
+            )
+            log.info(f"Inserted {len(vectors)} vectors into index '{collection_name}'.")
+        except Exception as e:
+            log.error(f"Error inserting vectors: {e}")
+            raise
+
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """
+        Insert or update vector items in the S3 Vector index. Create index if it does not exist.
+        """
+        if not items:
+            log.warning("No items to upsert")
+            return
+
+        dimension = len(items[0]["vector"])
+        log.info(f"Upsert dimension: {dimension}")
+
+        try:
+            if not self.has_collection(collection_name):
+                log.info(
+                    f"Index '{collection_name}' does not exist. Creating index for upsert."
+                )
+                self._create_index(
+                    index_name=collection_name,
+                    dimension=dimension,
+                    data_type="float32",
+                    distance_metric="cosine",
+                )
+
+            # Prepare vectors for upsert
+            vectors = []
+            for item in items:
+                # Ensure vector data is in the correct format for S3 Vector API
+                vector_data = item["vector"]
+                if isinstance(vector_data, list):
+                    # Convert list to float32 values as required by S3 Vector API
+                    vector_data = [float(x) for x in vector_data]
+
+                # Prepare metadata, ensuring the text field is preserved
+                metadata = item.get("metadata", {}).copy()
+                # Add the text field to metadata so it's available for retrieval
+                metadata["text"] = item["text"]
+
+                # Convert metadata to string format for consistency
+                metadata = stringify_metadata(metadata)
+
+                # Filter metadata to comply with S3 Vector API limit of 10 keys
+                metadata = self._filter_metadata(metadata, item["id"])
+
+                vectors.append(
+                    {
+                        "key": item["id"],
+                        "data": {"float32": vector_data},
+                        "metadata": metadata,
+                    }
+                )
+            # Upsert vectors (using put_vectors for upsert semantics)
+            log.info(
+                f"Upserting {len(vectors)} vectors. First vector sample: key={vectors[0]['key']}, data_type={type(vectors[0]['data']['float32'])}, data_len={len(vectors[0]['data']['float32'])}"
+            )
+            self.client.put_vectors(
+                vectorBucketName=self.bucket_name,
+                indexName=collection_name,
+                vectors=vectors,
+            )
+            log.info(f"Upserted {len(vectors)} vectors into index '{collection_name}'.")
+        except Exception as e:
+            log.error(f"Error upserting vectors: {e}")
+            raise
+
+    def search(
+        self, collection_name: str, vectors: List[List[Union[float, int]]], limit: int
+    ) -> Optional[SearchResult]:
+        """
+        Search for similar vectors in a collection using multiple query vectors.
+        """
+
+        if not self.has_collection(collection_name):
+            log.warning(f"Collection '{collection_name}' does not exist")
+            return None
+
+        if not vectors:
+            log.warning("No query vectors provided")
+            return None
+
+        try:
+            log.info(
+                f"Searching collection '{collection_name}' with {len(vectors)} query vectors, limit={limit}"
+            )
+
+            # Initialize result lists
+            all_ids = []
+            all_documents = []
+            all_metadatas = []
+            all_distances = []
+
+            # Process each query vector
+            for i, query_vector in enumerate(vectors):
+                log.debug(f"Processing query vector {i+1}/{len(vectors)}")
+
+                # Prepare the query vector in S3 Vector format
+                query_vector_dict = {"float32": [float(x) for x in query_vector]}
+
+                # Call S3 Vector query API
+                response = self.client.query_vectors(
+                    vectorBucketName=self.bucket_name,
+                    indexName=collection_name,
+                    topK=limit,
+                    queryVector=query_vector_dict,
+                    returnMetadata=True,
+                    returnDistance=True,
+                )
+
+                # Process results for this query
+                query_ids = []
+                query_documents = []
+                query_metadatas = []
+                query_distances = []
+
+                result_vectors = response.get("vectors", [])
+
+                for vector in result_vectors:
+                    vector_id = vector.get("key")
+                    vector_metadata = vector.get("metadata", {})
+                    vector_distance = vector.get("distance", 0.0)
+
+                    # Extract document text from metadata
+                    document_text = ""
+                    if isinstance(vector_metadata, dict):
+                        # Get the text field first (highest priority)
+                        document_text = vector_metadata.get("text")
+                        if not document_text:
+                            # Fallback to other possible text fields
+                            document_text = (
+                                vector_metadata.get("content")
+                                or vector_metadata.get("document")
+                                or vector_id
+                            )
+                    else:
+                        document_text = vector_id
+
+                    query_ids.append(vector_id)
+                    query_documents.append(document_text)
+                    query_metadatas.append(vector_metadata)
+                    query_distances.append(vector_distance)
+
+                # Add this query's results to the overall results
+                all_ids.append(query_ids)
+                all_documents.append(query_documents)
+                all_metadatas.append(query_metadatas)
+                all_distances.append(query_distances)
+
+            log.info(f"Search completed. Found results for {len(all_ids)} queries")
+
+            # Return SearchResult format
+            return SearchResult(
+                ids=all_ids if all_ids else None,
+                documents=all_documents if all_documents else None,
+                metadatas=all_metadatas if all_metadatas else None,
+                distances=all_distances if all_distances else None,
+            )
+
+        except Exception as e:
+            log.error(f"Error searching collection '{collection_name}': {str(e)}")
+            # Handle specific AWS exceptions
+            if hasattr(e, "response") and "Error" in e.response:
+                error_code = e.response["Error"]["Code"]
+                if error_code == "NotFoundException":
+                    log.warning(f"Collection '{collection_name}' not found")
+                    return None
+                elif error_code == "ValidationException":
+                    log.error(f"Invalid query vector dimensions or parameters")
+                    return None
+                elif error_code == "AccessDeniedException":
+                    log.error(
+                        f"Access denied for collection '{collection_name}'. Check permissions."
+                    )
+                    return None
+            raise
+
+    def query(
+        self, collection_name: str, filter: Dict, limit: Optional[int] = None
+    ) -> Optional[GetResult]:
+        """
+        Query vectors from a collection using metadata filter.
+        """
+
+        if not self.has_collection(collection_name):
+            log.warning(f"Collection '{collection_name}' does not exist")
+            return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+
+        if not filter:
+            log.warning("No filter provided, returning all vectors")
+            return self.get(collection_name)
+
+        try:
+            log.info(f"Querying collection '{collection_name}' with filter: {filter}")
+
+            # For S3 Vector, we need to use list_vectors and then filter results
+            # Since S3 Vector may not support complex server-side filtering,
+            # we'll retrieve all vectors and filter client-side
+
+            # Get all vectors first
+            all_vectors_result = self.get(collection_name)
+
+            if not all_vectors_result or not all_vectors_result.ids:
+                log.warning("No vectors found in collection")
+                return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+
+            # Extract the lists from the result
+            all_ids = all_vectors_result.ids[0] if all_vectors_result.ids else []
+            all_documents = (
+                all_vectors_result.documents[0] if all_vectors_result.documents else []
+            )
+            all_metadatas = (
+                all_vectors_result.metadatas[0] if all_vectors_result.metadatas else []
+            )
+
+            # Apply client-side filtering
+            filtered_ids = []
+            filtered_documents = []
+            filtered_metadatas = []
+
+            for i, metadata in enumerate(all_metadatas):
+                if self._matches_filter(metadata, filter):
+                    if i < len(all_ids):
+                        filtered_ids.append(all_ids[i])
+                    if i < len(all_documents):
+                        filtered_documents.append(all_documents[i])
+                    filtered_metadatas.append(metadata)
+
+                    # Apply limit if specified
+                    if limit and len(filtered_ids) >= limit:
+                        break
+
+            log.info(
+                f"Filter applied: {len(filtered_ids)} vectors match out of {len(all_ids)} total"
+            )
+
+            # Return GetResult format
+            if filtered_ids:
+                return GetResult(
+                    ids=[filtered_ids],
+                    documents=[filtered_documents],
+                    metadatas=[filtered_metadatas],
+                )
+            else:
+                return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+
+        except Exception as e:
+            log.error(f"Error querying collection '{collection_name}': {str(e)}")
+            # Handle specific AWS exceptions
+            if hasattr(e, "response") and "Error" in e.response:
+                error_code = e.response["Error"]["Code"]
+                if error_code == "NotFoundException":
+                    log.warning(f"Collection '{collection_name}' not found")
+                    return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+                elif error_code == "AccessDeniedException":
+                    log.error(
+                        f"Access denied for collection '{collection_name}'. Check permissions."
+                    )
+                    return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+            raise
+
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        """
+        Retrieve all vectors from a collection.
+        """
+
+        if not self.has_collection(collection_name):
+            log.warning(f"Collection '{collection_name}' does not exist")
+            return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+
+        try:
+            log.info(f"Retrieving all vectors from collection '{collection_name}'")
+
+            # Initialize result lists
+            all_ids = []
+            all_documents = []
+            all_metadatas = []
+
+            # Handle pagination
+            next_token = None
+
+            while True:
+                # Prepare request parameters
+                request_params = {
+                    "vectorBucketName": self.bucket_name,
+                    "indexName": collection_name,
+                    "returnData": False,  # Don't include vector data (not needed for get)
+                    "returnMetadata": True,  # Include metadata
+                    "maxResults": 500,  # Use reasonable page size
+                }
+
+                if next_token:
+                    request_params["nextToken"] = next_token
+
+                # Call S3 Vector API
+                response = self.client.list_vectors(**request_params)
+
+                # Process vectors in this page
+                vectors = response.get("vectors", [])
+
+                for vector in vectors:
+                    vector_id = vector.get("key")
+                    vector_data = vector.get("data", {})
+                    vector_metadata = vector.get("metadata", {})
+
+                    # Extract the actual vector array
+                    vector_array = vector_data.get("float32", [])
+
+                    # For documents, we try to extract text from metadata or use the vector ID
+                    document_text = ""
+                    if isinstance(vector_metadata, dict):
+                        # Get the text field first (highest priority)
+                        document_text = vector_metadata.get("text")
+                        if not document_text:
+                            # Fallback to other possible text fields
+                            document_text = (
+                                vector_metadata.get("content")
+                                or vector_metadata.get("document")
+                                or vector_id
+                            )
+
+                        # Log the actual content for debugging
+                        log.debug(
+                            f"Document text preview (first 200 chars): {str(document_text)[:200]}"
+                        )
+                    else:
+                        document_text = vector_id
+
+                    all_ids.append(vector_id)
+                    all_documents.append(document_text)
+                    all_metadatas.append(vector_metadata)
+
+                # Check if there are more pages
+                next_token = response.get("nextToken")
+                if not next_token:
+                    break
+
+            log.info(
+                f"Retrieved {len(all_ids)} vectors from collection '{collection_name}'"
+            )
+
+            # Return in GetResult format
+            # The Open WebUI GetResult expects lists of lists, so we wrap each list
+            if all_ids:
+                return GetResult(
+                    ids=[all_ids], documents=[all_documents], metadatas=[all_metadatas]
+                )
+            else:
+                return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+
+        except Exception as e:
+            log.error(
+                f"Error retrieving vectors from collection '{collection_name}': {str(e)}"
+            )
+            # Handle specific AWS exceptions
+            if hasattr(e, "response") and "Error" in e.response:
+                error_code = e.response["Error"]["Code"]
+                if error_code == "NotFoundException":
+                    log.warning(f"Collection '{collection_name}' not found")
+                    return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+                elif error_code == "AccessDeniedException":
+                    log.error(
+                        f"Access denied for collection '{collection_name}'. Check permissions."
+                    )
+                    return GetResult(ids=[[]], documents=[[]], metadatas=[[]])
+            raise
+
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[List[str]] = None,
+        filter: Optional[Dict] = None,
+    ) -> None:
+        """
+        Delete vectors by ID or filter from a collection.
+        """
+
+        if not self.has_collection(collection_name):
+            log.warning(
+                f"Collection '{collection_name}' does not exist, nothing to delete"
+            )
+            return
+
+        # Check if this is a knowledge collection (not file-specific)
+        is_knowledge_collection = not collection_name.startswith("file-")
+
+        try:
+            if ids:
+                # Delete by specific vector IDs/keys
+                log.info(
+                    f"Deleting {len(ids)} vectors by IDs from collection '{collection_name}'"
+                )
+                self.client.delete_vectors(
+                    vectorBucketName=self.bucket_name,
+                    indexName=collection_name,
+                    keys=ids,
+                )
+                log.info(f"Deleted {len(ids)} vectors from index '{collection_name}'")
+
+            elif filter:
+                # Handle filter-based deletion
+                log.info(
+                    f"Deleting vectors by filter from collection '{collection_name}': {filter}"
+                )
+
+                # If this is a knowledge collection and we have a file_id filter,
+                # also clean up the corresponding file-specific collection
+                if is_knowledge_collection and "file_id" in filter:
+                    file_id = filter["file_id"]
+                    file_collection_name = f"file-{file_id}"
+                    if self.has_collection(file_collection_name):
+                        log.info(
+                            f"Found related file-specific collection '{file_collection_name}', deleting it to prevent duplicates"
+                        )
+                        self.delete_collection(file_collection_name)
+
+                # For the main collection, implement query-then-delete
+                # First, query to get IDs matching the filter
+                query_result = self.query(collection_name, filter)
+                if query_result and query_result.ids and query_result.ids[0]:
+                    matching_ids = query_result.ids[0]
+                    log.info(
+                        f"Found {len(matching_ids)} vectors matching filter, deleting them"
+                    )
+
+                    # Delete the matching vectors by ID
+                    self.client.delete_vectors(
+                        vectorBucketName=self.bucket_name,
+                        indexName=collection_name,
+                        keys=matching_ids,
+                    )
+                    log.info(
+                        f"Deleted {len(matching_ids)} vectors from index '{collection_name}' using filter"
+                    )
+                else:
+                    log.warning("No vectors found matching the filter criteria")
+            else:
+                log.warning("No IDs or filter provided for deletion")
+        except Exception as e:
+            log.error(
+                f"Error deleting vectors from collection '{collection_name}': {e}"
+            )
+            raise
+
+    def reset(self) -> None:
+        """
+        Reset/clear all vector data. For S3 Vector, this deletes all indexes.
+        """
+
+        try:
+            log.warning(
+                "Reset called - this will delete all vector indexes in the S3 bucket"
+            )
+
+            # List all indexes
+            response = self.client.list_indexes(vectorBucketName=self.bucket_name)
+            indexes = response.get("indexes", [])
+
+            if not indexes:
+                log.warning("No indexes found to delete")
+                return
+
+            # Delete all indexes
+            deleted_count = 0
+            for index in indexes:
+                index_name = index.get("indexName")
+                if index_name:
+                    try:
+                        self.client.delete_index(
+                            vectorBucketName=self.bucket_name, indexName=index_name
+                        )
+                        deleted_count += 1
+                        log.info(f"Deleted index: {index_name}")
+                    except Exception as e:
+                        log.error(f"Error deleting index '{index_name}': {e}")
+
+            log.info(f"Reset completed: deleted {deleted_count} indexes")
+
+        except Exception as e:
+            log.error(f"Error during reset: {e}")
+            raise
+
+    def _matches_filter(self, metadata: Dict[str, Any], filter: Dict[str, Any]) -> bool:
+        """
+        Check if metadata matches the given filter conditions.
+        """
+        if not isinstance(metadata, dict) or not isinstance(filter, dict):
+            return False
+
+        # Check each filter condition
+        for key, expected_value in filter.items():
+            # Handle special operators
+            if key.startswith("$"):
+                if key == "$and":
+                    # All conditions must match
+                    if not isinstance(expected_value, list):
+                        continue
+                    for condition in expected_value:
+                        if not self._matches_filter(metadata, condition):
+                            return False
+                elif key == "$or":
+                    # At least one condition must match
+                    if not isinstance(expected_value, list):
+                        continue
+                    any_match = False
+                    for condition in expected_value:
+                        if self._matches_filter(metadata, condition):
+                            any_match = True
+                            break
+                    if not any_match:
+                        return False
+                continue
+
+            # Get the actual value from metadata
+            actual_value = metadata.get(key)
+
+            # Handle different types of expected values
+            if isinstance(expected_value, dict):
+                # Handle comparison operators
+                for op, op_value in expected_value.items():
+                    if op == "$eq":
+                        if actual_value != op_value:
+                            return False
+                    elif op == "$ne":
+                        if actual_value == op_value:
+                            return False
+                    elif op == "$in":
+                        if (
+                            not isinstance(op_value, list)
+                            or actual_value not in op_value
+                        ):
+                            return False
+                    elif op == "$nin":
+                        if isinstance(op_value, list) and actual_value in op_value:
+                            return False
+                    elif op == "$exists":
+                        if bool(op_value) != (key in metadata):
+                            return False
+                    # Add more operators as needed
+            else:
+                # Simple equality check
+                if actual_value != expected_value:
+                    return False
+
+        return True

backend/open_webui/retrieval/vector/factory.py

@@ -30,6 +30,10 @@ class Vector:
                 from open_webui.retrieval.vector.dbs.pinecone import PineconeClient
 
                 return PineconeClient()
+            case VectorType.S3VECTOR:
+                from open_webui.retrieval.vector.dbs.s3vector import S3VectorClient
+
+                return S3VectorClient()
             case VectorType.OPENSEARCH:
                 from open_webui.retrieval.vector.dbs.opensearch import OpenSearchClient
 
@@ -48,6 +52,10 @@ class Vector:
                 from open_webui.retrieval.vector.dbs.chroma import ChromaClient
 
                 return ChromaClient()
+            case VectorType.ORACLE23AI:
+                from open_webui.retrieval.vector.dbs.oracle23ai import Oracle23aiClient
+
+                return Oracle23aiClient()
             case _:
                 raise ValueError(f"Unsupported vector type: {vector_type}")
 

backend/open_webui/retrieval/vector/type.py

@@ -9,3 +9,5 @@ class VectorType(StrEnum):
     ELASTICSEARCH = "elasticsearch"
     OPENSEARCH = "opensearch"
     PGVECTOR = "pgvector"
+    ORACLE23AI = "oracle23ai"
+    S3VECTOR = "s3vector"

backend/open_webui/retrieval/vector/utils.py

@@ -0,0 +1,14 @@
+from datetime import datetime
+
+
+def stringify_metadata(
+    metadata: dict[str, any],
+) -> dict[str, any]:
+    for key, value in metadata.items():
+        if (
+            isinstance(value, datetime)
+            or isinstance(value, list)
+            or isinstance(value, dict)
+        ):
+            metadata[key] = str(value)
+    return metadata

backend/open_webui/retrieval/web/brave.py

@@ -36,7 +36,9 @@ def search_brave(
 
     return [
         SearchResult(
-            link=result["url"], title=result.get("title"), snippet=result.get("snippet")
+            link=result["url"],
+            title=result.get("title"),
+            snippet=result.get("description"),
         )
         for result in results[:count]
     ]

backend/open_webui/retrieval/web/duckduckgo.py

@@ -2,8 +2,8 @@ import logging
 from typing import Optional
 
 from open_webui.retrieval.web.main import SearchResult, get_filtered_results
-from duckduckgo_search import DDGS
-from duckduckgo_search.exceptions import RatelimitException
+from ddgs import DDGS
+from ddgs.exceptions import RatelimitException
 from open_webui.env import SRC_LOG_LEVELS
 
 log = logging.getLogger(__name__)

backend/open_webui/retrieval/web/perplexity.py

@@ -1,10 +1,20 @@
 import logging
-from typing import Optional, List
+from typing import Optional, Literal
 import requests
 
 from open_webui.retrieval.web.main import SearchResult, get_filtered_results
 from open_webui.env import SRC_LOG_LEVELS
 
+MODELS = Literal[
+    "sonar",
+    "sonar-pro",
+    "sonar-reasoning",
+    "sonar-reasoning-pro",
+    "sonar-deep-research",
+]
+SEARCH_CONTEXT_USAGE_LEVELS = Literal["low", "medium", "high"]
+
+
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
@@ -14,6 +24,8 @@ def search_perplexity(
     query: str,
     count: int,
     filter_list: Optional[list[str]] = None,
+    model: MODELS = "sonar",
+    search_context_usage: SEARCH_CONTEXT_USAGE_LEVELS = "medium",
 ) -> list[SearchResult]:
     """Search using Perplexity API and return the results as a list of SearchResult objects.
 
@@ -21,6 +33,9 @@ def search_perplexity(
       api_key (str): A Perplexity API key
       query (str): The query to search for
       count (int): Maximum number of results to return
+      filter_list (Optional[list[str]]): List of domains to filter results
+      model (str): The Perplexity model to use (sonar, sonar-pro)
+      search_context_usage (str): Search context usage level (low, medium, high)
 
     """
 
@@ -33,7 +48,7 @@ def search_perplexity(
 
         # Create payload for the API call
         payload = {
-            "model": "sonar",
+            "model": model,
             "messages": [
                 {
                     "role": "system",
@@ -43,6 +58,9 @@ def search_perplexity(
             ],
             "temperature": 0.2,  # Lower temperature for more factual responses
             "stream": False,
+            "web_search_options": {
+                "search_context_usage": search_context_usage,
+            },
         }
 
         headers = {

backend/open_webui/routers/audio.py

@@ -10,11 +10,12 @@ from pydub.silence import split_on_silence
 from concurrent.futures import ThreadPoolExecutor
 from typing import Optional
 
-
+from fnmatch import fnmatch
 import aiohttp
 import aiofiles
 import requests
 import mimetypes
+from urllib.parse import quote
 
 from fastapi import (
     Depends,
@@ -96,12 +97,9 @@ def is_audio_conversion_required(file_path):
             # File is AAC/mp4a audio, recommend mp3 conversion
             return True
 
-        # If the codec name or file extension is in the supported formats
-        if (
-            codec_name in SUPPORTED_FORMATS
-            or os.path.splitext(file_path)[1][1:].lower() in SUPPORTED_FORMATS
-        ):
-            return False  # Already supported
+        # If the codec name is in the supported formats
+        if codec_name in SUPPORTED_FORMATS:
+            return False
 
         return True
     except Exception as e:
@@ -171,6 +169,7 @@ class STTConfigForm(BaseModel):
     OPENAI_API_KEY: str
     ENGINE: str
     MODEL: str
+    SUPPORTED_CONTENT_TYPES: list[str] = []
     WHISPER_MODEL: str
     DEEPGRAM_API_KEY: str
     AZURE_API_KEY: str
@@ -205,6 +204,7 @@ async def get_audio_config(request: Request, user=Depends(get_admin_user)):
             "OPENAI_API_KEY": request.app.state.config.STT_OPENAI_API_KEY,
             "ENGINE": request.app.state.config.STT_ENGINE,
             "MODEL": request.app.state.config.STT_MODEL,
+            "SUPPORTED_CONTENT_TYPES": request.app.state.config.STT_SUPPORTED_CONTENT_TYPES,
             "WHISPER_MODEL": request.app.state.config.WHISPER_MODEL,
             "DEEPGRAM_API_KEY": request.app.state.config.DEEPGRAM_API_KEY,
             "AZURE_API_KEY": request.app.state.config.AUDIO_STT_AZURE_API_KEY,
@@ -239,6 +239,10 @@ async def update_audio_config(
     request.app.state.config.STT_OPENAI_API_KEY = form_data.stt.OPENAI_API_KEY
     request.app.state.config.STT_ENGINE = form_data.stt.ENGINE
     request.app.state.config.STT_MODEL = form_data.stt.MODEL
+    request.app.state.config.STT_SUPPORTED_CONTENT_TYPES = (
+        form_data.stt.SUPPORTED_CONTENT_TYPES
+    )
+
     request.app.state.config.WHISPER_MODEL = form_data.stt.WHISPER_MODEL
     request.app.state.config.DEEPGRAM_API_KEY = form_data.stt.DEEPGRAM_API_KEY
     request.app.state.config.AUDIO_STT_AZURE_API_KEY = form_data.stt.AZURE_API_KEY
@@ -253,6 +257,8 @@ async def update_audio_config(
         request.app.state.faster_whisper_model = set_faster_whisper_model(
             form_data.stt.WHISPER_MODEL, WHISPER_MODEL_AUTO_UPDATE
         )
+    else:
+        request.app.state.faster_whisper_model = None
 
     return {
         "tts": {
@@ -272,6 +278,7 @@ async def update_audio_config(
             "OPENAI_API_KEY": request.app.state.config.STT_OPENAI_API_KEY,
             "ENGINE": request.app.state.config.STT_ENGINE,
             "MODEL": request.app.state.config.STT_MODEL,
+            "SUPPORTED_CONTENT_TYPES": request.app.state.config.STT_SUPPORTED_CONTENT_TYPES,
             "WHISPER_MODEL": request.app.state.config.WHISPER_MODEL,
             "DEEPGRAM_API_KEY": request.app.state.config.DEEPGRAM_API_KEY,
             "AZURE_API_KEY": request.app.state.config.AUDIO_STT_AZURE_API_KEY,
@@ -321,6 +328,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
         log.exception(e)
         raise HTTPException(status_code=400, detail="Invalid JSON payload")
 
+    r = None
     if request.app.state.config.TTS_ENGINE == "openai":
         payload["model"] = request.app.state.config.TTS_MODEL
 
@@ -329,7 +337,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
             async with aiohttp.ClientSession(
                 timeout=timeout, trust_env=True
             ) as session:
-                async with session.post(
+                r = await session.post(
                     url=f"{request.app.state.config.TTS_OPENAI_API_BASE_URL}/audio/speech",
                     json=payload,
                     headers={
@@ -337,7 +345,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                         "Authorization": f"Bearer {request.app.state.config.TTS_OPENAI_API_KEY}",
                         **(
                             {
-                                "X-OpenWebUI-User-Name": user.name,
+                                "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                                 "X-OpenWebUI-User-Id": user.id,
                                 "X-OpenWebUI-User-Email": user.email,
                                 "X-OpenWebUI-User-Role": user.role,
@@ -347,14 +355,15 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                         ),
                     },
                     ssl=AIOHTTP_CLIENT_SESSION_SSL,
-                ) as r:
-                    r.raise_for_status()
+                )
 
-                    async with aiofiles.open(file_path, "wb") as f:
-                        await f.write(await r.read())
+                r.raise_for_status()
 
-                    async with aiofiles.open(file_body_path, "w") as f:
-                        await f.write(json.dumps(payload))
+                async with aiofiles.open(file_path, "wb") as f:
+                    await f.write(await r.read())
+
+                async with aiofiles.open(file_body_path, "w") as f:
+                    await f.write(json.dumps(payload))
 
             return FileResponse(file_path)
 
@@ -362,18 +371,22 @@ async def speech(request: Request, user=Depends(get_verified_user)):
             log.exception(e)
             detail = None
 
-            try:
-                if r.status != 200:
-                    res = await r.json()
+            status_code = 500
+            detail = f"Open WebUI: Server Connection Error"
 
+            if r is not None:
+                status_code = r.status
+
+                try:
+                    res = await r.json()
                     if "error" in res:
-                        detail = f"External: {res['error'].get('message', '')}"
-            except Exception:
-                detail = f"External: {e}"
+                        detail = f"External: {res['error']}"
+                except Exception:
+                    detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500) if r else 500,
-                detail=detail if detail else "Open WebUI: Server Connection Error",
+                status_code=status_code,
+                detail=detail,
             )
 
     elif request.app.state.config.TTS_ENGINE == "elevenlabs":
@@ -548,7 +561,11 @@ def transcription_handler(request, file_path, metadata):
             file_path,
             beam_size=5,
             vad_filter=request.app.state.config.WHISPER_VAD_FILTER,
-            language=metadata.get("language") or WHISPER_LANGUAGE,
+            language=(
+                metadata.get("language", None)
+                if WHISPER_LANGUAGE == ""
+                else WHISPER_LANGUAGE
+            ),
         )
         log.info(
             "Detected language '%s' with probability %f"
@@ -631,7 +648,7 @@ def transcription_handler(request, file_path, metadata):
 
             # Make request to Deepgram API
             r = requests.post(
-                "https://api.deepgram.com/v1/listen",
+                "https://api.deepgram.com/v1/listen?smart_format=true",
                 headers=headers,
                 params=params,
                 data=file_data,
@@ -913,10 +930,18 @@ def transcription(
 ):
     log.info(f"file.content_type: {file.content_type}")
 
-    SUPPORTED_CONTENT_TYPES = {"video/webm"}  # Extend if you add more video types!
-    if not (
-        file.content_type.startswith("audio/")
-        or file.content_type in SUPPORTED_CONTENT_TYPES
+    stt_supported_content_types = getattr(
+        request.app.state.config, "STT_SUPPORTED_CONTENT_TYPES", []
+    )
+
+    if not any(
+        fnmatch(file.content_type, content_type)
+        for content_type in (
+            stt_supported_content_types
+            if stt_supported_content_types
+            and any(t.strip() for t in stt_supported_content_types)
+            else ["audio/*", "video/webm"]
+        )
     ):
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,

backend/open_webui/routers/auths.py

@@ -55,9 +55,8 @@ from typing import Optional, List
 
 from ssl import CERT_NONE, CERT_REQUIRED, PROTOCOL_TLS
 
-if ENABLE_LDAP.value:
-    from ldap3 import Server, Connection, NONE, Tls
-    from ldap3.utils.conv import escape_filter_chars
+from ldap3 import Server, Connection, NONE, Tls
+from ldap3.utils.conv import escape_filter_chars
 
 router = APIRouter()
 
@@ -229,14 +228,30 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
         if not connection_app.bind():
             raise HTTPException(400, detail="Application account bind failed")
 
+        ENABLE_LDAP_GROUP_MANAGEMENT = (
+            request.app.state.config.ENABLE_LDAP_GROUP_MANAGEMENT
+        )
+        ENABLE_LDAP_GROUP_CREATION = request.app.state.config.ENABLE_LDAP_GROUP_CREATION
+        LDAP_ATTRIBUTE_FOR_GROUPS = request.app.state.config.LDAP_ATTRIBUTE_FOR_GROUPS
+
+        search_attributes = [
+            f"{LDAP_ATTRIBUTE_FOR_USERNAME}",
+            f"{LDAP_ATTRIBUTE_FOR_MAIL}",
+            "cn",
+        ]
+
+        if ENABLE_LDAP_GROUP_MANAGEMENT:
+            search_attributes.append(f"{LDAP_ATTRIBUTE_FOR_GROUPS}")
+            log.info(
+                f"LDAP Group Management enabled. Adding {LDAP_ATTRIBUTE_FOR_GROUPS} to search attributes"
+            )
+
+        log.info(f"LDAP search attributes: {search_attributes}")
+
         search_success = connection_app.search(
             search_base=LDAP_SEARCH_BASE,
             search_filter=f"(&({LDAP_ATTRIBUTE_FOR_USERNAME}={escape_filter_chars(form_data.user.lower())}){LDAP_SEARCH_FILTERS})",
-            attributes=[
-                f"{LDAP_ATTRIBUTE_FOR_USERNAME}",
-                f"{LDAP_ATTRIBUTE_FOR_MAIL}",
-                "cn",
-            ],
+            attributes=search_attributes,
         )
 
         if not search_success or not connection_app.entries:
@@ -259,6 +274,69 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
         cn = str(entry["cn"])
         user_dn = entry.entry_dn
 
+        user_groups = []
+        if ENABLE_LDAP_GROUP_MANAGEMENT and LDAP_ATTRIBUTE_FOR_GROUPS in entry:
+            group_dns = entry[LDAP_ATTRIBUTE_FOR_GROUPS]
+            log.info(f"LDAP raw group DNs for user {username}: {group_dns}")
+
+            if group_dns:
+                log.info(f"LDAP group_dns original: {group_dns}")
+                log.info(f"LDAP group_dns type: {type(group_dns)}")
+                log.info(f"LDAP group_dns length: {len(group_dns)}")
+
+                if hasattr(group_dns, "value"):
+                    group_dns = group_dns.value
+                    log.info(f"Extracted .value property: {group_dns}")
+                elif hasattr(group_dns, "__iter__") and not isinstance(
+                    group_dns, (str, bytes)
+                ):
+                    group_dns = list(group_dns)
+                    log.info(f"Converted to list: {group_dns}")
+
+                if isinstance(group_dns, list):
+                    group_dns = [str(item) for item in group_dns]
+                else:
+                    group_dns = [str(group_dns)]
+
+                log.info(
+                    f"LDAP group_dns after processing - type: {type(group_dns)}, length: {len(group_dns)}"
+                )
+
+                for group_idx, group_dn in enumerate(group_dns):
+                    group_dn = str(group_dn)
+                    log.info(f"Processing group DN #{group_idx + 1}: {group_dn}")
+
+                    try:
+                        group_cn = None
+
+                        for item in group_dn.split(","):
+                            item = item.strip()
+                            if item.upper().startswith("CN="):
+                                group_cn = item[3:]
+                                break
+
+                        if group_cn:
+                            user_groups.append(group_cn)
+
+                        else:
+                            log.warning(
+                                f"Could not extract CN from group DN: {group_dn}"
+                            )
+                    except Exception as e:
+                        log.warning(
+                            f"Failed to extract group name from DN {group_dn}: {e}"
+                        )
+
+                log.info(
+                    f"LDAP groups for user {username}: {user_groups} (total: {len(user_groups)})"
+                )
+            else:
+                log.info(f"No groups found for user {username}")
+        elif ENABLE_LDAP_GROUP_MANAGEMENT:
+            log.warning(
+                f"LDAP Group Management enabled but {LDAP_ATTRIBUTE_FOR_GROUPS} attribute not found in user entry"
+            )
+
         if username == form_data.user.lower():
             connection_user = Connection(
                 server,
@@ -273,11 +351,9 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
             user = Users.get_user_by_email(email)
             if not user:
                 try:
-                    user_count = Users.get_num_users()
-
                     role = (
                         "admin"
-                        if user_count == 0
+                        if not Users.has_users()
                         else request.app.state.config.DEFAULT_USER_ROLE
                     )
 
@@ -334,6 +410,22 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                     user.id, request.app.state.config.USER_PERMISSIONS
                 )
 
+                if (
+                    user.role != "admin"
+                    and ENABLE_LDAP_GROUP_MANAGEMENT
+                    and user_groups
+                ):
+                    if ENABLE_LDAP_GROUP_CREATION:
+                        Groups.create_groups_by_group_names(user.id, user_groups)
+
+                    try:
+                        Groups.sync_groups_by_group_names(user.id, user_groups)
+                        log.info(
+                            f"Successfully synced groups for user {user.id}: {user_groups}"
+                        )
+                    except Exception as e:
+                        log.error(f"Failed to sync groups for user {user.id}: {e}")
+
                 return {
                     "token": token,
                     "token_type": "Bearer",
@@ -386,7 +478,7 @@ async def signin(request: Request, response: Response, form_data: SigninForm):
             group_names = [name.strip() for name in group_names if name.strip()]
 
             if group_names:
-                Groups.sync_user_groups_by_group_names(user.id, group_names)
+                Groups.sync_groups_by_group_names(user.id, group_names)
 
     elif WEBUI_AUTH == False:
         admin_email = "admin@localhost"
@@ -395,7 +487,7 @@ async def signin(request: Request, response: Response, form_data: SigninForm):
         if Users.get_user_by_email(admin_email.lower()):
             user = Auths.authenticate_user(admin_email.lower(), admin_password)
         else:
-            if Users.get_num_users() != 0:
+            if Users.has_users():
                 raise HTTPException(400, detail=ERROR_MESSAGES.EXISTING_USERS)
 
             await signup(
@@ -462,6 +554,7 @@ async def signin(request: Request, response: Response, form_data: SigninForm):
 
 @router.post("/signup", response_model=SessionUserResponse)
 async def signup(request: Request, response: Response, form_data: SignupForm):
+    has_users = Users.has_users()
 
     if WEBUI_AUTH:
         if (
@@ -472,12 +565,11 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
                 status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.ACCESS_PROHIBITED
             )
     else:
-        if Users.get_num_users() != 0:
+        if has_users:
             raise HTTPException(
                 status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.ACCESS_PROHIBITED
             )
 
-    user_count = Users.get_num_users()
     if not validate_email_format(form_data.email.lower()):
         raise HTTPException(
             status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.INVALID_EMAIL_FORMAT
@@ -487,9 +579,7 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
         raise HTTPException(400, detail=ERROR_MESSAGES.EMAIL_TAKEN)
 
     try:
-        role = (
-            "admin" if user_count == 0 else request.app.state.config.DEFAULT_USER_ROLE
-        )
+        role = "admin" if not has_users else request.app.state.config.DEFAULT_USER_ROLE
 
         # The password passed to bcrypt must be 72 bytes or fewer. If it is longer, it will be truncated before hashing.
         if len(form_data.password.encode("utf-8")) > 72:
@@ -550,7 +640,7 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
                 user.id, request.app.state.config.USER_PERMISSIONS
             )
 
-            if user_count == 0:
+            if not has_users:
                 # Disable signup after the first user is created
                 request.app.state.config.ENABLE_SIGNUP = False
 
@@ -575,12 +665,13 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
 @router.get("/signout")
 async def signout(request: Request, response: Response):
     response.delete_cookie("token")
+    response.delete_cookie("oui-session")
 
     if ENABLE_OAUTH_SIGNUP.value:
         oauth_id_token = request.cookies.get("oauth_id_token")
-        if oauth_id_token:
+        if oauth_id_token and OPENID_PROVIDER_URL.value:
             try:
-                async with ClientSession() as session:
+                async with ClientSession(trust_env=True) as session:
                     async with session.get(OPENID_PROVIDER_URL.value) as resp:
                         if resp.status == 200:
                             openid_data = await resp.json()
@@ -592,7 +683,12 @@ async def signout(request: Request, response: Response):
                                     status_code=200,
                                     content={
                                         "status": True,
-                                        "redirect_url": f"{logout_url}?id_token_hint={oauth_id_token}",
+                                        "redirect_url": f"{logout_url}?id_token_hint={oauth_id_token}"
+                                        + (
+                                            f"&post_logout_redirect_uri={WEBUI_AUTH_SIGNOUT_REDIRECT_URL}"
+                                            if WEBUI_AUTH_SIGNOUT_REDIRECT_URL
+                                            else ""
+                                        ),
                                     },
                                     headers=response.headers,
                                 )

backend/open_webui/routers/channels.py

@@ -40,10 +40,14 @@ router = APIRouter()
 
 @router.get("/", response_model=list[ChannelModel])
 async def get_channels(user=Depends(get_verified_user)):
+    return Channels.get_channels_by_user_id(user.id)
+
+
+@router.get("/list", response_model=list[ChannelModel])
+async def get_all_channels(user=Depends(get_verified_user)):
     if user.role == "admin":
         return Channels.get_channels()
-    else:
-        return Channels.get_channels_by_user_id(user.id)
+    return Channels.get_channels_by_user_id(user.id)
 
 
 ############################
@@ -430,13 +434,6 @@ async def update_message_by_id(
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if user.role != "admin" and not has_access(
-        user.id, type="read", access_control=channel.access_control
-    ):
-        raise HTTPException(
-            status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
-        )
-
     message = Messages.get_message_by_id(message_id)
     if not message:
         raise HTTPException(
@@ -448,6 +445,15 @@ async def update_message_by_id(
             status_code=status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.DEFAULT()
         )
 
+    if (
+        user.role != "admin"
+        and message.user_id != user.id
+        and not has_access(user.id, type="read", access_control=channel.access_control)
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
+        )
+
     try:
         message = Messages.update_message_by_id(message_id, form_data)
         message = Messages.get_message_by_id(message_id)
@@ -637,13 +643,6 @@ async def delete_message_by_id(
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if user.role != "admin" and not has_access(
-        user.id, type="read", access_control=channel.access_control
-    ):
-        raise HTTPException(
-            status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
-        )
-
     message = Messages.get_message_by_id(message_id)
     if not message:
         raise HTTPException(
@@ -655,6 +654,15 @@ async def delete_message_by_id(
             status_code=status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.DEFAULT()
         )
 
+    if (
+        user.role != "admin"
+        and message.user_id != user.id
+        and not has_access(user.id, type="read", access_control=channel.access_control)
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
+        )
+
     try:
         Messages.delete_message_by_id(message_id)
         await sio.emit(

backend/open_webui/routers/chats.py

@@ -39,13 +39,21 @@ router = APIRouter()
 async def get_session_user_chat_list(
     user=Depends(get_verified_user), page: Optional[int] = None
 ):
-    if page is not None:
-        limit = 60
-        skip = (page - 1) * limit
+    try:
+        if page is not None:
+            limit = 60
+            skip = (page - 1) * limit
 
-        return Chats.get_chat_title_id_list_by_user_id(user.id, skip=skip, limit=limit)
-    else:
-        return Chats.get_chat_title_id_list_by_user_id(user.id)
+            return Chats.get_chat_title_id_list_by_user_id(
+                user.id, skip=skip, limit=limit
+            )
+        else:
+            return Chats.get_chat_title_id_list_by_user_id(user.id)
+    except Exception as e:
+        log.exception(e)
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.DEFAULT()
+        )
 
 
 ############################
@@ -609,7 +617,18 @@ async def clone_chat_by_id(
             "title": form_data.title if form_data.title else f"Clone of {chat.title}",
         }
 
-        chat = Chats.insert_new_chat(user.id, ChatForm(**{"chat": updated_chat}))
+        chat = Chats.import_chat(
+            user.id,
+            ChatImportForm(
+                **{
+                    "chat": updated_chat,
+                    "meta": chat.meta,
+                    "pinned": chat.pinned,
+                    "folder_id": chat.folder_id,
+                }
+            ),
+        )
+
         return ChatResponse(**chat.model_dump())
     else:
         raise HTTPException(
@@ -638,7 +657,17 @@ async def clone_shared_chat_by_id(id: str, user=Depends(get_verified_user)):
             "title": f"Clone of {chat.title}",
         }
 
-        chat = Chats.insert_new_chat(user.id, ChatForm(**{"chat": updated_chat}))
+        chat = Chats.import_chat(
+            user.id,
+            ChatImportForm(
+                **{
+                    "chat": updated_chat,
+                    "meta": chat.meta,
+                    "pinned": chat.pinned,
+                    "folder_id": chat.folder_id,
+                }
+            ),
+        )
         return ChatResponse(**chat.model_dump())
     else:
         raise HTTPException(
@@ -684,8 +713,10 @@ async def archive_chat_by_id(id: str, user=Depends(get_verified_user)):
 
 @router.post("/{id}/share", response_model=Optional[ChatResponse])
 async def share_chat_by_id(request: Request, id: str, user=Depends(get_verified_user)):
-    if not has_permission(
-        user.id, "chat.share", request.app.state.config.USER_PERMISSIONS
+    if (user.role != "admin") and (
+        not has_permission(
+            user.id, "chat.share", request.app.state.config.USER_PERMISSIONS
+        )
     ):
         raise HTTPException(
             status_code=status.HTTP_401_UNAUTHORIZED,

backend/open_webui/routers/configs.py

@@ -7,7 +7,11 @@ from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.config import get_config, save_config
 from open_webui.config import BannerModel
 
-from open_webui.utils.tools import get_tool_server_data, get_tool_servers_data
+from open_webui.utils.tools import (
+    get_tool_server_data,
+    get_tool_servers_data,
+    get_tool_server_url,
+)
 
 
 router = APIRouter()
@@ -39,32 +43,39 @@ async def export_config(user=Depends(get_admin_user)):
 
 
 ############################
-# Direct Connections Config
+# Connections Config
 ############################
 
 
-class DirectConnectionsConfigForm(BaseModel):
+class ConnectionsConfigForm(BaseModel):
     ENABLE_DIRECT_CONNECTIONS: bool
+    ENABLE_BASE_MODELS_CACHE: bool
 
 
-@router.get("/direct_connections", response_model=DirectConnectionsConfigForm)
-async def get_direct_connections_config(request: Request, user=Depends(get_admin_user)):
+@router.get("/connections", response_model=ConnectionsConfigForm)
+async def get_connections_config(request: Request, user=Depends(get_admin_user)):
     return {
         "ENABLE_DIRECT_CONNECTIONS": request.app.state.config.ENABLE_DIRECT_CONNECTIONS,
+        "ENABLE_BASE_MODELS_CACHE": request.app.state.config.ENABLE_BASE_MODELS_CACHE,
     }
 
 
-@router.post("/direct_connections", response_model=DirectConnectionsConfigForm)
-async def set_direct_connections_config(
+@router.post("/connections", response_model=ConnectionsConfigForm)
+async def set_connections_config(
     request: Request,
-    form_data: DirectConnectionsConfigForm,
+    form_data: ConnectionsConfigForm,
     user=Depends(get_admin_user),
 ):
     request.app.state.config.ENABLE_DIRECT_CONNECTIONS = (
         form_data.ENABLE_DIRECT_CONNECTIONS
     )
+    request.app.state.config.ENABLE_BASE_MODELS_CACHE = (
+        form_data.ENABLE_BASE_MODELS_CACHE
+    )
+
     return {
         "ENABLE_DIRECT_CONNECTIONS": request.app.state.config.ENABLE_DIRECT_CONNECTIONS,
+        "ENABLE_BASE_MODELS_CACHE": request.app.state.config.ENABLE_BASE_MODELS_CACHE,
     }
 
 
@@ -128,7 +139,7 @@ async def verify_tool_servers_config(
         elif form_data.auth_type == "session":
             token = request.state.token.credentials
 
-        url = f"{form_data.url}/{form_data.path}"
+        url = get_tool_server_url(form_data.url, form_data.path)
         return await get_tool_server_data(token, url)
     except Exception as e:
         raise HTTPException(

backend/open_webui/routers/evaluations.py

@@ -129,7 +129,10 @@ async def create_feedback(
 
 @router.get("/feedback/{id}", response_model=FeedbackModel)
 async def get_feedback_by_id(id: str, user=Depends(get_verified_user)):
-    feedback = Feedbacks.get_feedback_by_id_and_user_id(id=id, user_id=user.id)
+    if user.role == "admin":
+        feedback = Feedbacks.get_feedback_by_id(id=id)
+    else:
+        feedback = Feedbacks.get_feedback_by_id_and_user_id(id=id, user_id=user.id)
 
     if not feedback:
         raise HTTPException(
@@ -143,9 +146,12 @@ async def get_feedback_by_id(id: str, user=Depends(get_verified_user)):
 async def update_feedback_by_id(
     id: str, form_data: FeedbackForm, user=Depends(get_verified_user)
 ):
-    feedback = Feedbacks.update_feedback_by_id_and_user_id(
-        id=id, user_id=user.id, form_data=form_data
-    )
+    if user.role == "admin":
+        feedback = Feedbacks.update_feedback_by_id(id=id, form_data=form_data)
+    else:
+        feedback = Feedbacks.update_feedback_by_id_and_user_id(
+            id=id, user_id=user.id, form_data=form_data
+        )
 
     if not feedback:
         raise HTTPException(

backend/open_webui/routers/files.py

@@ -21,6 +21,7 @@ from fastapi import (
 from fastapi.responses import FileResponse, StreamingResponse
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import SRC_LOG_LEVELS
+from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT
 
 from open_webui.models.users import Users
 from open_webui.models.files import (
@@ -155,9 +156,19 @@ def upload_file(
         if process:
             try:
                 if file.content_type:
-                    if file.content_type.startswith("audio/") or file.content_type in {
-                        "video/webm"
-                    }:
+                    stt_supported_content_types = getattr(
+                        request.app.state.config, "STT_SUPPORTED_CONTENT_TYPES", []
+                    )
+
+                    if any(
+                        fnmatch(file.content_type, content_type)
+                        for content_type in (
+                            stt_supported_content_types
+                            if stt_supported_content_types
+                            and any(t.strip() for t in stt_supported_content_types)
+                            else ["audio/*", "video/webm"]
+                        )
+                    ):
                         file_path = Storage.get_file(file_path)
                         result = transcribe(request, file_path, file_metadata)
 
@@ -276,6 +287,7 @@ async def delete_all_files(user=Depends(get_admin_user)):
     if result:
         try:
             Storage.delete_all_files()
+            VECTOR_DB_CLIENT.reset()
         except Exception as e:
             log.exception(e)
             log.error("Error deleting files")
@@ -593,12 +605,12 @@ async def delete_file_by_id(id: str, user=Depends(get_verified_user)):
         or user.role == "admin"
         or has_access_to_file(id, "write", user)
     ):
-        # We should add Chroma cleanup here
 
         result = Files.delete_file_by_id(id)
         if result:
             try:
                 Storage.delete_file(file.path)
+                VECTOR_DB_CLIENT.delete(collection_name=f"file-{id}")
             except Exception as e:
                 log.exception(e)
                 log.error("Error deleting files")

backend/open_webui/routers/folders.py

@@ -49,7 +49,7 @@ async def get_folders(user=Depends(get_verified_user)):
             **folder.model_dump(),
             "items": {
                 "chats": [
-                    {"title": chat.title, "id": chat.id}
+                    {"title": chat.title, "id": chat.id, "updated_at": chat.updated_at}
                     for chat in Chats.get_chats_by_folder_id_and_user_id(
                         folder.id, user.id
                     )
@@ -78,7 +78,7 @@ def create_folder(form_data: FolderForm, user=Depends(get_verified_user)):
         )
 
     try:
-        folder = Folders.insert_new_folder(user.id, form_data.name)
+        folder = Folders.insert_new_folder(user.id, form_data)
         return folder
     except Exception as e:
         log.exception(e)
@@ -120,16 +120,14 @@ async def update_folder_name_by_id(
         existing_folder = Folders.get_folder_by_parent_id_and_user_id_and_name(
             folder.parent_id, user.id, form_data.name
         )
-        if existing_folder:
+        if existing_folder and existing_folder.id != id:
             raise HTTPException(
                 status_code=status.HTTP_400_BAD_REQUEST,
                 detail=ERROR_MESSAGES.DEFAULT("Folder already exists"),
             )
 
         try:
-            folder = Folders.update_folder_name_by_id_and_user_id(
-                id, user.id, form_data.name
-            )
+            folder = Folders.update_folder_by_id_and_user_id(id, user.id, form_data)
 
             return folder
         except Exception as e:
@@ -246,11 +244,11 @@ async def delete_folder_by_id(
     folder = Folders.get_folder_by_id_and_user_id(id, user.id)
     if folder:
         try:
-            result = Folders.delete_folder_by_id_and_user_id(id, user.id)
-            if result:
-                return result
-            else:
-                raise Exception("Error deleting folder")
+            folder_ids = Folders.delete_folder_by_id_and_user_id(id, user.id)
+            for folder_id in folder_ids:
+                Chats.delete_chats_by_user_id_and_folder_id(user.id, folder_id)
+
+            return True
         except Exception as e:
             log.exception(e)
             log.error(f"Error deleting folder: {id}")

backend/open_webui/routers/functions.py

@@ -105,7 +105,7 @@ async def load_function_from_url(
     )
 
     try:
-        async with aiohttp.ClientSession() as session:
+        async with aiohttp.ClientSession(trust_env=True) as session:
             async with session.get(
                 url, headers={"Content-Type": "application/json"}
             ) as resp:
@@ -131,15 +131,29 @@ async def load_function_from_url(
 ############################
 
 
-class SyncFunctionsForm(FunctionForm):
+class SyncFunctionsForm(BaseModel):
     functions: list[FunctionModel] = []
 
 
-@router.post("/sync", response_model=Optional[FunctionModel])
+@router.post("/sync", response_model=list[FunctionModel])
 async def sync_functions(
     request: Request, form_data: SyncFunctionsForm, user=Depends(get_admin_user)
 ):
-    return Functions.sync_functions(user.id, form_data.functions)
+    try:
+        for function in form_data.functions:
+            function.content = replace_imports(function.content)
+            function_module, function_type, frontmatter = load_function_module_by_id(
+                function.id,
+                content=function.content,
+            )
+
+        return Functions.sync_functions(user.id, form_data.functions)
+    except Exception as e:
+        log.exception(f"Failed to load a function: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.DEFAULT(e),
+        )
 
 
 ############################

backend/open_webui/routers/groups.py

@@ -9,6 +9,7 @@ from open_webui.models.groups import (
     GroupForm,
     GroupUpdateForm,
     GroupResponse,
+    UserIdsForm,
 )
 
 from open_webui.config import CACHE_DIR
@@ -107,6 +108,56 @@ async def update_group_by_id(
         )
 
 
+############################
+# AddUserToGroupByUserIdAndGroupId
+############################
+
+
+@router.post("/id/{id}/users/add", response_model=Optional[GroupResponse])
+async def add_user_to_group(
+    id: str, form_data: UserIdsForm, user=Depends(get_admin_user)
+):
+    try:
+        if form_data.user_ids:
+            form_data.user_ids = Users.get_valid_user_ids(form_data.user_ids)
+
+        group = Groups.add_users_to_group(id, form_data.user_ids)
+        if group:
+            return group
+        else:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail=ERROR_MESSAGES.DEFAULT("Error adding users to group"),
+            )
+    except Exception as e:
+        log.exception(f"Error adding users to group {id}: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.DEFAULT(e),
+        )
+
+
+@router.post("/id/{id}/users/remove", response_model=Optional[GroupResponse])
+async def remove_users_from_group(
+    id: str, form_data: UserIdsForm, user=Depends(get_admin_user)
+):
+    try:
+        group = Groups.remove_users_from_group(id, form_data.user_ids)
+        if group:
+            return group
+        else:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail=ERROR_MESSAGES.DEFAULT("Error removing users from group"),
+            )
+    except Exception as e:
+        log.exception(f"Error removing users from group {id}: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.DEFAULT(e),
+        )
+
+
 ############################
 # DeleteGroupById
 ############################

backend/open_webui/routers/images.py

@@ -8,6 +8,7 @@ import re
 from pathlib import Path
 from typing import Optional
 
+from urllib.parse import quote
 import requests
 from fastapi import APIRouter, Depends, HTTPException, Request, UploadFile
 from open_webui.config import CACHE_DIR
@@ -302,8 +303,16 @@ async def update_image_config(
 ):
     set_image_model(request, form_data.MODEL)
 
+    if form_data.IMAGE_SIZE == "auto" and form_data.MODEL != "gpt-image-1":
+        raise HTTPException(
+            status_code=400,
+            detail=ERROR_MESSAGES.INCORRECT_FORMAT(
+                "  (auto is only allowed with gpt-image-1)."
+            ),
+        )
+
     pattern = r"^\d+x\d+$"
-    if re.match(pattern, form_data.IMAGE_SIZE):
+    if form_data.IMAGE_SIZE == "auto" or re.match(pattern, form_data.IMAGE_SIZE):
         request.app.state.config.IMAGE_SIZE = form_data.IMAGE_SIZE
     else:
         raise HTTPException(
@@ -420,7 +429,7 @@ def load_b64_image_data(b64_str):
     try:
         if "," in b64_str:
             header, encoded = b64_str.split(",", 1)
-            mime_type = header.split(";")[0]
+            mime_type = header.split(";")[0].lstrip("data:")
             img_data = base64.b64decode(encoded)
         else:
             mime_type = "image/png"
@@ -428,7 +437,7 @@ def load_b64_image_data(b64_str):
         return img_data, mime_type
     except Exception as e:
         log.exception(f"Error loading image data: {e}")
-        return None
+        return None, None
 
 
 def load_url_image_data(url, headers=None):
@@ -471,7 +480,14 @@ async def image_generations(
     form_data: GenerateImageForm,
     user=Depends(get_verified_user),
 ):
-    width, height = tuple(map(int, request.app.state.config.IMAGE_SIZE.split("x")))
+    # if IMAGE_SIZE = 'auto', default WidthxHeight to the 512x512 default
+    # This is only relevant when the user has set IMAGE_SIZE to 'auto' with an
+    # image model other than gpt-image-1, which is warned about on settings save
+    width, height = (
+        tuple(map(int, request.app.state.config.IMAGE_SIZE.split("x")))
+        if "x" in request.app.state.config.IMAGE_SIZE
+        else (512, 512)
+    )
 
     r = None
     try:
@@ -483,7 +499,7 @@ async def image_generations(
             headers["Content-Type"] = "application/json"
 
             if ENABLE_FORWARD_USER_INFO_HEADERS:
-                headers["X-OpenWebUI-User-Name"] = user.name
+                headers["X-OpenWebUI-User-Name"] = quote(user.name, safe=" ")
                 headers["X-OpenWebUI-User-Id"] = user.id
                 headers["X-OpenWebUI-User-Email"] = user.email
                 headers["X-OpenWebUI-User-Role"] = user.role

backend/open_webui/routers/knowledge.py

@@ -25,6 +25,7 @@ from open_webui.utils.access_control import has_access, has_permission
 
 
 from open_webui.env import SRC_LOG_LEVELS
+from open_webui.config import ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS
 from open_webui.models.models import Models, ModelForm
 
 
@@ -42,7 +43,7 @@ router = APIRouter()
 async def get_knowledge(user=Depends(get_verified_user)):
     knowledge_bases = []
 
-    if user.role == "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
         knowledge_bases = Knowledges.get_knowledge_bases()
     else:
         knowledge_bases = Knowledges.get_knowledge_bases_by_user_id(user.id, "read")
@@ -90,7 +91,7 @@ async def get_knowledge(user=Depends(get_verified_user)):
 async def get_knowledge_list(user=Depends(get_verified_user)):
     knowledge_bases = []
 
-    if user.role == "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
         knowledge_bases = Knowledges.get_knowledge_bases()
     else:
         knowledge_bases = Knowledges.get_knowledge_bases_by_user_id(user.id, "write")

backend/open_webui/routers/memories.py

@@ -82,6 +82,10 @@ class QueryMemoryForm(BaseModel):
 async def query_memory(
     request: Request, form_data: QueryMemoryForm, user=Depends(get_verified_user)
 ):
+    memories = Memories.get_memories_by_user_id(user.id)
+    if not memories:
+        raise HTTPException(status_code=404, detail="No memories found for user")
+
     results = VECTOR_DB_CLIENT.search(
         collection_name=f"user-memory-{user.id}",
         vectors=[request.app.state.EMBEDDING_FUNCTION(form_data.content, user=user)],

backend/open_webui/routers/models.py

@@ -7,13 +7,15 @@ from open_webui.models.models import (
     ModelUserResponse,
     Models,
 )
+
+from pydantic import BaseModel
 from open_webui.constants import ERROR_MESSAGES
 from fastapi import APIRouter, Depends, HTTPException, Request, status
 
 
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.access_control import has_access, has_permission
-
+from open_webui.config import ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS
 
 router = APIRouter()
 
@@ -25,7 +27,7 @@ router = APIRouter()
 
 @router.get("/", response_model=list[ModelUserResponse])
 async def get_models(id: Optional[str] = None, user=Depends(get_verified_user)):
-    if user.role == "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
         return Models.get_models()
     else:
         return Models.get_models_by_user_id(user.id)
@@ -78,6 +80,32 @@ async def create_new_model(
             )
 
 
+############################
+# ExportModels
+############################
+
+
+@router.get("/export", response_model=list[ModelModel])
+async def export_models(user=Depends(get_admin_user)):
+    return Models.get_models()
+
+
+############################
+# SyncModels
+############################
+
+
+class SyncModelsForm(BaseModel):
+    models: list[ModelModel] = []
+
+
+@router.post("/sync", response_model=list[ModelModel])
+async def sync_models(
+    request: Request, form_data: SyncModelsForm, user=Depends(get_admin_user)
+):
+    return Models.sync_models(user.id, form_data.models)
+
+
 ###########################
 # GetModelById
 ###########################
@@ -102,7 +130,7 @@ async def get_model_by_id(id: str, user=Depends(get_verified_user)):
 
 
 ############################
-# ToggelModelById
+# ToggleModelById
 ############################
 
 

backend/open_webui/routers/notes.py

@@ -6,6 +6,9 @@ from typing import Optional
 from fastapi import APIRouter, Depends, HTTPException, Request, status, BackgroundTasks
 from pydantic import BaseModel
 
+from open_webui.socket.main import sio
+
+
 from open_webui.models.users import Users, UserResponse
 from open_webui.models.notes import Notes, NoteModel, NoteForm, NoteUserResponse
 
@@ -51,7 +54,14 @@ async def get_notes(request: Request, user=Depends(get_verified_user)):
     return notes
 
 
-@router.get("/list", response_model=list[NoteUserResponse])
+class NoteTitleIdResponse(BaseModel):
+    id: str
+    title: str
+    updated_at: int
+    created_at: int
+
+
+@router.get("/list", response_model=list[NoteTitleIdResponse])
 async def get_note_list(request: Request, user=Depends(get_verified_user)):
 
     if user.role != "admin" and not has_permission(
@@ -63,13 +73,8 @@ async def get_note_list(request: Request, user=Depends(get_verified_user)):
         )
 
     notes = [
-        NoteUserResponse(
-            **{
-                **note.model_dump(),
-                "user": UserResponse(**Users.get_user_by_id(note.user_id).model_dump()),
-            }
-        )
-        for note in Notes.get_notes_by_user_id(user.id, "read")
+        NoteTitleIdResponse(**note.model_dump())
+        for note in Notes.get_notes_by_user_id(user.id, "write")
     ]
 
     return notes
@@ -124,10 +129,9 @@ async def get_note_by_id(request: Request, id: str, user=Depends(get_verified_us
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if (
-        user.role != "admin"
-        and user.id != note.user_id
-        and not has_access(user.id, type="read", access_control=note.access_control)
+    if user.role != "admin" and (
+        user.id != note.user_id
+        and (not has_access(user.id, type="read", access_control=note.access_control))
     ):
         raise HTTPException(
             status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
@@ -159,9 +163,8 @@ async def update_note_by_id(
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if (
-        user.role != "admin"
-        and user.id != note.user_id
+    if user.role != "admin" and (
+        user.id != note.user_id
         and not has_access(user.id, type="write", access_control=note.access_control)
     ):
         raise HTTPException(
@@ -170,6 +173,12 @@ async def update_note_by_id(
 
     try:
         note = Notes.update_note_by_id(id, form_data)
+        await sio.emit(
+            "note-events",
+            note.model_dump(),
+            to=f"note:{note.id}",
+        )
+
         return note
     except Exception as e:
         log.exception(e)
@@ -199,9 +208,8 @@ async def delete_note_by_id(request: Request, id: str, user=Depends(get_verified
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if (
-        user.role != "admin"
-        and user.id != note.user_id
+    if user.role != "admin" and (
+        user.id != note.user_id
         and not has_access(user.id, type="write", access_control=note.access_control)
     ):
         raise HTTPException(

backend/open_webui/routers/ollama.py

@@ -16,6 +16,9 @@ from urllib.parse import urlparse
 import aiohttp
 from aiocache import cached
 import requests
+from urllib.parse import quote
+
+from open_webui.models.chats import Chats
 from open_webui.models.users import UserModel
 
 from open_webui.env import (
@@ -56,6 +59,7 @@ from open_webui.config import (
 from open_webui.env import (
     ENV,
     SRC_LOG_LEVELS,
+    MODELS_CACHE_TTL,
     AIOHTTP_CLIENT_SESSION_SSL,
     AIOHTTP_CLIENT_TIMEOUT,
     AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST,
@@ -85,7 +89,7 @@ async def send_get_request(url, key=None, user: UserModel = None):
                     **({"Authorization": f"Bearer {key}"} if key else {}),
                     **(
                         {
-                            "X-OpenWebUI-User-Name": user.name,
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                             "X-OpenWebUI-User-Id": user.id,
                             "X-OpenWebUI-User-Email": user.email,
                             "X-OpenWebUI-User-Role": user.role,
@@ -120,6 +124,7 @@ async def send_post_request(
     key: Optional[str] = None,
     content_type: Optional[str] = None,
     user: UserModel = None,
+    metadata: Optional[dict] = None,
 ):
 
     r = None
@@ -136,10 +141,15 @@ async def send_post_request(
                 **({"Authorization": f"Bearer {key}"} if key else {}),
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
+                        **(
+                            {"X-OpenWebUI-Chat-Id": metadata.get("chat_id")}
+                            if metadata and metadata.get("chat_id")
+                            else {}
+                        ),
                     }
                     if ENABLE_FORWARD_USER_INFO_HEADERS and user
                     else {}
@@ -147,8 +157,23 @@ async def send_post_request(
             },
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
         )
-        r.raise_for_status()
 
+        if r.ok is False:
+            try:
+                res = await r.json()
+                await cleanup_response(r, session)
+                if "error" in res:
+                    raise HTTPException(status_code=r.status, detail=res["error"])
+            except HTTPException as e:
+                raise e  # Re-raise HTTPException to be handled by FastAPI
+            except Exception as e:
+                log.error(f"Failed to parse error response: {e}")
+                raise HTTPException(
+                    status_code=r.status,
+                    detail=f"Open WebUI: Server Connection Error",
+                )
+
+        r.raise_for_status()  # Raises an error for bad responses (4xx, 5xx)
         if stream:
             response_headers = dict(r.headers)
 
@@ -165,24 +190,20 @@ async def send_post_request(
             )
         else:
             res = await r.json()
-            await cleanup_response(r, session)
             return res
 
+    except HTTPException as e:
+        raise e  # Re-raise HTTPException to be handled by FastAPI
     except Exception as e:
-        detail = None
-
-        if r is not None:
-            try:
-                res = await r.json()
-                if "error" in res:
-                    detail = f"Ollama: {res.get('error', 'Unknown error')}"
-            except Exception:
-                detail = f"Ollama: {e}"
+        detail = f"Ollama: {e}"
 
         raise HTTPException(
             status_code=r.status if r else 500,
-            detail=detail if detail else "Open WebUI: Server Connection Error",
+            detail=detail if e else "Open WebUI: Server Connection Error",
         )
+    finally:
+        if not stream:
+            await cleanup_response(r, session)
 
 
 def get_api_key(idx, url, configs):
@@ -231,7 +252,7 @@ async def verify_connection(
                     **({"Authorization": f"Bearer {key}"} if key else {}),
                     **(
                         {
-                            "X-OpenWebUI-User-Name": user.name,
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                             "X-OpenWebUI-User-Id": user.id,
                             "X-OpenWebUI-User-Email": user.email,
                             "X-OpenWebUI-User-Role": user.role,
@@ -318,7 +339,7 @@ def merge_ollama_models_lists(model_lists):
     return list(merged_models.values())
 
 
-@cached(ttl=1)
+@cached(ttl=MODELS_CACHE_TTL)
 async def get_all_models(request: Request, user: UserModel = None):
     log.info("get_all_models()")
     if request.app.state.config.ENABLE_OLLAMA_API:
@@ -451,7 +472,7 @@ async def get_ollama_tags(
                     **({"Authorization": f"Bearer {key}"} if key else {}),
                     **(
                         {
-                            "X-OpenWebUI-User-Name": user.name,
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                             "X-OpenWebUI-User-Id": user.id,
                             "X-OpenWebUI-User-Email": user.email,
                             "X-OpenWebUI-User-Role": user.role,
@@ -623,7 +644,10 @@ async def get_ollama_versions(request: Request, url_idx: Optional[int] = None):
 
 
 class ModelNameForm(BaseModel):
-    name: str
+    model: Optional[str] = None
+    model_config = ConfigDict(
+        extra="allow",
+    )
 
 
 @router.post("/api/unload")
@@ -632,10 +656,12 @@ async def unload_model(
     form_data: ModelNameForm,
     user=Depends(get_admin_user),
 ):
-    model_name = form_data.name
+    form_data = form_data.model_dump(exclude_none=True)
+    model_name = form_data.get("model", form_data.get("name"))
+
     if not model_name:
         raise HTTPException(
-            status_code=400, detail="Missing 'name' of model to unload."
+            status_code=400, detail="Missing name of the model to unload."
         )
 
     # Refresh/load models if needed, get mapping from name to URLs
@@ -698,11 +724,14 @@ async def pull_model(
     url_idx: int = 0,
     user=Depends(get_admin_user),
 ):
+    form_data = form_data.model_dump(exclude_none=True)
+    form_data["model"] = form_data.get("model", form_data.get("name"))
+
     url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
     log.info(f"url: {url}")
 
     # Admin should be able to pull models from any source
-    payload = {**form_data.model_dump(exclude_none=True), "insecure": True}
+    payload = {**form_data, "insecure": True}
 
     return await send_post_request(
         url=f"{url}/api/pull",
@@ -713,7 +742,7 @@ async def pull_model(
 
 
 class PushModelForm(BaseModel):
-    name: str
+    model: str
     insecure: Optional[bool] = None
     stream: Optional[bool] = None
 
@@ -730,12 +759,12 @@ async def push_model(
         await get_all_models(request, user=user)
         models = request.app.state.OLLAMA_MODELS
 
-        if form_data.name in models:
-            url_idx = models[form_data.name]["urls"][0]
+        if form_data.model in models:
+            url_idx = models[form_data.model]["urls"][0]
         else:
             raise HTTPException(
                 status_code=400,
-                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.name),
+                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.model),
             )
 
     url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
@@ -813,7 +842,7 @@ async def copy_model(
                 **({"Authorization": f"Bearer {key}"} if key else {}),
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -854,16 +883,21 @@ async def delete_model(
     url_idx: Optional[int] = None,
     user=Depends(get_admin_user),
 ):
+    form_data = form_data.model_dump(exclude_none=True)
+    form_data["model"] = form_data.get("model", form_data.get("name"))
+
+    model = form_data.get("model")
+
     if url_idx is None:
         await get_all_models(request, user=user)
         models = request.app.state.OLLAMA_MODELS
 
-        if form_data.name in models:
-            url_idx = models[form_data.name]["urls"][0]
+        if model in models:
+            url_idx = models[model]["urls"][0]
         else:
             raise HTTPException(
                 status_code=400,
-                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.name),
+                detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model),
             )
 
     url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
@@ -873,13 +907,13 @@ async def delete_model(
         r = requests.request(
             method="DELETE",
             url=f"{url}/api/delete",
-            data=form_data.model_dump_json(exclude_none=True).encode(),
+            data=json.dumps(form_data).encode(),
             headers={
                 "Content-Type": "application/json",
                 **({"Authorization": f"Bearer {key}"} if key else {}),
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -915,16 +949,21 @@ async def delete_model(
 async def show_model_info(
     request: Request, form_data: ModelNameForm, user=Depends(get_verified_user)
 ):
+    form_data = form_data.model_dump(exclude_none=True)
+    form_data["model"] = form_data.get("model", form_data.get("name"))
+
     await get_all_models(request, user=user)
     models = request.app.state.OLLAMA_MODELS
 
-    if form_data.name not in models:
+    model = form_data.get("model")
+
+    if model not in models:
         raise HTTPException(
             status_code=400,
-            detail=ERROR_MESSAGES.MODEL_NOT_FOUND(form_data.name),
+            detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model),
         )
 
-    url_idx = random.choice(models[form_data.name]["urls"])
+    url_idx = random.choice(models[model]["urls"])
 
     url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
     key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)
@@ -938,7 +977,7 @@ async def show_model_info(
                 **({"Authorization": f"Bearer {key}"} if key else {}),
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -947,7 +986,7 @@ async def show_model_info(
                     else {}
                 ),
             },
-            data=form_data.model_dump_json(exclude_none=True).encode(),
+            data=json.dumps(form_data).encode(),
         )
         r.raise_for_status()
 
@@ -1025,7 +1064,7 @@ async def embed(
                 **({"Authorization": f"Bearer {key}"} if key else {}),
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -1112,7 +1151,7 @@ async def embeddings(
                 **({"Authorization": f"Bearer {key}"} if key else {}),
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -1232,6 +1271,9 @@ class GenerateChatCompletionForm(BaseModel):
     stream: Optional[bool] = True
     keep_alive: Optional[Union[int, str]] = None
     tools: Optional[list[dict]] = None
+    model_config = ConfigDict(
+        extra="allow",
+    )
 
 
 async def get_ollama_url(request: Request, model: str, url_idx: Optional[int] = None):
@@ -1269,7 +1311,9 @@ async def generate_chat_completion(
             detail=str(e),
         )
 
-    payload = {**form_data.model_dump(exclude_none=True)}
+    if isinstance(form_data, BaseModel):
+        payload = {**form_data.model_dump(exclude_none=True)}
+
     if "metadata" in payload:
         del payload["metadata"]
 
@@ -1283,13 +1327,10 @@ async def generate_chat_completion(
         params = model_info.params.model_dump()
 
         if params:
-            if payload.get("options") is None:
-                payload["options"] = {}
+            system = params.pop("system", None)
 
-            payload["options"] = apply_model_params_to_body_ollama(
-                params, payload["options"]
-            )
-            payload = apply_model_system_prompt_to_body(params, payload, metadata, user)
+            payload = apply_model_params_to_body_ollama(params, payload)
+            payload = apply_model_system_prompt_to_body(system, payload, metadata, user)
 
         # Check if user has access to the model
         if not bypass_filter and user.role == "user":
@@ -1322,7 +1363,7 @@ async def generate_chat_completion(
     prefix_id = api_config.get("prefix_id", None)
     if prefix_id:
         payload["model"] = payload["model"].replace(f"{prefix_id}.", "")
-    # payload["keep_alive"] = -1 # keep alive forever
+
     return await send_post_request(
         url=f"{url}/api/chat",
         payload=json.dumps(payload),
@@ -1330,6 +1371,7 @@ async def generate_chat_completion(
         key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
         content_type="application/x-ndjson",
         user=user,
+        metadata=metadata,
     )
 
 
@@ -1368,6 +1410,8 @@ async def generate_openai_completion(
     url_idx: Optional[int] = None,
     user=Depends(get_verified_user),
 ):
+    metadata = form_data.pop("metadata", None)
+
     try:
         form_data = OpenAICompletionForm(**form_data)
     except Exception as e:
@@ -1433,6 +1477,7 @@ async def generate_openai_completion(
         stream=payload.get("stream", False),
         key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
         user=user,
+        metadata=metadata,
     )
 
 
@@ -1471,8 +1516,10 @@ async def generate_openai_chat_completion(
         params = model_info.params.model_dump()
 
         if params:
+            system = params.pop("system", None)
+
             payload = apply_model_params_to_body_openai(params, payload)
-            payload = apply_model_system_prompt_to_body(params, payload, metadata, user)
+            payload = apply_model_system_prompt_to_body(system, payload, metadata, user)
 
         # Check if user has access to the model
         if user.role == "user":
@@ -1512,6 +1559,7 @@ async def generate_openai_chat_completion(
         stream=payload.get("stream", False),
         key=get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS),
         user=user,
+        metadata=metadata,
     )
 
 

backend/open_webui/routers/openai.py

@@ -2,17 +2,20 @@ import asyncio
 import hashlib
 import json
 import logging
-from pathlib import Path
-from typing import Literal, Optional, overload
+from typing import Optional
 
 import aiohttp
 from aiocache import cached
 import requests
+from urllib.parse import quote
 
-
-from fastapi import Depends, FastAPI, HTTPException, Request, APIRouter
-from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import FileResponse, StreamingResponse
+from fastapi import Depends, HTTPException, Request, APIRouter
+from fastapi.responses import (
+    FileResponse,
+    StreamingResponse,
+    JSONResponse,
+    PlainTextResponse,
+)
 from pydantic import BaseModel
 from starlette.background import BackgroundTask
 
@@ -21,6 +24,7 @@ from open_webui.config import (
     CACHE_DIR,
 )
 from open_webui.env import (
+    MODELS_CACHE_TTL,
     AIOHTTP_CLIENT_SESSION_SSL,
     AIOHTTP_CLIENT_TIMEOUT,
     AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST,
@@ -30,7 +34,7 @@ from open_webui.env import (
 from open_webui.models.users import UserModel
 
 from open_webui.constants import ERROR_MESSAGES
-from open_webui.env import ENV, SRC_LOG_LEVELS
+from open_webui.env import SRC_LOG_LEVELS
 
 
 from open_webui.utils.payload import (
@@ -66,7 +70,7 @@ async def send_get_request(url, key=None, user: UserModel = None):
                     **({"Authorization": f"Bearer {key}"} if key else {}),
                     **(
                         {
-                            "X-OpenWebUI-User-Name": user.name,
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                             "X-OpenWebUI-User-Id": user.id,
                             "X-OpenWebUI-User-Email": user.email,
                             "X-OpenWebUI-User-Role": user.role,
@@ -94,12 +98,12 @@ async def cleanup_response(
         await session.close()
 
 
-def openai_o_series_handler(payload):
+def openai_reasoning_model_handler(payload):
     """
-    Handle "o" series specific parameters
+    Handle reasoning model specific parameters
     """
     if "max_tokens" in payload:
-        # Convert "max_tokens" to "max_completion_tokens" for all o-series models
+        # Convert "max_tokens" to "max_completion_tokens" for all reasoning models
         payload["max_completion_tokens"] = payload["max_tokens"]
         del payload["max_tokens"]
 
@@ -225,7 +229,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                     ),
                     **(
                         {
-                            "X-OpenWebUI-User-Name": user.name,
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                             "X-OpenWebUI-User-Id": user.id,
                             "X-OpenWebUI-User-Email": user.email,
                             "X-OpenWebUI-User-Role": user.role,
@@ -361,7 +365,9 @@ async def get_all_models_responses(request: Request, user: UserModel) -> list:
                 response if isinstance(response, list) else response.get("data", [])
             ):
                 if prefix_id:
-                    model["id"] = f"{prefix_id}.{model['id']}"
+                    model["id"] = (
+                        f"{prefix_id}.{model.get('id', model.get('name', ''))}"
+                    )
 
                 if tags:
                     model["tags"] = tags
@@ -386,7 +392,7 @@ async def get_filtered_models(models, user):
     return filtered_models
 
 
-@cached(ttl=1)
+@cached(ttl=MODELS_CACHE_TTL)
 async def get_all_models(request: Request, user: UserModel) -> dict[str, list]:
     log.info("get_all_models()")
 
@@ -478,7 +484,7 @@ async def get_models(
                     "Content-Type": "application/json",
                     **(
                         {
-                            "X-OpenWebUI-User-Name": user.name,
+                            "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                             "X-OpenWebUI-User-Id": user.id,
                             "X-OpenWebUI-User-Email": user.email,
                             "X-OpenWebUI-User-Role": user.role,
@@ -573,7 +579,7 @@ async def verify_connection(
                 "Content-Type": "application/json",
                 **(
                     {
-                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                         "X-OpenWebUI-User-Id": user.id,
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
@@ -592,15 +598,21 @@ async def verify_connection(
                     headers=headers,
                     ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as r:
+                    try:
+                        response_data = await r.json()
+                    except Exception:
+                        response_data = await r.text()
+
                     if r.status != 200:
-                        # Extract response error details if available
-                        error_detail = f"HTTP Error: {r.status}"
-                        res = await r.json()
-                        if "error" in res:
-                            error_detail = f"External Error: {res['error']}"
-                        raise Exception(error_detail)
+                        if isinstance(response_data, (dict, list)):
+                            return JSONResponse(
+                                status_code=r.status, content=response_data
+                            )
+                        else:
+                            return PlainTextResponse(
+                                status_code=r.status, content=response_data
+                            )
 
-                    response_data = await r.json()
                     return response_data
             else:
                 headers["Authorization"] = f"Bearer {key}"
@@ -610,15 +622,21 @@ async def verify_connection(
                     headers=headers,
                     ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as r:
+                    try:
+                        response_data = await r.json()
+                    except Exception:
+                        response_data = await r.text()
+
                     if r.status != 200:
-                        # Extract response error details if available
-                        error_detail = f"HTTP Error: {r.status}"
-                        res = await r.json()
-                        if "error" in res:
-                            error_detail = f"External Error: {res['error']}"
-                        raise Exception(error_detail)
+                        if isinstance(response_data, (dict, list)):
+                            return JSONResponse(
+                                status_code=r.status, content=response_data
+                            )
+                        else:
+                            return PlainTextResponse(
+                                status_code=r.status, content=response_data
+                            )
 
-                    response_data = await r.json()
                     return response_data
 
         except aiohttp.ClientError as e:
@@ -629,17 +647,12 @@ async def verify_connection(
             )
         except Exception as e:
             log.exception(f"Unexpected error: {e}")
-            error_detail = f"Unexpected error: {str(e)}"
-            raise HTTPException(status_code=500, detail=error_detail)
+            raise HTTPException(
+                status_code=500, detail="Open WebUI: Server Connection Error"
+            )
 
 
-def convert_to_azure_payload(
-    url,
-    payload: dict,
-):
-    model = payload.get("model", "")
-
-    # Filter allowed parameters based on Azure OpenAI API
+def get_azure_allowed_params(api_version: str) -> set[str]:
     allowed_params = {
         "messages",
         "temperature",
@@ -669,6 +682,23 @@ def convert_to_azure_payload(
         "max_completion_tokens",
     }
 
+    try:
+        if api_version >= "2024-09-01-preview":
+            allowed_params.add("stream_options")
+    except ValueError:
+        log.debug(
+            f"Invalid API version {api_version} for Azure OpenAI. Defaulting to allowed parameters."
+        )
+
+    return allowed_params
+
+
+def convert_to_azure_payload(url, payload: dict, api_version: str):
+    model = payload.get("model", "")
+
+    # Filter allowed parameters based on Azure OpenAI API
+    allowed_params = get_azure_allowed_params(api_version)
+
     # Special handling for o-series models
     if model.startswith("o") and model.endswith("-mini"):
         # Convert max_tokens to max_completion_tokens for o-series models
@@ -715,8 +745,12 @@ async def generate_chat_completion(
             model_id = model_info.base_model_id
 
         params = model_info.params.model_dump()
-        payload = apply_model_params_to_body_openai(params, payload)
-        payload = apply_model_system_prompt_to_body(params, payload, metadata, user)
+
+        if params:
+            system = params.pop("system", None)
+
+            payload = apply_model_params_to_body_openai(params, payload)
+            payload = apply_model_system_prompt_to_body(system, payload, metadata, user)
 
         # Check if user has access to the model
         if not bypass_filter and user.role == "user":
@@ -771,10 +805,12 @@ async def generate_chat_completion(
     url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
     key = request.app.state.config.OPENAI_API_KEYS[idx]
 
-    # Check if model is from "o" series
-    is_o_series = payload["model"].lower().startswith(("o1", "o3", "o4"))
-    if is_o_series:
-        payload = openai_o_series_handler(payload)
+    # Check if model is a reasoning model that needs special handling
+    is_reasoning_model = (
+        payload["model"].lower().startswith(("o1", "o3", "o4", "gpt-5"))
+    )
+    if is_reasoning_model:
+        payload = openai_reasoning_model_handler(payload)
     elif "api.openai.com" not in url:
         # Remove "max_completion_tokens" from the payload for backward compatibility
         if "max_completion_tokens" in payload:
@@ -802,10 +838,15 @@ async def generate_chat_completion(
         ),
         **(
             {
-                "X-OpenWebUI-User-Name": user.name,
+                "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                 "X-OpenWebUI-User-Id": user.id,
                 "X-OpenWebUI-User-Email": user.email,
                 "X-OpenWebUI-User-Role": user.role,
+                **(
+                    {"X-OpenWebUI-Chat-Id": metadata.get("chat_id")}
+                    if metadata and metadata.get("chat_id")
+                    else {}
+                ),
             }
             if ENABLE_FORWARD_USER_INFO_HEADERS
             else {}
@@ -813,8 +854,8 @@ async def generate_chat_completion(
     }
 
     if api_config.get("azure", False):
-        request_url, payload = convert_to_azure_payload(url, payload)
-        api_version = api_config.get("api_version", "") or "2023-03-15-preview"
+        api_version = api_config.get("api_version", "2023-03-15-preview")
+        request_url, payload = convert_to_azure_payload(url, payload, api_version)
         headers["api-key"] = key
         headers["api-version"] = api_version
         request_url = f"{request_url}/chat/completions?api-version={api_version}"
@@ -860,27 +901,107 @@ async def generate_chat_completion(
                 log.error(e)
                 response = await r.text()
 
-            r.raise_for_status()
+            if r.status >= 400:
+                if isinstance(response, (dict, list)):
+                    return JSONResponse(status_code=r.status, content=response)
+                else:
+                    return PlainTextResponse(status_code=r.status, content=response)
+
             return response
     except Exception as e:
         log.exception(e)
 
-        detail = None
-        if isinstance(response, dict):
-            if "error" in response:
-                detail = f"{response['error']['message'] if 'message' in response['error'] else response['error']}"
-        elif isinstance(response, str):
-            detail = response
-
         raise HTTPException(
             status_code=r.status if r else 500,
-            detail=detail if detail else "Open WebUI: Server Connection Error",
+            detail="Open WebUI: Server Connection Error",
         )
     finally:
-        if not streaming and session:
-            if r:
-                r.close()
-            await session.close()
+        if not streaming:
+            await cleanup_response(r, session)
+
+
+async def embeddings(request: Request, form_data: dict, user):
+    """
+    Calls the embeddings endpoint for OpenAI-compatible providers.
+
+    Args:
+        request (Request): The FastAPI request context.
+        form_data (dict): OpenAI-compatible embeddings payload.
+        user (UserModel): The authenticated user.
+
+    Returns:
+        dict: OpenAI-compatible embeddings response.
+    """
+    idx = 0
+    # Prepare payload/body
+    body = json.dumps(form_data)
+    # Find correct backend url/key based on model
+    await get_all_models(request, user=user)
+    model_id = form_data.get("model")
+    models = request.app.state.OPENAI_MODELS
+    if model_id in models:
+        idx = models[model_id]["urlIdx"]
+    url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
+    key = request.app.state.config.OPENAI_API_KEYS[idx]
+    r = None
+    session = None
+    streaming = False
+    try:
+        session = aiohttp.ClientSession(trust_env=True)
+        r = await session.request(
+            method="POST",
+            url=f"{url}/embeddings",
+            data=body,
+            headers={
+                "Authorization": f"Bearer {key}",
+                "Content-Type": "application/json",
+                **(
+                    {
+                        "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
+                        "X-OpenWebUI-User-Id": user.id,
+                        "X-OpenWebUI-User-Email": user.email,
+                        "X-OpenWebUI-User-Role": user.role,
+                    }
+                    if ENABLE_FORWARD_USER_INFO_HEADERS and user
+                    else {}
+                ),
+            },
+        )
+
+        if "text/event-stream" in r.headers.get("Content-Type", ""):
+            streaming = True
+            return StreamingResponse(
+                r.content,
+                status_code=r.status,
+                headers=dict(r.headers),
+                background=BackgroundTask(
+                    cleanup_response, response=r, session=session
+                ),
+            )
+        else:
+            try:
+                response_data = await r.json()
+            except Exception:
+                response_data = await r.text()
+
+            if r.status >= 400:
+                if isinstance(response_data, (dict, list)):
+                    return JSONResponse(status_code=r.status, content=response_data)
+                else:
+                    return PlainTextResponse(
+                        status_code=r.status, content=response_data
+                    )
+
+            return response_data
+    except Exception as e:
+        log.exception(e)
+        raise HTTPException(
+            status_code=r.status if r else 500,
+            detail="Open WebUI: Server Connection Error",
+        )
+    finally:
+        if not streaming:
+            await cleanup_response(r, session)
 
 
 @router.api_route("/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])
@@ -910,7 +1031,7 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
             "Content-Type": "application/json",
             **(
                 {
-                    "X-OpenWebUI-User-Name": user.name,
+                    "X-OpenWebUI-User-Name": quote(user.name, safe=" "),
                     "X-OpenWebUI-User-Id": user.id,
                     "X-OpenWebUI-User-Email": user.email,
                     "X-OpenWebUI-User-Role": user.role,
@@ -921,16 +1042,15 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
         }
 
         if api_config.get("azure", False):
+            api_version = api_config.get("api_version", "2023-03-15-preview")
             headers["api-key"] = key
-            headers["api-version"] = (
-                api_config.get("api_version", "") or "2023-03-15-preview"
-            )
+            headers["api-version"] = api_version
 
             payload = json.loads(body)
-            url, payload = convert_to_azure_payload(url, payload)
+            url, payload = convert_to_azure_payload(url, payload, api_version)
             body = json.dumps(payload).encode()
 
-            request_url = f"{url}/{path}?api-version={api_config.get('api_version', '2023-03-15-preview')}"
+            request_url = f"{url}/{path}?api-version={api_version}"
         else:
             headers["Authorization"] = f"Bearer {key}"
             request_url = f"{url}/{path}"
@@ -943,7 +1063,6 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
             headers=headers,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
         )
-        r.raise_for_status()
 
         # Check if response is SSE
         if "text/event-stream" in r.headers.get("Content-Type", ""):
@@ -957,27 +1076,27 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
                 ),
             )
         else:
-            response_data = await r.json()
+            try:
+                response_data = await r.json()
+            except Exception:
+                response_data = await r.text()
+
+            if r.status >= 400:
+                if isinstance(response_data, (dict, list)):
+                    return JSONResponse(status_code=r.status, content=response_data)
+                else:
+                    return PlainTextResponse(
+                        status_code=r.status, content=response_data
+                    )
+
             return response_data
 
     except Exception as e:
         log.exception(e)
-
-        detail = None
-        if r is not None:
-            try:
-                res = await r.json()
-                log.error(res)
-                if "error" in res:
-                    detail = f"External: {res['error']['message'] if 'message' in res['error'] else res['error']}"
-            except Exception:
-                detail = f"External: {e}"
         raise HTTPException(
             status_code=r.status if r else 500,
-            detail=detail if detail else "Open WebUI: Server Connection Error",
+            detail="Open WebUI: Server Connection Error",
         )
     finally:
-        if not streaming and session:
-            if r:
-                r.close()
-            await session.close()
+        if not streaming:
+            await cleanup_response(r, session)

backend/open_webui/routers/prompts.py

@@ -1,4 +1,5 @@
 from typing import Optional
+from fastapi import APIRouter, Depends, HTTPException, status, Request
 
 from open_webui.models.prompts import (
     PromptForm,
@@ -7,9 +8,9 @@ from open_webui.models.prompts import (
     Prompts,
 )
 from open_webui.constants import ERROR_MESSAGES
-from fastapi import APIRouter, Depends, HTTPException, status, Request
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.access_control import has_access, has_permission
+from open_webui.config import ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS
 
 router = APIRouter()
 
@@ -20,7 +21,7 @@ router = APIRouter()
 
 @router.get("/", response_model=list[PromptModel])
 async def get_prompts(user=Depends(get_verified_user)):
-    if user.role == "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
         prompts = Prompts.get_prompts()
     else:
         prompts = Prompts.get_prompts_by_user_id(user.id, "read")
@@ -30,7 +31,7 @@ async def get_prompts(user=Depends(get_verified_user)):
 
 @router.get("/list", response_model=list[PromptUserResponse])
 async def get_prompt_list(user=Depends(get_verified_user)):
-    if user.role == "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
         prompts = Prompts.get_prompts()
     else:
         prompts = Prompts.get_prompts_by_user_id(user.id, "write")

backend/open_webui/routers/retrieval.py

@@ -29,6 +29,7 @@ import tiktoken
 
 
 from langchain.text_splitter import RecursiveCharacterTextSplitter, TokenTextSplitter
+from langchain_text_splitters import MarkdownHeaderTextSplitter
 from langchain_core.documents import Document
 
 from open_webui.models.files import FileModel, Files
@@ -69,6 +70,7 @@ from open_webui.retrieval.web.external import search_external
 
 from open_webui.retrieval.utils import (
     get_embedding_function,
+    get_reranking_function,
     get_model_path,
     query_collection,
     query_collection_with_hybrid_search,
@@ -239,6 +241,11 @@ async def get_embedding_config(request: Request, user=Depends(get_admin_user)):
             "url": request.app.state.config.RAG_OLLAMA_BASE_URL,
             "key": request.app.state.config.RAG_OLLAMA_API_KEY,
         },
+        "azure_openai_config": {
+            "url": request.app.state.config.RAG_AZURE_OPENAI_BASE_URL,
+            "key": request.app.state.config.RAG_AZURE_OPENAI_API_KEY,
+            "version": request.app.state.config.RAG_AZURE_OPENAI_API_VERSION,
+        },
     }
 
 
@@ -252,9 +259,16 @@ class OllamaConfigForm(BaseModel):
     key: str
 
 
+class AzureOpenAIConfigForm(BaseModel):
+    url: str
+    key: str
+    version: str
+
+
 class EmbeddingModelUpdateForm(BaseModel):
     openai_config: Optional[OpenAIConfigForm] = None
     ollama_config: Optional[OllamaConfigForm] = None
+    azure_openai_config: Optional[AzureOpenAIConfigForm] = None
     embedding_engine: str
     embedding_model: str
     embedding_batch_size: Optional[int] = 1
@@ -271,7 +285,11 @@ async def update_embedding_config(
         request.app.state.config.RAG_EMBEDDING_ENGINE = form_data.embedding_engine
         request.app.state.config.RAG_EMBEDDING_MODEL = form_data.embedding_model
 
-        if request.app.state.config.RAG_EMBEDDING_ENGINE in ["ollama", "openai"]:
+        if request.app.state.config.RAG_EMBEDDING_ENGINE in [
+            "ollama",
+            "openai",
+            "azure_openai",
+        ]:
             if form_data.openai_config is not None:
                 request.app.state.config.RAG_OPENAI_API_BASE_URL = (
                     form_data.openai_config.url
@@ -288,6 +306,17 @@ async def update_embedding_config(
                     form_data.ollama_config.key
                 )
 
+            if form_data.azure_openai_config is not None:
+                request.app.state.config.RAG_AZURE_OPENAI_BASE_URL = (
+                    form_data.azure_openai_config.url
+                )
+                request.app.state.config.RAG_AZURE_OPENAI_API_KEY = (
+                    form_data.azure_openai_config.key
+                )
+                request.app.state.config.RAG_AZURE_OPENAI_API_VERSION = (
+                    form_data.azure_openai_config.version
+                )
+
             request.app.state.config.RAG_EMBEDDING_BATCH_SIZE = (
                 form_data.embedding_batch_size
             )
@@ -304,14 +333,27 @@ async def update_embedding_config(
             (
                 request.app.state.config.RAG_OPENAI_API_BASE_URL
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_BASE_URL
+                else (
+                    request.app.state.config.RAG_OLLAMA_BASE_URL
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_BASE_URL
+                )
             ),
             (
                 request.app.state.config.RAG_OPENAI_API_KEY
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_API_KEY
+                else (
+                    request.app.state.config.RAG_OLLAMA_API_KEY
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_API_KEY
+                )
             ),
             request.app.state.config.RAG_EMBEDDING_BATCH_SIZE,
+            azure_api_version=(
+                request.app.state.config.RAG_AZURE_OPENAI_API_VERSION
+                if request.app.state.config.RAG_EMBEDDING_ENGINE == "azure_openai"
+                else None
+            ),
         )
 
         return {
@@ -327,6 +369,11 @@ async def update_embedding_config(
                 "url": request.app.state.config.RAG_OLLAMA_BASE_URL,
                 "key": request.app.state.config.RAG_OLLAMA_API_KEY,
             },
+            "azure_openai_config": {
+                "url": request.app.state.config.RAG_AZURE_OPENAI_BASE_URL,
+                "key": request.app.state.config.RAG_AZURE_OPENAI_API_KEY,
+                "version": request.app.state.config.RAG_AZURE_OPENAI_API_VERSION,
+            },
         }
     except Exception as e:
         log.exception(f"Problem updating embedding model: {e}")
@@ -354,12 +401,14 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         "CONTENT_EXTRACTION_ENGINE": request.app.state.config.CONTENT_EXTRACTION_ENGINE,
         "PDF_EXTRACT_IMAGES": request.app.state.config.PDF_EXTRACT_IMAGES,
         "DATALAB_MARKER_API_KEY": request.app.state.config.DATALAB_MARKER_API_KEY,
-        "DATALAB_MARKER_LANGS": request.app.state.config.DATALAB_MARKER_LANGS,
+        "DATALAB_MARKER_API_BASE_URL": request.app.state.config.DATALAB_MARKER_API_BASE_URL,
+        "DATALAB_MARKER_ADDITIONAL_CONFIG": request.app.state.config.DATALAB_MARKER_ADDITIONAL_CONFIG,
         "DATALAB_MARKER_SKIP_CACHE": request.app.state.config.DATALAB_MARKER_SKIP_CACHE,
         "DATALAB_MARKER_FORCE_OCR": request.app.state.config.DATALAB_MARKER_FORCE_OCR,
         "DATALAB_MARKER_PAGINATE": request.app.state.config.DATALAB_MARKER_PAGINATE,
         "DATALAB_MARKER_STRIP_EXISTING_OCR": request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR,
         "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION": request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+        "DATALAB_MARKER_FORMAT_LINES": request.app.state.config.DATALAB_MARKER_FORMAT_LINES,
         "DATALAB_MARKER_USE_LLM": request.app.state.config.DATALAB_MARKER_USE_LLM,
         "DATALAB_MARKER_OUTPUT_FORMAT": request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT,
         "EXTERNAL_DOCUMENT_LOADER_URL": request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL,
@@ -369,6 +418,9 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         "DOCLING_OCR_ENGINE": request.app.state.config.DOCLING_OCR_ENGINE,
         "DOCLING_OCR_LANG": request.app.state.config.DOCLING_OCR_LANG,
         "DOCLING_DO_PICTURE_DESCRIPTION": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+        "DOCLING_PICTURE_DESCRIPTION_MODE": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL,
+        "DOCLING_PICTURE_DESCRIPTION_API": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API,
         "DOCUMENT_INTELLIGENCE_ENDPOINT": request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
         "DOCUMENT_INTELLIGENCE_KEY": request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
         "MISTRAL_OCR_API_KEY": request.app.state.config.MISTRAL_OCR_API_KEY,
@@ -384,6 +436,8 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         # File upload settings
         "FILE_MAX_SIZE": request.app.state.config.FILE_MAX_SIZE,
         "FILE_MAX_COUNT": request.app.state.config.FILE_MAX_COUNT,
+        "FILE_IMAGE_COMPRESSION_WIDTH": request.app.state.config.FILE_IMAGE_COMPRESSION_WIDTH,
+        "FILE_IMAGE_COMPRESSION_HEIGHT": request.app.state.config.FILE_IMAGE_COMPRESSION_HEIGHT,
         "ALLOWED_FILE_EXTENSIONS": request.app.state.config.ALLOWED_FILE_EXTENSIONS,
         # Integration settings
         "ENABLE_GOOGLE_DRIVE_INTEGRATION": request.app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
@@ -422,6 +476,8 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
             "BING_SEARCH_V7_SUBSCRIPTION_KEY": request.app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY,
             "EXA_API_KEY": request.app.state.config.EXA_API_KEY,
             "PERPLEXITY_API_KEY": request.app.state.config.PERPLEXITY_API_KEY,
+            "PERPLEXITY_MODEL": request.app.state.config.PERPLEXITY_MODEL,
+            "PERPLEXITY_SEARCH_CONTEXT_USAGE": request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE,
             "SOUGOU_API_SID": request.app.state.config.SOUGOU_API_SID,
             "SOUGOU_API_SK": request.app.state.config.SOUGOU_API_SK,
             "WEB_LOADER_ENGINE": request.app.state.config.WEB_LOADER_ENGINE,
@@ -475,6 +531,8 @@ class WebConfig(BaseModel):
     BING_SEARCH_V7_SUBSCRIPTION_KEY: Optional[str] = None
     EXA_API_KEY: Optional[str] = None
     PERPLEXITY_API_KEY: Optional[str] = None
+    PERPLEXITY_MODEL: Optional[str] = None
+    PERPLEXITY_SEARCH_CONTEXT_USAGE: Optional[str] = None
     SOUGOU_API_SID: Optional[str] = None
     SOUGOU_API_SK: Optional[str] = None
     WEB_LOADER_ENGINE: Optional[str] = None
@@ -510,12 +568,14 @@ class ConfigForm(BaseModel):
     CONTENT_EXTRACTION_ENGINE: Optional[str] = None
     PDF_EXTRACT_IMAGES: Optional[bool] = None
     DATALAB_MARKER_API_KEY: Optional[str] = None
-    DATALAB_MARKER_LANGS: Optional[str] = None
+    DATALAB_MARKER_API_BASE_URL: Optional[str] = None
+    DATALAB_MARKER_ADDITIONAL_CONFIG: Optional[str] = None
     DATALAB_MARKER_SKIP_CACHE: Optional[bool] = None
     DATALAB_MARKER_FORCE_OCR: Optional[bool] = None
     DATALAB_MARKER_PAGINATE: Optional[bool] = None
     DATALAB_MARKER_STRIP_EXISTING_OCR: Optional[bool] = None
     DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION: Optional[bool] = None
+    DATALAB_MARKER_FORMAT_LINES: Optional[bool] = None
     DATALAB_MARKER_USE_LLM: Optional[bool] = None
     DATALAB_MARKER_OUTPUT_FORMAT: Optional[str] = None
     EXTERNAL_DOCUMENT_LOADER_URL: Optional[str] = None
@@ -526,6 +586,9 @@ class ConfigForm(BaseModel):
     DOCLING_OCR_ENGINE: Optional[str] = None
     DOCLING_OCR_LANG: Optional[str] = None
     DOCLING_DO_PICTURE_DESCRIPTION: Optional[bool] = None
+    DOCLING_PICTURE_DESCRIPTION_MODE: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_LOCAL: Optional[dict] = None
+    DOCLING_PICTURE_DESCRIPTION_API: Optional[dict] = None
     DOCUMENT_INTELLIGENCE_ENDPOINT: Optional[str] = None
     DOCUMENT_INTELLIGENCE_KEY: Optional[str] = None
     MISTRAL_OCR_API_KEY: Optional[str] = None
@@ -544,6 +607,8 @@ class ConfigForm(BaseModel):
     # File upload settings
     FILE_MAX_SIZE: Optional[int] = None
     FILE_MAX_COUNT: Optional[int] = None
+    FILE_IMAGE_COMPRESSION_WIDTH: Optional[int] = None
+    FILE_IMAGE_COMPRESSION_HEIGHT: Optional[int] = None
     ALLOWED_FILE_EXTENSIONS: Optional[List[str]] = None
 
     # Integration settings
@@ -622,10 +687,15 @@ async def update_rag_config(
         if form_data.DATALAB_MARKER_API_KEY is not None
         else request.app.state.config.DATALAB_MARKER_API_KEY
     )
-    request.app.state.config.DATALAB_MARKER_LANGS = (
-        form_data.DATALAB_MARKER_LANGS
-        if form_data.DATALAB_MARKER_LANGS is not None
-        else request.app.state.config.DATALAB_MARKER_LANGS
+    request.app.state.config.DATALAB_MARKER_API_BASE_URL = (
+        form_data.DATALAB_MARKER_API_BASE_URL
+        if form_data.DATALAB_MARKER_API_BASE_URL is not None
+        else request.app.state.config.DATALAB_MARKER_API_BASE_URL
+    )
+    request.app.state.config.DATALAB_MARKER_ADDITIONAL_CONFIG = (
+        form_data.DATALAB_MARKER_ADDITIONAL_CONFIG
+        if form_data.DATALAB_MARKER_ADDITIONAL_CONFIG is not None
+        else request.app.state.config.DATALAB_MARKER_ADDITIONAL_CONFIG
     )
     request.app.state.config.DATALAB_MARKER_SKIP_CACHE = (
         form_data.DATALAB_MARKER_SKIP_CACHE
@@ -652,6 +722,11 @@ async def update_rag_config(
         if form_data.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION is not None
         else request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
     )
+    request.app.state.config.DATALAB_MARKER_FORMAT_LINES = (
+        form_data.DATALAB_MARKER_FORMAT_LINES
+        if form_data.DATALAB_MARKER_FORMAT_LINES is not None
+        else request.app.state.config.DATALAB_MARKER_FORMAT_LINES
+    )
     request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT = (
         form_data.DATALAB_MARKER_OUTPUT_FORMAT
         if form_data.DATALAB_MARKER_OUTPUT_FORMAT is not None
@@ -699,6 +774,22 @@ async def update_rag_config(
         else request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION
     )
 
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_MODE
+        if form_data.DOCLING_PICTURE_DESCRIPTION_MODE is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL
+        if form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_API
+        if form_data.DOCLING_PICTURE_DESCRIPTION_API is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API
+    )
+
     request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT = (
         form_data.DOCUMENT_INTELLIGENCE_ENDPOINT
         if form_data.DOCUMENT_INTELLIGENCE_ENDPOINT is not None
@@ -738,7 +829,11 @@ async def update_rag_config(
         f"Updating reranking model: {request.app.state.config.RAG_RERANKING_MODEL} to {form_data.RAG_RERANKING_MODEL}"
     )
     try:
-        request.app.state.config.RAG_RERANKING_MODEL = form_data.RAG_RERANKING_MODEL
+        request.app.state.config.RAG_RERANKING_MODEL = (
+            form_data.RAG_RERANKING_MODEL
+            if form_data.RAG_RERANKING_MODEL is not None
+            else request.app.state.config.RAG_RERANKING_MODEL
+        )
 
         try:
             request.app.state.rf = get_rf(
@@ -748,6 +843,12 @@ async def update_rag_config(
                 request.app.state.config.RAG_EXTERNAL_RERANKER_API_KEY,
                 True,
             )
+
+            request.app.state.RERANKING_FUNCTION = get_reranking_function(
+                request.app.state.config.RAG_RERANKING_ENGINE,
+                request.app.state.config.RAG_RERANKING_MODEL,
+                request.app.state.rf,
+            )
         except Exception as e:
             log.error(f"Error loading reranking model: {e}")
             request.app.state.config.ENABLE_RAG_HYBRID_SEARCH = False
@@ -776,15 +877,13 @@ async def update_rag_config(
     )
 
     # File upload settings
-    request.app.state.config.FILE_MAX_SIZE = (
-        form_data.FILE_MAX_SIZE
-        if form_data.FILE_MAX_SIZE is not None
-        else request.app.state.config.FILE_MAX_SIZE
+    request.app.state.config.FILE_MAX_SIZE = form_data.FILE_MAX_SIZE
+    request.app.state.config.FILE_MAX_COUNT = form_data.FILE_MAX_COUNT
+    request.app.state.config.FILE_IMAGE_COMPRESSION_WIDTH = (
+        form_data.FILE_IMAGE_COMPRESSION_WIDTH
     )
-    request.app.state.config.FILE_MAX_COUNT = (
-        form_data.FILE_MAX_COUNT
-        if form_data.FILE_MAX_COUNT is not None
-        else request.app.state.config.FILE_MAX_COUNT
+    request.app.state.config.FILE_IMAGE_COMPRESSION_HEIGHT = (
+        form_data.FILE_IMAGE_COMPRESSION_HEIGHT
     )
     request.app.state.config.ALLOWED_FILE_EXTENSIONS = (
         form_data.ALLOWED_FILE_EXTENSIONS
@@ -862,6 +961,10 @@ async def update_rag_config(
         )
         request.app.state.config.EXA_API_KEY = form_data.web.EXA_API_KEY
         request.app.state.config.PERPLEXITY_API_KEY = form_data.web.PERPLEXITY_API_KEY
+        request.app.state.config.PERPLEXITY_MODEL = form_data.web.PERPLEXITY_MODEL
+        request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE = (
+            form_data.web.PERPLEXITY_SEARCH_CONTEXT_USAGE
+        )
         request.app.state.config.SOUGOU_API_SID = form_data.web.SOUGOU_API_SID
         request.app.state.config.SOUGOU_API_SK = form_data.web.SOUGOU_API_SK
 
@@ -917,7 +1020,8 @@ async def update_rag_config(
         "CONTENT_EXTRACTION_ENGINE": request.app.state.config.CONTENT_EXTRACTION_ENGINE,
         "PDF_EXTRACT_IMAGES": request.app.state.config.PDF_EXTRACT_IMAGES,
         "DATALAB_MARKER_API_KEY": request.app.state.config.DATALAB_MARKER_API_KEY,
-        "DATALAB_MARKER_LANGS": request.app.state.config.DATALAB_MARKER_LANGS,
+        "DATALAB_MARKER_API_BASE_URL": request.app.state.config.DATALAB_MARKER_API_BASE_URL,
+        "DATALAB_MARKER_ADDITIONAL_CONFIG": request.app.state.config.DATALAB_MARKER_ADDITIONAL_CONFIG,
         "DATALAB_MARKER_SKIP_CACHE": request.app.state.config.DATALAB_MARKER_SKIP_CACHE,
         "DATALAB_MARKER_FORCE_OCR": request.app.state.config.DATALAB_MARKER_FORCE_OCR,
         "DATALAB_MARKER_PAGINATE": request.app.state.config.DATALAB_MARKER_PAGINATE,
@@ -932,6 +1036,9 @@ async def update_rag_config(
         "DOCLING_OCR_ENGINE": request.app.state.config.DOCLING_OCR_ENGINE,
         "DOCLING_OCR_LANG": request.app.state.config.DOCLING_OCR_LANG,
         "DOCLING_DO_PICTURE_DESCRIPTION": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+        "DOCLING_PICTURE_DESCRIPTION_MODE": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL,
+        "DOCLING_PICTURE_DESCRIPTION_API": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API,
         "DOCUMENT_INTELLIGENCE_ENDPOINT": request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
         "DOCUMENT_INTELLIGENCE_KEY": request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
         "MISTRAL_OCR_API_KEY": request.app.state.config.MISTRAL_OCR_API_KEY,
@@ -947,6 +1054,8 @@ async def update_rag_config(
         # File upload settings
         "FILE_MAX_SIZE": request.app.state.config.FILE_MAX_SIZE,
         "FILE_MAX_COUNT": request.app.state.config.FILE_MAX_COUNT,
+        "FILE_IMAGE_COMPRESSION_WIDTH": request.app.state.config.FILE_IMAGE_COMPRESSION_WIDTH,
+        "FILE_IMAGE_COMPRESSION_HEIGHT": request.app.state.config.FILE_IMAGE_COMPRESSION_HEIGHT,
         "ALLOWED_FILE_EXTENSIONS": request.app.state.config.ALLOWED_FILE_EXTENSIONS,
         # Integration settings
         "ENABLE_GOOGLE_DRIVE_INTEGRATION": request.app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
@@ -985,6 +1094,8 @@ async def update_rag_config(
             "BING_SEARCH_V7_SUBSCRIPTION_KEY": request.app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY,
             "EXA_API_KEY": request.app.state.config.EXA_API_KEY,
             "PERPLEXITY_API_KEY": request.app.state.config.PERPLEXITY_API_KEY,
+            "PERPLEXITY_MODEL": request.app.state.config.PERPLEXITY_MODEL,
+            "PERPLEXITY_SEARCH_CONTEXT_USAGE": request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE,
             "SOUGOU_API_SID": request.app.state.config.SOUGOU_API_SID,
             "SOUGOU_API_SK": request.app.state.config.SOUGOU_API_SK,
             "WEB_LOADER_ENGINE": request.app.state.config.WEB_LOADER_ENGINE,
@@ -1062,6 +1173,7 @@ def save_docs_to_vector_db(
                 chunk_overlap=request.app.state.config.CHUNK_OVERLAP,
                 add_start_index=True,
             )
+            docs = text_splitter.split_documents(docs)
         elif request.app.state.config.TEXT_SPLITTER == "token":
             log.info(
                 f"Using token text splitter: {request.app.state.config.TIKTOKEN_ENCODING_NAME}"
@@ -1074,11 +1186,56 @@ def save_docs_to_vector_db(
                 chunk_overlap=request.app.state.config.CHUNK_OVERLAP,
                 add_start_index=True,
             )
+            docs = text_splitter.split_documents(docs)
+        elif request.app.state.config.TEXT_SPLITTER == "markdown_header":
+            log.info("Using markdown header text splitter")
+
+            # Define headers to split on - covering most common markdown header levels
+            headers_to_split_on = [
+                ("#", "Header 1"),
+                ("##", "Header 2"),
+                ("###", "Header 3"),
+                ("####", "Header 4"),
+                ("#####", "Header 5"),
+                ("######", "Header 6"),
+            ]
+
+            markdown_splitter = MarkdownHeaderTextSplitter(
+                headers_to_split_on=headers_to_split_on,
+                strip_headers=False,  # Keep headers in content for context
+            )
+
+            md_split_docs = []
+            for doc in docs:
+                md_header_splits = markdown_splitter.split_text(doc.page_content)
+                text_splitter = RecursiveCharacterTextSplitter(
+                    chunk_size=request.app.state.config.CHUNK_SIZE,
+                    chunk_overlap=request.app.state.config.CHUNK_OVERLAP,
+                    add_start_index=True,
+                )
+                md_header_splits = text_splitter.split_documents(md_header_splits)
+
+                # Convert back to Document objects, preserving original metadata
+                for split_chunk in md_header_splits:
+                    headings_list = []
+                    # Extract header values in order based on headers_to_split_on
+                    for _, header_meta_key_name in headers_to_split_on:
+                        if header_meta_key_name in split_chunk.metadata:
+                            headings_list.append(
+                                split_chunk.metadata[header_meta_key_name]
+                            )
+
+                    md_split_docs.append(
+                        Document(
+                            page_content=split_chunk.page_content,
+                            metadata={**doc.metadata, "headings": headings_list},
+                        )
+                    )
+
+            docs = md_split_docs
         else:
             raise ValueError(ERROR_MESSAGES.DEFAULT("Invalid text splitter"))
 
-        docs = text_splitter.split_documents(docs)
-
     if len(docs) == 0:
         raise ValueError(ERROR_MESSAGES.EMPTY_CONTENT)
 
@@ -1087,27 +1244,14 @@ def save_docs_to_vector_db(
         {
             **doc.metadata,
             **(metadata if metadata else {}),
-            "embedding_config": json.dumps(
-                {
-                    "engine": request.app.state.config.RAG_EMBEDDING_ENGINE,
-                    "model": request.app.state.config.RAG_EMBEDDING_MODEL,
-                }
-            ),
+            "embedding_config": {
+                "engine": request.app.state.config.RAG_EMBEDDING_ENGINE,
+                "model": request.app.state.config.RAG_EMBEDDING_MODEL,
+            },
         }
         for doc in docs
     ]
 
-    # ChromaDB does not like datetime formats
-    # for meta-data so convert them to string.
-    for metadata in metadatas:
-        for key, value in metadata.items():
-            if (
-                isinstance(value, datetime)
-                or isinstance(value, list)
-                or isinstance(value, dict)
-            ):
-                metadata[key] = str(value)
-
     try:
         if VECTOR_DB_CLIENT.has_collection(collection_name=collection_name):
             log.info(f"collection {collection_name} already exists")
@@ -1129,14 +1273,27 @@ def save_docs_to_vector_db(
             (
                 request.app.state.config.RAG_OPENAI_API_BASE_URL
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_BASE_URL
+                else (
+                    request.app.state.config.RAG_OLLAMA_BASE_URL
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_BASE_URL
+                )
             ),
             (
                 request.app.state.config.RAG_OPENAI_API_KEY
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_API_KEY
+                else (
+                    request.app.state.config.RAG_OLLAMA_API_KEY
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_API_KEY
+                )
             ),
             request.app.state.config.RAG_EMBEDDING_BATCH_SIZE,
+            azure_api_version=(
+                request.app.state.config.RAG_AZURE_OPENAI_API_VERSION
+                if request.app.state.config.RAG_EMBEDDING_ENGINE == "azure_openai"
+                else None
+            ),
         )
 
         embeddings = embedding_function(
@@ -1251,21 +1408,28 @@ def process_file(
                 loader = Loader(
                     engine=request.app.state.config.CONTENT_EXTRACTION_ENGINE,
                     DATALAB_MARKER_API_KEY=request.app.state.config.DATALAB_MARKER_API_KEY,
-                    DATALAB_MARKER_LANGS=request.app.state.config.DATALAB_MARKER_LANGS,
+                    DATALAB_MARKER_API_BASE_URL=request.app.state.config.DATALAB_MARKER_API_BASE_URL,
+                    DATALAB_MARKER_ADDITIONAL_CONFIG=request.app.state.config.DATALAB_MARKER_ADDITIONAL_CONFIG,
                     DATALAB_MARKER_SKIP_CACHE=request.app.state.config.DATALAB_MARKER_SKIP_CACHE,
                     DATALAB_MARKER_FORCE_OCR=request.app.state.config.DATALAB_MARKER_FORCE_OCR,
                     DATALAB_MARKER_PAGINATE=request.app.state.config.DATALAB_MARKER_PAGINATE,
                     DATALAB_MARKER_STRIP_EXISTING_OCR=request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR,
                     DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION=request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+                    DATALAB_MARKER_FORMAT_LINES=request.app.state.config.DATALAB_MARKER_FORMAT_LINES,
                     DATALAB_MARKER_USE_LLM=request.app.state.config.DATALAB_MARKER_USE_LLM,
                     DATALAB_MARKER_OUTPUT_FORMAT=request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT,
                     EXTERNAL_DOCUMENT_LOADER_URL=request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL,
                     EXTERNAL_DOCUMENT_LOADER_API_KEY=request.app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY,
                     TIKA_SERVER_URL=request.app.state.config.TIKA_SERVER_URL,
                     DOCLING_SERVER_URL=request.app.state.config.DOCLING_SERVER_URL,
-                    DOCLING_OCR_ENGINE=request.app.state.config.DOCLING_OCR_ENGINE,
-                    DOCLING_OCR_LANG=request.app.state.config.DOCLING_OCR_LANG,
-                    DOCLING_DO_PICTURE_DESCRIPTION=request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+                    DOCLING_PARAMS={
+                        "ocr_engine": request.app.state.config.DOCLING_OCR_ENGINE,
+                        "ocr_lang": request.app.state.config.DOCLING_OCR_LANG,
+                        "do_picture_description": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+                        "picture_description_mode": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+                        "picture_description_local": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL,
+                        "picture_description_api": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API,
+                    },
                     PDF_EXTRACT_IMAGES=request.app.state.config.PDF_EXTRACT_IMAGES,
                     DOCUMENT_INTELLIGENCE_ENDPOINT=request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
                     DOCUMENT_INTELLIGENCE_KEY=request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
@@ -1625,7 +1789,7 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
                 request.app.state.config.SERPLY_API_KEY,
                 query,
                 request.app.state.config.WEB_SEARCH_RESULT_COUNT,
-                request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
+                filter_list=request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
             )
         else:
             raise Exception("No SERPLY_API_KEY found in environment variables")
@@ -1645,6 +1809,16 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
             )
         else:
             raise Exception("No TAVILY_API_KEY found in environment variables")
+    elif engine == "exa":
+        if request.app.state.config.EXA_API_KEY:
+            return search_exa(
+                request.app.state.config.EXA_API_KEY,
+                query,
+                request.app.state.config.WEB_SEARCH_RESULT_COUNT,
+                request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
+            )
+        else:
+            raise Exception("No EXA_API_KEY found in environment variables")
     elif engine == "searchapi":
         if request.app.state.config.SEARCHAPI_API_KEY:
             return search_searchapi(
@@ -1695,6 +1869,8 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
             query,
             request.app.state.config.WEB_SEARCH_RESULT_COUNT,
             request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
+            model=request.app.state.config.PERPLEXITY_MODEL,
+            search_context_usage=request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE,
         )
     elif engine == "sougou":
         if (
@@ -1774,6 +1950,10 @@ async def process_web_search(
 
     try:
         if request.app.state.config.BYPASS_WEB_SEARCH_WEB_LOADER:
+            search_results = [
+                item for result in search_results for item in result if result
+            ]
+
             docs = [
                 Document(
                     page_content=result.snippet,
@@ -1785,7 +1965,7 @@ async def process_web_search(
                     },
                 )
                 for result in search_results
-                if hasattr(result, "snippet")
+                if hasattr(result, "snippet") and result.snippet is not None
             ]
         else:
             loader = get_web_loader(
@@ -1877,7 +2057,15 @@ def query_doc_handler(
                     query, prefix=prefix, user=user
                 ),
                 k=form_data.k if form_data.k else request.app.state.config.TOP_K,
-                reranking_function=request.app.state.rf,
+                reranking_function=(
+                    (
+                        lambda sentences: request.app.state.RERANKING_FUNCTION(
+                            sentences, user=user
+                        )
+                    )
+                    if request.app.state.RERANKING_FUNCTION
+                    else None
+                ),
                 k_reranker=form_data.k_reranker
                 or request.app.state.config.TOP_K_RERANKER,
                 r=(
@@ -1934,7 +2122,15 @@ def query_collection_handler(
                     query, prefix=prefix, user=user
                 ),
                 k=form_data.k if form_data.k else request.app.state.config.TOP_K,
-                reranking_function=request.app.state.rf,
+                reranking_function=(
+                    (
+                        lambda sentences: request.app.state.RERANKING_FUNCTION(
+                            sentences, user=user
+                        )
+                    )
+                    if request.app.state.RERANKING_FUNCTION
+                    else None
+                ),
                 k_reranker=form_data.k_reranker
                 or request.app.state.config.TOP_K_RERANKER,
                 r=(

backend/open_webui/routers/scim.py

@@ -0,0 +1,926 @@
+"""
+Experimental SCIM 2.0 Implementation for Open WebUI
+Provides System for Cross-domain Identity Management endpoints for users and groups
+
+NOTE: This is an experimental implementation and may not fully comply with SCIM 2.0 standards, and is subject to change.
+"""
+
+import logging
+import uuid
+import time
+from typing import Optional, List, Dict, Any
+from datetime import datetime, timezone
+
+from fastapi import APIRouter, Depends, HTTPException, Request, Query, Header, status
+from fastapi.responses import JSONResponse
+from pydantic import BaseModel, Field, ConfigDict
+
+from open_webui.models.users import Users, UserModel
+from open_webui.models.groups import Groups, GroupModel
+from open_webui.utils.auth import (
+    get_admin_user,
+    get_current_user,
+    decode_token,
+    get_verified_user,
+)
+from open_webui.constants import ERROR_MESSAGES
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["MAIN"])
+
+router = APIRouter()
+
+# SCIM 2.0 Schema URIs
+SCIM_USER_SCHEMA = "urn:ietf:params:scim:schemas:core:2.0:User"
+SCIM_GROUP_SCHEMA = "urn:ietf:params:scim:schemas:core:2.0:Group"
+SCIM_LIST_RESPONSE_SCHEMA = "urn:ietf:params:scim:api:messages:2.0:ListResponse"
+SCIM_ERROR_SCHEMA = "urn:ietf:params:scim:api:messages:2.0:Error"
+
+# SCIM Resource Types
+SCIM_RESOURCE_TYPE_USER = "User"
+SCIM_RESOURCE_TYPE_GROUP = "Group"
+
+
+def scim_error(status_code: int, detail: str, scim_type: Optional[str] = None):
+    """Create a SCIM-compliant error response"""
+    error_body = {
+        "schemas": [SCIM_ERROR_SCHEMA],
+        "status": str(status_code),
+        "detail": detail,
+    }
+
+    if scim_type:
+        error_body["scimType"] = scim_type
+    elif status_code == 404:
+        error_body["scimType"] = "invalidValue"
+    elif status_code == 409:
+        error_body["scimType"] = "uniqueness"
+    elif status_code == 400:
+        error_body["scimType"] = "invalidSyntax"
+
+    return JSONResponse(status_code=status_code, content=error_body)
+
+
+class SCIMError(BaseModel):
+    """SCIM Error Response"""
+
+    schemas: List[str] = [SCIM_ERROR_SCHEMA]
+    status: str
+    scimType: Optional[str] = None
+    detail: Optional[str] = None
+
+
+class SCIMMeta(BaseModel):
+    """SCIM Resource Metadata"""
+
+    resourceType: str
+    created: str
+    lastModified: str
+    location: Optional[str] = None
+    version: Optional[str] = None
+
+
+class SCIMName(BaseModel):
+    """SCIM User Name"""
+
+    formatted: Optional[str] = None
+    familyName: Optional[str] = None
+    givenName: Optional[str] = None
+    middleName: Optional[str] = None
+    honorificPrefix: Optional[str] = None
+    honorificSuffix: Optional[str] = None
+
+
+class SCIMEmail(BaseModel):
+    """SCIM Email"""
+
+    value: str
+    type: Optional[str] = "work"
+    primary: bool = True
+    display: Optional[str] = None
+
+
+class SCIMPhoto(BaseModel):
+    """SCIM Photo"""
+
+    value: str
+    type: Optional[str] = "photo"
+    primary: bool = True
+    display: Optional[str] = None
+
+
+class SCIMGroupMember(BaseModel):
+    """SCIM Group Member"""
+
+    value: str  # User ID
+    ref: Optional[str] = Field(None, alias="$ref")
+    type: Optional[str] = "User"
+    display: Optional[str] = None
+
+
+class SCIMUser(BaseModel):
+    """SCIM User Resource"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    schemas: List[str] = [SCIM_USER_SCHEMA]
+    id: str
+    externalId: Optional[str] = None
+    userName: str
+    name: Optional[SCIMName] = None
+    displayName: str
+    emails: List[SCIMEmail]
+    active: bool = True
+    photos: Optional[List[SCIMPhoto]] = None
+    groups: Optional[List[Dict[str, str]]] = None
+    meta: SCIMMeta
+
+
+class SCIMUserCreateRequest(BaseModel):
+    """SCIM User Create Request"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    schemas: List[str] = [SCIM_USER_SCHEMA]
+    externalId: Optional[str] = None
+    userName: str
+    name: Optional[SCIMName] = None
+    displayName: str
+    emails: List[SCIMEmail]
+    active: bool = True
+    password: Optional[str] = None
+    photos: Optional[List[SCIMPhoto]] = None
+
+
+class SCIMUserUpdateRequest(BaseModel):
+    """SCIM User Update Request"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    schemas: List[str] = [SCIM_USER_SCHEMA]
+    id: Optional[str] = None
+    externalId: Optional[str] = None
+    userName: Optional[str] = None
+    name: Optional[SCIMName] = None
+    displayName: Optional[str] = None
+    emails: Optional[List[SCIMEmail]] = None
+    active: Optional[bool] = None
+    photos: Optional[List[SCIMPhoto]] = None
+
+
+class SCIMGroup(BaseModel):
+    """SCIM Group Resource"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    schemas: List[str] = [SCIM_GROUP_SCHEMA]
+    id: str
+    displayName: str
+    members: Optional[List[SCIMGroupMember]] = []
+    meta: SCIMMeta
+
+
+class SCIMGroupCreateRequest(BaseModel):
+    """SCIM Group Create Request"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    schemas: List[str] = [SCIM_GROUP_SCHEMA]
+    displayName: str
+    members: Optional[List[SCIMGroupMember]] = []
+
+
+class SCIMGroupUpdateRequest(BaseModel):
+    """SCIM Group Update Request"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    schemas: List[str] = [SCIM_GROUP_SCHEMA]
+    displayName: Optional[str] = None
+    members: Optional[List[SCIMGroupMember]] = None
+
+
+class SCIMListResponse(BaseModel):
+    """SCIM List Response"""
+
+    schemas: List[str] = [SCIM_LIST_RESPONSE_SCHEMA]
+    totalResults: int
+    itemsPerPage: int
+    startIndex: int
+    Resources: List[Any]
+
+
+class SCIMPatchOperation(BaseModel):
+    """SCIM Patch Operation"""
+
+    op: str  # "add", "replace", "remove"
+    path: Optional[str] = None
+    value: Optional[Any] = None
+
+
+class SCIMPatchRequest(BaseModel):
+    """SCIM Patch Request"""
+
+    schemas: List[str] = ["urn:ietf:params:scim:api:messages:2.0:PatchOp"]
+    Operations: List[SCIMPatchOperation]
+
+
+def get_scim_auth(
+    request: Request, authorization: Optional[str] = Header(None)
+) -> bool:
+    """
+    Verify SCIM authentication
+    Checks for SCIM-specific bearer token configured in the system
+    """
+    if not authorization:
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail="Authorization header required",
+            headers={"WWW-Authenticate": "Bearer"},
+        )
+
+    try:
+        parts = authorization.split()
+        if len(parts) != 2:
+            raise HTTPException(
+                status_code=status.HTTP_401_UNAUTHORIZED,
+                detail="Invalid authorization format. Expected: Bearer <token>",
+            )
+
+        scheme, token = parts
+        if scheme.lower() != "bearer":
+            raise HTTPException(
+                status_code=status.HTTP_401_UNAUTHORIZED,
+                detail="Invalid authentication scheme",
+            )
+
+        # Check if SCIM is enabled
+        scim_enabled = getattr(request.app.state, "SCIM_ENABLED", False)
+        log.info(
+            f"SCIM auth check - raw SCIM_ENABLED: {scim_enabled}, type: {type(scim_enabled)}"
+        )
+        # Handle both PersistentConfig and direct value
+        if hasattr(scim_enabled, "value"):
+            scim_enabled = scim_enabled.value
+        log.info(f"SCIM enabled status after conversion: {scim_enabled}")
+        if not scim_enabled:
+            raise HTTPException(
+                status_code=status.HTTP_403_FORBIDDEN,
+                detail="SCIM is not enabled",
+            )
+
+        # Verify the SCIM token
+        scim_token = getattr(request.app.state, "SCIM_TOKEN", None)
+        # Handle both PersistentConfig and direct value
+        if hasattr(scim_token, "value"):
+            scim_token = scim_token.value
+        log.debug(f"SCIM token configured: {bool(scim_token)}")
+        if not scim_token or token != scim_token:
+            raise HTTPException(
+                status_code=status.HTTP_401_UNAUTHORIZED,
+                detail="Invalid SCIM token",
+            )
+
+        return True
+    except HTTPException:
+        # Re-raise HTTP exceptions as-is
+        raise
+    except Exception as e:
+        log.error(f"SCIM authentication error: {e}")
+        import traceback
+
+        log.error(f"Traceback: {traceback.format_exc()}")
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail="Authentication failed",
+        )
+
+
+def user_to_scim(user: UserModel, request: Request) -> SCIMUser:
+    """Convert internal User model to SCIM User"""
+    # Parse display name into name components
+    name_parts = user.name.split(" ", 1) if user.name else ["", ""]
+    given_name = name_parts[0] if name_parts else ""
+    family_name = name_parts[1] if len(name_parts) > 1 else ""
+
+    # Get user's groups
+    user_groups = Groups.get_groups_by_member_id(user.id)
+    groups = [
+        {
+            "value": group.id,
+            "display": group.name,
+            "$ref": f"{request.base_url}api/v1/scim/v2/Groups/{group.id}",
+            "type": "direct",
+        }
+        for group in user_groups
+    ]
+
+    return SCIMUser(
+        id=user.id,
+        userName=user.email,
+        name=SCIMName(
+            formatted=user.name,
+            givenName=given_name,
+            familyName=family_name,
+        ),
+        displayName=user.name,
+        emails=[SCIMEmail(value=user.email)],
+        active=user.role != "pending",
+        photos=(
+            [SCIMPhoto(value=user.profile_image_url)]
+            if user.profile_image_url
+            else None
+        ),
+        groups=groups if groups else None,
+        meta=SCIMMeta(
+            resourceType=SCIM_RESOURCE_TYPE_USER,
+            created=datetime.fromtimestamp(
+                user.created_at, tz=timezone.utc
+            ).isoformat(),
+            lastModified=datetime.fromtimestamp(
+                user.updated_at, tz=timezone.utc
+            ).isoformat(),
+            location=f"{request.base_url}api/v1/scim/v2/Users/{user.id}",
+        ),
+    )
+
+
+def group_to_scim(group: GroupModel, request: Request) -> SCIMGroup:
+    """Convert internal Group model to SCIM Group"""
+    members = []
+    for user_id in group.user_ids:
+        user = Users.get_user_by_id(user_id)
+        if user:
+            members.append(
+                SCIMGroupMember(
+                    value=user.id,
+                    ref=f"{request.base_url}api/v1/scim/v2/Users/{user.id}",
+                    display=user.name,
+                )
+            )
+
+    return SCIMGroup(
+        id=group.id,
+        displayName=group.name,
+        members=members,
+        meta=SCIMMeta(
+            resourceType=SCIM_RESOURCE_TYPE_GROUP,
+            created=datetime.fromtimestamp(
+                group.created_at, tz=timezone.utc
+            ).isoformat(),
+            lastModified=datetime.fromtimestamp(
+                group.updated_at, tz=timezone.utc
+            ).isoformat(),
+            location=f"{request.base_url}api/v1/scim/v2/Groups/{group.id}",
+        ),
+    )
+
+
+# SCIM Service Provider Config
+@router.get("/ServiceProviderConfig")
+async def get_service_provider_config():
+    """Get SCIM Service Provider Configuration"""
+    return {
+        "schemas": ["urn:ietf:params:scim:schemas:core:2.0:ServiceProviderConfig"],
+        "patch": {"supported": True},
+        "bulk": {"supported": False, "maxOperations": 1000, "maxPayloadSize": 1048576},
+        "filter": {"supported": True, "maxResults": 200},
+        "changePassword": {"supported": False},
+        "sort": {"supported": False},
+        "etag": {"supported": False},
+        "authenticationSchemes": [
+            {
+                "type": "oauthbearertoken",
+                "name": "OAuth Bearer Token",
+                "description": "Authentication using OAuth 2.0 Bearer Token",
+            }
+        ],
+    }
+
+
+# SCIM Resource Types
+@router.get("/ResourceTypes")
+async def get_resource_types(request: Request):
+    """Get SCIM Resource Types"""
+    return [
+        {
+            "schemas": ["urn:ietf:params:scim:schemas:core:2.0:ResourceType"],
+            "id": "User",
+            "name": "User",
+            "endpoint": "/Users",
+            "schema": SCIM_USER_SCHEMA,
+            "meta": {
+                "location": f"{request.base_url}api/v1/scim/v2/ResourceTypes/User",
+                "resourceType": "ResourceType",
+            },
+        },
+        {
+            "schemas": ["urn:ietf:params:scim:schemas:core:2.0:ResourceType"],
+            "id": "Group",
+            "name": "Group",
+            "endpoint": "/Groups",
+            "schema": SCIM_GROUP_SCHEMA,
+            "meta": {
+                "location": f"{request.base_url}api/v1/scim/v2/ResourceTypes/Group",
+                "resourceType": "ResourceType",
+            },
+        },
+    ]
+
+
+# SCIM Schemas
+@router.get("/Schemas")
+async def get_schemas():
+    """Get SCIM Schemas"""
+    return [
+        {
+            "schemas": ["urn:ietf:params:scim:schemas:core:2.0:Schema"],
+            "id": SCIM_USER_SCHEMA,
+            "name": "User",
+            "description": "User Account",
+            "attributes": [
+                {
+                    "name": "userName",
+                    "type": "string",
+                    "required": True,
+                    "uniqueness": "server",
+                },
+                {"name": "displayName", "type": "string", "required": True},
+                {
+                    "name": "emails",
+                    "type": "complex",
+                    "multiValued": True,
+                    "required": True,
+                },
+                {"name": "active", "type": "boolean", "required": False},
+            ],
+        },
+        {
+            "schemas": ["urn:ietf:params:scim:schemas:core:2.0:Schema"],
+            "id": SCIM_GROUP_SCHEMA,
+            "name": "Group",
+            "description": "Group",
+            "attributes": [
+                {"name": "displayName", "type": "string", "required": True},
+                {
+                    "name": "members",
+                    "type": "complex",
+                    "multiValued": True,
+                    "required": False,
+                },
+            ],
+        },
+    ]
+
+
+# Users endpoints
+@router.get("/Users", response_model=SCIMListResponse)
+async def get_users(
+    request: Request,
+    startIndex: int = Query(1, ge=1),
+    count: int = Query(20, ge=1, le=100),
+    filter: Optional[str] = None,
+    _: bool = Depends(get_scim_auth),
+):
+    """List SCIM Users"""
+    skip = startIndex - 1
+    limit = count
+
+    # Get users from database
+    if filter:
+        # Simple filter parsing - supports userName eq "email"
+        # In production, you'd want a more robust filter parser
+        if "userName eq" in filter:
+            email = filter.split('"')[1]
+            user = Users.get_user_by_email(email)
+            users_list = [user] if user else []
+            total = 1 if user else 0
+        else:
+            response = Users.get_users(skip=skip, limit=limit)
+            users_list = response["users"]
+            total = response["total"]
+    else:
+        response = Users.get_users(skip=skip, limit=limit)
+        users_list = response["users"]
+        total = response["total"]
+
+    # Convert to SCIM format
+    scim_users = [user_to_scim(user, request) for user in users_list]
+
+    return SCIMListResponse(
+        totalResults=total,
+        itemsPerPage=len(scim_users),
+        startIndex=startIndex,
+        Resources=scim_users,
+    )
+
+
+@router.get("/Users/{user_id}", response_model=SCIMUser)
+async def get_user(
+    user_id: str,
+    request: Request,
+    _: bool = Depends(get_scim_auth),
+):
+    """Get SCIM User by ID"""
+    user = Users.get_user_by_id(user_id)
+    if not user:
+        return scim_error(
+            status_code=status.HTTP_404_NOT_FOUND, detail=f"User {user_id} not found"
+        )
+
+    return user_to_scim(user, request)
+
+
+@router.post("/Users", response_model=SCIMUser, status_code=status.HTTP_201_CREATED)
+async def create_user(
+    request: Request,
+    user_data: SCIMUserCreateRequest,
+    _: bool = Depends(get_scim_auth),
+):
+    """Create SCIM User"""
+    # Check if user already exists
+    existing_user = Users.get_user_by_email(user_data.userName)
+    if existing_user:
+        raise HTTPException(
+            status_code=status.HTTP_409_CONFLICT,
+            detail=f"User with email {user_data.userName} already exists",
+        )
+
+    # Create user
+    user_id = str(uuid.uuid4())
+    email = user_data.emails[0].value if user_data.emails else user_data.userName
+
+    # Parse name if provided
+    name = user_data.displayName
+    if user_data.name:
+        if user_data.name.formatted:
+            name = user_data.name.formatted
+        elif user_data.name.givenName or user_data.name.familyName:
+            name = f"{user_data.name.givenName or ''} {user_data.name.familyName or ''}".strip()
+
+    # Get profile image if provided
+    profile_image = "/user.png"
+    if user_data.photos and len(user_data.photos) > 0:
+        profile_image = user_data.photos[0].value
+
+    # Create user
+    new_user = Users.insert_new_user(
+        id=user_id,
+        name=name,
+        email=email,
+        profile_image_url=profile_image,
+        role="user" if user_data.active else "pending",
+    )
+
+    if not new_user:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to create user",
+        )
+
+    return user_to_scim(new_user, request)
+
+
+@router.put("/Users/{user_id}", response_model=SCIMUser)
+async def update_user(
+    user_id: str,
+    request: Request,
+    user_data: SCIMUserUpdateRequest,
+    _: bool = Depends(get_scim_auth),
+):
+    """Update SCIM User (full update)"""
+    user = Users.get_user_by_id(user_id)
+    if not user:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"User {user_id} not found",
+        )
+
+    # Build update dict
+    update_data = {}
+
+    if user_data.userName:
+        update_data["email"] = user_data.userName
+
+    if user_data.displayName:
+        update_data["name"] = user_data.displayName
+    elif user_data.name:
+        if user_data.name.formatted:
+            update_data["name"] = user_data.name.formatted
+        elif user_data.name.givenName or user_data.name.familyName:
+            update_data["name"] = (
+                f"{user_data.name.givenName or ''} {user_data.name.familyName or ''}".strip()
+            )
+
+    if user_data.emails and len(user_data.emails) > 0:
+        update_data["email"] = user_data.emails[0].value
+
+    if user_data.active is not None:
+        update_data["role"] = "user" if user_data.active else "pending"
+
+    if user_data.photos and len(user_data.photos) > 0:
+        update_data["profile_image_url"] = user_data.photos[0].value
+
+    # Update user
+    updated_user = Users.update_user_by_id(user_id, update_data)
+    if not updated_user:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to update user",
+        )
+
+    return user_to_scim(updated_user, request)
+
+
+@router.patch("/Users/{user_id}", response_model=SCIMUser)
+async def patch_user(
+    user_id: str,
+    request: Request,
+    patch_data: SCIMPatchRequest,
+    _: bool = Depends(get_scim_auth),
+):
+    """Update SCIM User (partial update)"""
+    user = Users.get_user_by_id(user_id)
+    if not user:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"User {user_id} not found",
+        )
+
+    update_data = {}
+
+    for operation in patch_data.Operations:
+        op = operation.op.lower()
+        path = operation.path
+        value = operation.value
+
+        if op == "replace":
+            if path == "active":
+                update_data["role"] = "user" if value else "pending"
+            elif path == "userName":
+                update_data["email"] = value
+            elif path == "displayName":
+                update_data["name"] = value
+            elif path == "emails[primary eq true].value":
+                update_data["email"] = value
+            elif path == "name.formatted":
+                update_data["name"] = value
+
+    # Update user
+    if update_data:
+        updated_user = Users.update_user_by_id(user_id, update_data)
+        if not updated_user:
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail="Failed to update user",
+            )
+    else:
+        updated_user = user
+
+    return user_to_scim(updated_user, request)
+
+
+@router.delete("/Users/{user_id}", status_code=status.HTTP_204_NO_CONTENT)
+async def delete_user(
+    user_id: str,
+    request: Request,
+    _: bool = Depends(get_scim_auth),
+):
+    """Delete SCIM User"""
+    user = Users.get_user_by_id(user_id)
+    if not user:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"User {user_id} not found",
+        )
+
+    success = Users.delete_user_by_id(user_id)
+    if not success:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to delete user",
+        )
+
+    return None
+
+
+# Groups endpoints
+@router.get("/Groups", response_model=SCIMListResponse)
+async def get_groups(
+    request: Request,
+    startIndex: int = Query(1, ge=1),
+    count: int = Query(20, ge=1, le=100),
+    filter: Optional[str] = None,
+    _: bool = Depends(get_scim_auth),
+):
+    """List SCIM Groups"""
+    # Get all groups
+    groups_list = Groups.get_groups()
+
+    # Apply pagination
+    total = len(groups_list)
+    start = startIndex - 1
+    end = start + count
+    paginated_groups = groups_list[start:end]
+
+    # Convert to SCIM format
+    scim_groups = [group_to_scim(group, request) for group in paginated_groups]
+
+    return SCIMListResponse(
+        totalResults=total,
+        itemsPerPage=len(scim_groups),
+        startIndex=startIndex,
+        Resources=scim_groups,
+    )
+
+
+@router.get("/Groups/{group_id}", response_model=SCIMGroup)
+async def get_group(
+    group_id: str,
+    request: Request,
+    _: bool = Depends(get_scim_auth),
+):
+    """Get SCIM Group by ID"""
+    group = Groups.get_group_by_id(group_id)
+    if not group:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"Group {group_id} not found",
+        )
+
+    return group_to_scim(group, request)
+
+
+@router.post("/Groups", response_model=SCIMGroup, status_code=status.HTTP_201_CREATED)
+async def create_group(
+    request: Request,
+    group_data: SCIMGroupCreateRequest,
+    _: bool = Depends(get_scim_auth),
+):
+    """Create SCIM Group"""
+    # Extract member IDs
+    member_ids = []
+    if group_data.members:
+        for member in group_data.members:
+            member_ids.append(member.value)
+
+    # Create group
+    from open_webui.models.groups import GroupForm
+
+    form = GroupForm(
+        name=group_data.displayName,
+        description="",
+    )
+
+    # Need to get the creating user's ID - we'll use the first admin
+    admin_user = Users.get_super_admin_user()
+    if not admin_user:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="No admin user found",
+        )
+
+    new_group = Groups.insert_new_group(admin_user.id, form)
+    if not new_group:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to create group",
+        )
+
+    # Add members if provided
+    if member_ids:
+        from open_webui.models.groups import GroupUpdateForm
+
+        update_form = GroupUpdateForm(
+            name=new_group.name,
+            description=new_group.description,
+            user_ids=member_ids,
+        )
+        Groups.update_group_by_id(new_group.id, update_form)
+        new_group = Groups.get_group_by_id(new_group.id)
+
+    return group_to_scim(new_group, request)
+
+
+@router.put("/Groups/{group_id}", response_model=SCIMGroup)
+async def update_group(
+    group_id: str,
+    request: Request,
+    group_data: SCIMGroupUpdateRequest,
+    _: bool = Depends(get_scim_auth),
+):
+    """Update SCIM Group (full update)"""
+    group = Groups.get_group_by_id(group_id)
+    if not group:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"Group {group_id} not found",
+        )
+
+    # Build update form
+    from open_webui.models.groups import GroupUpdateForm
+
+    update_form = GroupUpdateForm(
+        name=group_data.displayName if group_data.displayName else group.name,
+        description=group.description,
+    )
+
+    # Handle members if provided
+    if group_data.members is not None:
+        member_ids = [member.value for member in group_data.members]
+        update_form.user_ids = member_ids
+
+    # Update group
+    updated_group = Groups.update_group_by_id(group_id, update_form)
+    if not updated_group:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to update group",
+        )
+
+    return group_to_scim(updated_group, request)
+
+
+@router.patch("/Groups/{group_id}", response_model=SCIMGroup)
+async def patch_group(
+    group_id: str,
+    request: Request,
+    patch_data: SCIMPatchRequest,
+    _: bool = Depends(get_scim_auth),
+):
+    """Update SCIM Group (partial update)"""
+    group = Groups.get_group_by_id(group_id)
+    if not group:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"Group {group_id} not found",
+        )
+
+    from open_webui.models.groups import GroupUpdateForm
+
+    update_form = GroupUpdateForm(
+        name=group.name,
+        description=group.description,
+        user_ids=group.user_ids.copy() if group.user_ids else [],
+    )
+
+    for operation in patch_data.Operations:
+        op = operation.op.lower()
+        path = operation.path
+        value = operation.value
+
+        if op == "replace":
+            if path == "displayName":
+                update_form.name = value
+            elif path == "members":
+                # Replace all members
+                update_form.user_ids = [member["value"] for member in value]
+        elif op == "add":
+            if path == "members":
+                # Add members
+                if isinstance(value, list):
+                    for member in value:
+                        if isinstance(member, dict) and "value" in member:
+                            if member["value"] not in update_form.user_ids:
+                                update_form.user_ids.append(member["value"])
+        elif op == "remove":
+            if path and path.startswith("members[value eq"):
+                # Remove specific member
+                member_id = path.split('"')[1]
+                if member_id in update_form.user_ids:
+                    update_form.user_ids.remove(member_id)
+
+    # Update group
+    updated_group = Groups.update_group_by_id(group_id, update_form)
+    if not updated_group:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to update group",
+        )
+
+    return group_to_scim(updated_group, request)
+
+
+@router.delete("/Groups/{group_id}", status_code=status.HTTP_204_NO_CONTENT)
+async def delete_group(
+    group_id: str,
+    request: Request,
+    _: bool = Depends(get_scim_auth),
+):
+    """Delete SCIM Group"""
+    group = Groups.get_group_by_id(group_id)
+    if not group:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"Group {group_id} not found",
+        )
+
+    success = Groups.delete_group_by_id(group_id)
+    if not success:
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Failed to delete group",
+        )
+
+    return None

backend/open_webui/routers/tasks.py

@@ -9,6 +9,7 @@ import re
 from open_webui.utils.chat import generate_chat_completion
 from open_webui.utils.task import (
     title_generation_template,
+    follow_up_generation_template,
     query_generation_template,
     image_prompt_generation_template,
     autocomplete_generation_template,
@@ -25,6 +26,7 @@ from open_webui.utils.task import get_task_model_id
 
 from open_webui.config import (
     DEFAULT_TITLE_GENERATION_PROMPT_TEMPLATE,
+    DEFAULT_FOLLOW_UP_GENERATION_PROMPT_TEMPLATE,
     DEFAULT_TAGS_GENERATION_PROMPT_TEMPLATE,
     DEFAULT_IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE,
     DEFAULT_QUERY_GENERATION_PROMPT_TEMPLATE,
@@ -58,6 +60,8 @@ async def get_task_config(request: Request, user=Depends(get_verified_user)):
         "ENABLE_AUTOCOMPLETE_GENERATION": request.app.state.config.ENABLE_AUTOCOMPLETE_GENERATION,
         "AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH": request.app.state.config.AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH,
         "TAGS_GENERATION_PROMPT_TEMPLATE": request.app.state.config.TAGS_GENERATION_PROMPT_TEMPLATE,
+        "FOLLOW_UP_GENERATION_PROMPT_TEMPLATE": request.app.state.config.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE,
+        "ENABLE_FOLLOW_UP_GENERATION": request.app.state.config.ENABLE_FOLLOW_UP_GENERATION,
         "ENABLE_TAGS_GENERATION": request.app.state.config.ENABLE_TAGS_GENERATION,
         "ENABLE_TITLE_GENERATION": request.app.state.config.ENABLE_TITLE_GENERATION,
         "ENABLE_SEARCH_QUERY_GENERATION": request.app.state.config.ENABLE_SEARCH_QUERY_GENERATION,
@@ -76,6 +80,8 @@ class TaskConfigForm(BaseModel):
     ENABLE_AUTOCOMPLETE_GENERATION: bool
     AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH: int
     TAGS_GENERATION_PROMPT_TEMPLATE: str
+    FOLLOW_UP_GENERATION_PROMPT_TEMPLATE: str
+    ENABLE_FOLLOW_UP_GENERATION: bool
     ENABLE_TAGS_GENERATION: bool
     ENABLE_SEARCH_QUERY_GENERATION: bool
     ENABLE_RETRIEVAL_QUERY_GENERATION: bool
@@ -94,6 +100,13 @@ async def update_task_config(
         form_data.TITLE_GENERATION_PROMPT_TEMPLATE
     )
 
+    request.app.state.config.ENABLE_FOLLOW_UP_GENERATION = (
+        form_data.ENABLE_FOLLOW_UP_GENERATION
+    )
+    request.app.state.config.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE = (
+        form_data.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE
+    )
+
     request.app.state.config.IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE = (
         form_data.IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE
     )
@@ -133,6 +146,8 @@ async def update_task_config(
         "AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH": request.app.state.config.AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH,
         "TAGS_GENERATION_PROMPT_TEMPLATE": request.app.state.config.TAGS_GENERATION_PROMPT_TEMPLATE,
         "ENABLE_TAGS_GENERATION": request.app.state.config.ENABLE_TAGS_GENERATION,
+        "ENABLE_FOLLOW_UP_GENERATION": request.app.state.config.ENABLE_FOLLOW_UP_GENERATION,
+        "FOLLOW_UP_GENERATION_PROMPT_TEMPLATE": request.app.state.config.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE,
         "ENABLE_SEARCH_QUERY_GENERATION": request.app.state.config.ENABLE_SEARCH_QUERY_GENERATION,
         "ENABLE_RETRIEVAL_QUERY_GENERATION": request.app.state.config.ENABLE_RETRIEVAL_QUERY_GENERATION,
         "QUERY_GENERATION_PROMPT_TEMPLATE": request.app.state.config.QUERY_GENERATION_PROMPT_TEMPLATE,
@@ -231,6 +246,86 @@ async def generate_title(
         )
 
 
+@router.post("/follow_up/completions")
+async def generate_follow_ups(
+    request: Request, form_data: dict, user=Depends(get_verified_user)
+):
+
+    if not request.app.state.config.ENABLE_FOLLOW_UP_GENERATION:
+        return JSONResponse(
+            status_code=status.HTTP_200_OK,
+            content={"detail": "Follow-up generation is disabled"},
+        )
+
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS
+
+    model_id = form_data["model"]
+    if model_id not in models:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="Model not found",
+        )
+
+    # Check if the user has a custom task model
+    # If the user has a custom task model, use that model
+    task_model_id = get_task_model_id(
+        model_id,
+        request.app.state.config.TASK_MODEL,
+        request.app.state.config.TASK_MODEL_EXTERNAL,
+        models,
+    )
+
+    log.debug(
+        f"generating chat title using model {task_model_id} for user {user.email} "
+    )
+
+    if request.app.state.config.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE != "":
+        template = request.app.state.config.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE
+    else:
+        template = DEFAULT_FOLLOW_UP_GENERATION_PROMPT_TEMPLATE
+
+    content = follow_up_generation_template(
+        template,
+        form_data["messages"],
+        {
+            "name": user.name,
+            "location": user.info.get("location") if user.info else None,
+        },
+    )
+
+    payload = {
+        "model": task_model_id,
+        "messages": [{"role": "user", "content": content}],
+        "stream": False,
+        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
+            "task": str(TASKS.FOLLOW_UP_GENERATION),
+            "task_body": form_data,
+            "chat_id": form_data.get("chat_id", None),
+        },
+    }
+
+    # Process the payload through the pipeline
+    try:
+        payload = await process_pipeline_inlet_filter(request, payload, user, models)
+    except Exception as e:
+        raise e
+
+    try:
+        return await generate_chat_completion(request, form_data=payload, user=user)
+    except Exception as e:
+        log.error("Exception occurred", exc_info=True)
+        return JSONResponse(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            content={"detail": "An internal error has occurred."},
+        )
+
+
 @router.post("/tags/completions")
 async def generate_chat_tags(
     request: Request, form_data: dict, user=Depends(get_verified_user)
@@ -600,11 +695,11 @@ async def generate_emoji(
                 "max_completion_tokens": 4,
             }
         ),
-        "chat_id": form_data.get("chat_id", None),
         "metadata": {
             **(request.state.metadata if hasattr(request.state, "metadata") else {}),
             "task": str(TASKS.EMOJI_GENERATION),
             "task_body": form_data,
+            "chat_id": form_data.get("chat_id", None),
         },
     }
 

backend/open_webui/routers/tools.py

@@ -5,6 +5,8 @@ import time
 import re
 import aiohttp
 from pydantic import BaseModel, HttpUrl
+from fastapi import APIRouter, Depends, HTTPException, Request, status
+
 
 from open_webui.models.tools import (
     ToolForm,
@@ -14,16 +16,15 @@ from open_webui.models.tools import (
     Tools,
 )
 from open_webui.utils.plugin import load_tool_module_by_id, replace_imports
-from open_webui.config import CACHE_DIR
-from open_webui.constants import ERROR_MESSAGES
-from fastapi import APIRouter, Depends, HTTPException, Request, status
 from open_webui.utils.tools import get_tool_specs
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.access_control import has_access, has_permission
-from open_webui.env import SRC_LOG_LEVELS
-
 from open_webui.utils.tools import get_tool_servers_data
 
+from open_webui.env import SRC_LOG_LEVELS
+from open_webui.config import CACHE_DIR, ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS
+from open_webui.constants import ERROR_MESSAGES
+
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
@@ -74,15 +75,17 @@ async def get_tools(request: Request, user=Depends(get_verified_user)):
             )
         )
 
-    if user.role != "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
+        # Admin can see all tools
+        return tools
+    else:
         tools = [
             tool
             for tool in tools
             if tool.user_id == user.id
             or has_access(user.id, "read", tool.access_control)
         ]
-
-    return tools
+        return tools
 
 
 ############################
@@ -92,7 +95,7 @@ async def get_tools(request: Request, user=Depends(get_verified_user)):
 
 @router.get("/list", response_model=list[ToolUserResponse])
 async def get_tool_list(user=Depends(get_verified_user)):
-    if user.role == "admin":
+    if user.role == "admin" and ENABLE_ADMIN_WORKSPACE_CONTENT_ACCESS:
         tools = Tools.get_tools()
     else:
         tools = Tools.get_tools_by_user_id(user.id, "write")
@@ -153,7 +156,7 @@ async def load_tool_from_url(
     )
 
     try:
-        async with aiohttp.ClientSession() as session:
+        async with aiohttp.ClientSession(trust_env=True) as session:
             async with session.get(
                 url, headers={"Content-Type": "application/json"}
             ) as resp:

backend/open_webui/routers/users.py

@@ -1,5 +1,13 @@
 import logging
 from typing import Optional
+import base64
+import io
+
+
+from fastapi import APIRouter, Depends, HTTPException, Request, status
+from fastapi.responses import Response, StreamingResponse, FileResponse
+from pydantic import BaseModel
+
 
 from open_webui.models.auths import Auths
 from open_webui.models.groups import Groups
@@ -7,6 +15,7 @@ from open_webui.models.chats import Chats
 from open_webui.models.users import (
     UserModel,
     UserListResponse,
+    UserInfoListResponse,
     UserRoleUpdateForm,
     Users,
     UserSettings,
@@ -14,11 +23,14 @@ from open_webui.models.users import (
 )
 
 
-from open_webui.socket.main import get_active_status_by_user_id
+from open_webui.socket.main import (
+    get_active_status_by_user_id,
+    get_active_user_ids,
+    get_user_active_status,
+)
 from open_webui.constants import ERROR_MESSAGES
-from open_webui.env import SRC_LOG_LEVELS
-from fastapi import APIRouter, Depends, HTTPException, Request, status
-from pydantic import BaseModel
+from open_webui.env import SRC_LOG_LEVELS, STATIC_DIR
+
 
 from open_webui.utils.auth import get_admin_user, get_password_hash, get_verified_user
 from open_webui.utils.access_control import get_permissions, has_permission
@@ -29,6 +41,24 @@ log.setLevel(SRC_LOG_LEVELS["MODELS"])
 
 router = APIRouter()
 
+
+############################
+# GetActiveUsers
+############################
+
+
+@router.get("/active")
+async def get_active_users(
+    user=Depends(get_verified_user),
+):
+    """
+    Get a list of active users.
+    """
+    return {
+        "user_ids": get_active_user_ids(),
+    }
+
+
 ############################
 # GetUsers
 ############################
@@ -61,7 +91,7 @@ async def get_users(
     return Users.get_users(filter=filter, skip=skip, limit=limit)
 
 
-@router.get("/all", response_model=UserListResponse)
+@router.get("/all", response_model=UserInfoListResponse)
 async def get_all_users(
     user=Depends(get_admin_user),
 ):
@@ -111,6 +141,9 @@ class SharingPermissions(BaseModel):
 
 class ChatPermissions(BaseModel):
     controls: bool = True
+    valves: bool = True
+    system_prompt: bool = True
+    params: bool = True
     file_upload: bool = True
     delete: bool = True
     edit: bool = True
@@ -165,22 +198,6 @@ async def update_default_user_permissions(
     return request.app.state.config.USER_PERMISSIONS
 
 
-############################
-# UpdateUserRole
-############################
-
-
-@router.post("/update/role", response_model=Optional[UserModel])
-async def update_user_role(form_data: UserRoleUpdateForm, user=Depends(get_admin_user)):
-    if user.id != form_data.id and form_data.id != Users.get_first_user().id:
-        return Users.update_user_role_by_id(form_data.id, form_data.role)
-
-    raise HTTPException(
-        status_code=status.HTTP_403_FORBIDDEN,
-        detail=ERROR_MESSAGES.ACTION_PROHIBITED,
-    )
-
-
 ############################
 # GetUserSettingsBySessionUser
 ############################
@@ -319,6 +336,55 @@ async def get_user_by_id(user_id: str, user=Depends(get_verified_user)):
         )
 
 
+############################
+# GetUserProfileImageById
+############################
+
+
+@router.get("/{user_id}/profile/image")
+async def get_user_profile_image_by_id(user_id: str, user=Depends(get_verified_user)):
+    user = Users.get_user_by_id(user_id)
+    if user:
+        if user.profile_image_url:
+            # check if it's url or base64
+            if user.profile_image_url.startswith("http"):
+                return Response(
+                    status_code=status.HTTP_302_FOUND,
+                    headers={"Location": user.profile_image_url},
+                )
+            elif user.profile_image_url.startswith("data:image"):
+                try:
+                    header, base64_data = user.profile_image_url.split(",", 1)
+                    image_data = base64.b64decode(base64_data)
+                    image_buffer = io.BytesIO(image_data)
+
+                    return StreamingResponse(
+                        image_buffer,
+                        media_type="image/png",
+                        headers={"Content-Disposition": "inline; filename=image.png"},
+                    )
+                except Exception as e:
+                    pass
+        return FileResponse(f"{STATIC_DIR}/user.png")
+    else:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.USER_NOT_FOUND,
+        )
+
+
+############################
+# GetUserActiveStatusById
+############################
+
+
+@router.get("/{user_id}/active", response_model=dict)
+async def get_user_active_status_by_id(user_id: str, user=Depends(get_verified_user)):
+    return {
+        "active": get_user_active_status(user_id),
+    }
+
+
 ############################
 # UpdateUserById
 ############################
@@ -333,11 +399,22 @@ async def update_user_by_id(
     # Prevent modification of the primary admin user by other admins
     try:
         first_user = Users.get_first_user()
-        if first_user and user_id == first_user.id and session_user.id != user_id:
-            raise HTTPException(
-                status_code=status.HTTP_403_FORBIDDEN,
-                detail=ERROR_MESSAGES.ACTION_PROHIBITED,
-            )
+        if first_user:
+            if user_id == first_user.id:
+                if session_user.id != user_id:
+                    # If the user trying to update is the primary admin, and they are not the primary admin themselves
+                    raise HTTPException(
+                        status_code=status.HTTP_403_FORBIDDEN,
+                        detail=ERROR_MESSAGES.ACTION_PROHIBITED,
+                    )
+
+                if form_data.role != "admin":
+                    # If the primary admin is trying to change their own role, prevent it
+                    raise HTTPException(
+                        status_code=status.HTTP_403_FORBIDDEN,
+                        detail=ERROR_MESSAGES.ACTION_PROHIBITED,
+                    )
+
     except Exception as e:
         log.error(f"Error checking primary admin status: {e}")
         raise HTTPException(
@@ -365,6 +442,7 @@ async def update_user_by_id(
         updated_user = Users.update_user_by_id(
             user_id,
             {
+                "role": form_data.role,
                 "name": form_data.name,
                 "email": form_data.email.lower(),
                 "profile_image_url": form_data.profile_image_url,

backend/open_webui/routers/utils.py

@@ -33,7 +33,7 @@ class CodeForm(BaseModel):
 
 
 @router.post("/code/format")
-async def format_code(form_data: CodeForm, user=Depends(get_verified_user)):
+async def format_code(form_data: CodeForm, user=Depends(get_admin_user)):
     try:
         formatted_code = black.format_str(form_data.code, mode=black.Mode())
         return {"code": formatted_code}

backend/open_webui/socket/main.py

@@ -1,13 +1,18 @@
 import asyncio
+import random
+
 import socketio
 import logging
 import sys
 import time
+from typing import Dict, Set
 from redis import asyncio as aioredis
+import pycrdt as Y
 
 from open_webui.models.users import Users, UserNameResponse
 from open_webui.models.channels import Channels
 from open_webui.models.chats import Chats
+from open_webui.models.notes import Notes, NoteUpdateForm
 from open_webui.utils.redis import (
     get_sentinels_from_env,
     get_sentinel_url_from_env,
@@ -17,12 +22,18 @@ from open_webui.env import (
     ENABLE_WEBSOCKET_SUPPORT,
     WEBSOCKET_MANAGER,
     WEBSOCKET_REDIS_URL,
+    WEBSOCKET_REDIS_CLUSTER,
     WEBSOCKET_REDIS_LOCK_TIMEOUT,
     WEBSOCKET_SENTINEL_PORT,
     WEBSOCKET_SENTINEL_HOSTS,
+    REDIS_KEY_PREFIX,
 )
 from open_webui.utils.auth import decode_token
-from open_webui.socket.utils import RedisDict, RedisLock
+from open_webui.socket.utils import RedisDict, RedisLock, YdocManager
+from open_webui.tasks import create_task, stop_item_tasks
+from open_webui.utils.redis import get_redis_connection
+from open_webui.utils.access_control import has_access, get_users_with_access
+
 
 from open_webui.env import (
     GLOBAL_LOG_LEVEL,
@@ -35,6 +46,8 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["SOCKET"])
 
 
+REDIS = None
+
 if WEBSOCKET_MANAGER == "redis":
     if WEBSOCKET_SENTINEL_HOSTS:
         mgr = socketio.AsyncRedisManager(
@@ -69,23 +82,35 @@ TIMEOUT_DURATION = 3
 
 if WEBSOCKET_MANAGER == "redis":
     log.debug("Using Redis to manage websockets.")
+    REDIS = get_redis_connection(
+        redis_url=WEBSOCKET_REDIS_URL,
+        redis_sentinels=get_sentinels_from_env(
+            WEBSOCKET_SENTINEL_HOSTS, WEBSOCKET_SENTINEL_PORT
+        ),
+        redis_cluster=WEBSOCKET_REDIS_CLUSTER,
+        async_mode=True,
+    )
+
     redis_sentinels = get_sentinels_from_env(
         WEBSOCKET_SENTINEL_HOSTS, WEBSOCKET_SENTINEL_PORT
     )
     SESSION_POOL = RedisDict(
-        "open-webui:session_pool",
+        f"{REDIS_KEY_PREFIX}:session_pool",
         redis_url=WEBSOCKET_REDIS_URL,
         redis_sentinels=redis_sentinels,
+        redis_cluster=WEBSOCKET_REDIS_CLUSTER,
     )
     USER_POOL = RedisDict(
-        "open-webui:user_pool",
+        f"{REDIS_KEY_PREFIX}:user_pool",
         redis_url=WEBSOCKET_REDIS_URL,
         redis_sentinels=redis_sentinels,
+        redis_cluster=WEBSOCKET_REDIS_CLUSTER,
     )
     USAGE_POOL = RedisDict(
-        "open-webui:usage_pool",
+        f"{REDIS_KEY_PREFIX}:usage_pool",
         redis_url=WEBSOCKET_REDIS_URL,
         redis_sentinels=redis_sentinels,
+        redis_cluster=WEBSOCKET_REDIS_CLUSTER,
     )
 
     clean_up_lock = RedisLock(
@@ -93,6 +118,7 @@ if WEBSOCKET_MANAGER == "redis":
         lock_name="usage_cleanup_lock",
         timeout_secs=WEBSOCKET_REDIS_LOCK_TIMEOUT,
         redis_sentinels=redis_sentinels,
+        redis_cluster=WEBSOCKET_REDIS_CLUSTER,
     )
     aquire_func = clean_up_lock.aquire_lock
     renew_func = clean_up_lock.renew_lock
@@ -101,14 +127,37 @@ else:
     SESSION_POOL = {}
     USER_POOL = {}
     USAGE_POOL = {}
+
     aquire_func = release_func = renew_func = lambda: True
 
 
+YDOC_MANAGER = YdocManager(
+    redis=REDIS,
+    redis_key_prefix=f"{REDIS_KEY_PREFIX}:ydoc:documents",
+)
+
+
 async def periodic_usage_pool_cleanup():
-    if not aquire_func():
-        log.debug("Usage pool cleanup lock already exists. Not running it.")
-        return
-    log.debug("Running periodic_usage_pool_cleanup")
+    max_retries = 2
+    retry_delay = random.uniform(
+        WEBSOCKET_REDIS_LOCK_TIMEOUT / 2, WEBSOCKET_REDIS_LOCK_TIMEOUT
+    )
+    for attempt in range(max_retries + 1):
+        if aquire_func():
+            break
+        else:
+            if attempt < max_retries:
+                log.debug(
+                    f"Cleanup lock already exists. Retry {attempt + 1} after {retry_delay}s..."
+                )
+                await asyncio.sleep(retry_delay)
+            else:
+                log.warning(
+                    "Failed to acquire cleanup lock after retries. Skipping cleanup."
+                )
+                return
+
+    log.debug("Running periodic_cleanup")
     try:
         while True:
             if not renew_func():
@@ -135,11 +184,6 @@ async def periodic_usage_pool_cleanup():
                     USAGE_POOL[model_id] = connections
 
                 send_usage = True
-
-            if send_usage:
-                # Emit updated usage information after cleaning
-                await sio.emit("usage", {"models": get_models_in_use()})
-
             await asyncio.sleep(TIMEOUT_DURATION)
     finally:
         release_func()
@@ -157,6 +201,47 @@ def get_models_in_use():
     return models_in_use
 
 
+def get_active_user_ids():
+    """Get the list of active user IDs."""
+    return list(USER_POOL.keys())
+
+
+def get_user_active_status(user_id):
+    """Check if a user is currently active."""
+    return user_id in USER_POOL
+
+
+def get_user_id_from_session_pool(sid):
+    user = SESSION_POOL.get(sid)
+    if user:
+        return user["id"]
+    return None
+
+
+def get_session_ids_from_room(room):
+    """Get all session IDs from a specific room."""
+    active_session_ids = sio.manager.get_participants(
+        namespace="/",
+        room=room,
+    )
+    return [session_id[0] for session_id in active_session_ids]
+
+
+def get_user_ids_from_room(room):
+    active_session_ids = get_session_ids_from_room(room)
+
+    active_user_ids = list(
+        set([SESSION_POOL.get(session_id)["id"] for session_id in active_session_ids])
+    )
+    return active_user_ids
+
+
+def get_active_status_by_user_id(user_id):
+    if user_id in USER_POOL:
+        return True
+    return False
+
+
 @sio.on("usage")
 async def usage(sid, data):
     if sid in SESSION_POOL:
@@ -170,9 +255,6 @@ async def usage(sid, data):
             sid: {"updated_at": current_time},
         }
 
-        # Broadcast the usage data to all clients
-        await sio.emit("usage", {"models": get_models_in_use()})
-
 
 @sio.event
 async def connect(sid, environ, auth):
@@ -190,10 +272,6 @@ async def connect(sid, environ, auth):
             else:
                 USER_POOL[user.id] = [sid]
 
-            # print(f"user {user.name}({user.id}) connected with session ID {sid}")
-            await sio.emit("user-list", {"user_ids": list(USER_POOL.keys())})
-            await sio.emit("usage", {"models": get_models_in_use()})
-
 
 @sio.on("user-join")
 async def user_join(sid, data):
@@ -221,10 +299,6 @@ async def user_join(sid, data):
     log.debug(f"{channels=}")
     for channel in channels:
         await sio.enter_room(sid, f"channel:{channel.id}")
-
-    # print(f"user {user.name}({user.id}) connected with session ID {sid}")
-
-    await sio.emit("user-list", {"user_ids": list(USER_POOL.keys())})
     return {"id": user.id, "name": user.name}
 
 
@@ -249,6 +323,37 @@ async def join_channel(sid, data):
         await sio.enter_room(sid, f"channel:{channel.id}")
 
 
+@sio.on("join-note")
+async def join_note(sid, data):
+    auth = data["auth"] if "auth" in data else None
+    if not auth or "token" not in auth:
+        return
+
+    token_data = decode_token(auth["token"])
+    if token_data is None or "id" not in token_data:
+        return
+
+    user = Users.get_user_by_id(token_data["id"])
+    if not user:
+        return
+
+    note = Notes.get_note_by_id(data["note_id"])
+    if not note:
+        log.error(f"Note {data['note_id']} not found for user {user.id}")
+        return
+
+    if (
+        user.role != "admin"
+        and user.id != note.user_id
+        and not has_access(user.id, type="read", access_control=note.access_control)
+    ):
+        log.error(f"User {user.id} does not have access to note {data['note_id']}")
+        return
+
+    log.debug(f"Joining note {note.id} for user {user.id}")
+    await sio.enter_room(sid, f"note:{note.id}")
+
+
 @sio.on("channel-events")
 async def channel_events(sid, data):
     room = f"channel:{data['channel_id']}"
@@ -277,10 +382,240 @@ async def channel_events(sid, data):
         )
 
 
-@sio.on("user-list")
-async def user_list(sid):
-    if sid in SESSION_POOL:
-        await sio.emit("user-list", {"user_ids": list(USER_POOL.keys())})
+@sio.on("ydoc:document:join")
+async def ydoc_document_join(sid, data):
+    """Handle user joining a document"""
+    user = SESSION_POOL.get(sid)
+
+    try:
+        document_id = data["document_id"]
+
+        if document_id.startswith("note:"):
+            note_id = document_id.split(":")[1]
+            note = Notes.get_note_by_id(note_id)
+            if not note:
+                log.error(f"Note {note_id} not found")
+                return
+
+            if (
+                user.get("role") != "admin"
+                and user.get("id") != note.user_id
+                and not has_access(
+                    user.get("id"), type="read", access_control=note.access_control
+                )
+            ):
+                log.error(
+                    f"User {user.get('id')} does not have access to note {note_id}"
+                )
+                return
+
+        user_id = data.get("user_id", sid)
+        user_name = data.get("user_name", "Anonymous")
+        user_color = data.get("user_color", "#000000")
+
+        log.info(f"User {user_id} joining document {document_id}")
+        await YDOC_MANAGER.add_user(document_id=document_id, user_id=sid)
+
+        # Join Socket.IO room
+        await sio.enter_room(sid, f"doc_{document_id}")
+
+        active_session_ids = get_session_ids_from_room(f"doc_{document_id}")
+
+        # Get the Yjs document state
+        ydoc = Y.Doc()
+        updates = await YDOC_MANAGER.get_updates(document_id)
+        for update in updates:
+            ydoc.apply_update(bytes(update))
+
+        # Encode the entire document state as an update
+        state_update = ydoc.get_update()
+        await sio.emit(
+            "ydoc:document:state",
+            {
+                "document_id": document_id,
+                "state": list(state_update),  # Convert bytes to list for JSON
+                "sessions": active_session_ids,
+            },
+            room=sid,
+        )
+
+        # Notify other users about the new user
+        await sio.emit(
+            "ydoc:user:joined",
+            {
+                "document_id": document_id,
+                "user_id": user_id,
+                "user_name": user_name,
+                "user_color": user_color,
+            },
+            room=f"doc_{document_id}",
+            skip_sid=sid,
+        )
+
+        log.info(f"User {user_id} successfully joined document {document_id}")
+
+    except Exception as e:
+        log.error(f"Error in yjs_document_join: {e}")
+        await sio.emit("error", {"message": "Failed to join document"}, room=sid)
+
+
+async def document_save_handler(document_id, data, user):
+    if document_id.startswith("note:"):
+        note_id = document_id.split(":")[1]
+        note = Notes.get_note_by_id(note_id)
+        if not note:
+            log.error(f"Note {note_id} not found")
+            return
+
+        if (
+            user.get("role") != "admin"
+            and user.get("id") != note.user_id
+            and not has_access(
+                user.get("id"), type="read", access_control=note.access_control
+            )
+        ):
+            log.error(f"User {user.get('id')} does not have access to note {note_id}")
+            return
+
+        Notes.update_note_by_id(note_id, NoteUpdateForm(data=data))
+
+
+@sio.on("ydoc:document:state")
+async def yjs_document_state(sid, data):
+    """Send the current state of the Yjs document to the user"""
+    try:
+        document_id = data["document_id"]
+        room = f"doc_{document_id}"
+
+        active_session_ids = get_session_ids_from_room(room)
+
+        if sid not in active_session_ids:
+            log.warning(f"Session {sid} not in room {room}. Cannot send state.")
+            return
+
+        if not await YDOC_MANAGER.document_exists(document_id):
+            log.warning(f"Document {document_id} not found")
+            return
+
+        # Get the Yjs document state
+        ydoc = Y.Doc()
+        updates = await YDOC_MANAGER.get_updates(document_id)
+        for update in updates:
+            ydoc.apply_update(bytes(update))
+
+        # Encode the entire document state as an update
+        state_update = ydoc.get_update()
+
+        await sio.emit(
+            "ydoc:document:state",
+            {
+                "document_id": document_id,
+                "state": list(state_update),  # Convert bytes to list for JSON
+                "sessions": active_session_ids,
+            },
+            room=sid,
+        )
+    except Exception as e:
+        log.error(f"Error in yjs_document_state: {e}")
+
+
+@sio.on("ydoc:document:update")
+async def yjs_document_update(sid, data):
+    """Handle Yjs document updates"""
+    try:
+        document_id = data["document_id"]
+
+        try:
+            await stop_item_tasks(REDIS, document_id)
+        except:
+            pass
+
+        user_id = data.get("user_id", sid)
+
+        update = data["update"]  # List of bytes from frontend
+
+        await YDOC_MANAGER.append_to_updates(
+            document_id=document_id,
+            update=update,  # Convert list of bytes to bytes
+        )
+
+        # Broadcast update to all other users in the document
+        await sio.emit(
+            "ydoc:document:update",
+            {
+                "document_id": document_id,
+                "user_id": user_id,
+                "update": update,
+                "socket_id": sid,  # Add socket_id to match frontend filtering
+            },
+            room=f"doc_{document_id}",
+            skip_sid=sid,
+        )
+
+        async def debounced_save():
+            await asyncio.sleep(0.5)
+            await document_save_handler(
+                document_id, data.get("data", {}), SESSION_POOL.get(sid)
+            )
+
+        if data.get("data"):
+            await create_task(REDIS, debounced_save(), document_id)
+
+    except Exception as e:
+        log.error(f"Error in yjs_document_update: {e}")
+
+
+@sio.on("ydoc:document:leave")
+async def yjs_document_leave(sid, data):
+    """Handle user leaving a document"""
+    try:
+        document_id = data["document_id"]
+        user_id = data.get("user_id", sid)
+
+        log.info(f"User {user_id} leaving document {document_id}")
+
+        # Remove user from the document
+        await YDOC_MANAGER.remove_user(document_id=document_id, user_id=sid)
+
+        # Leave Socket.IO room
+        await sio.leave_room(sid, f"doc_{document_id}")
+
+        # Notify other users
+        await sio.emit(
+            "ydoc:user:left",
+            {"document_id": document_id, "user_id": user_id},
+            room=f"doc_{document_id}",
+        )
+
+        if (
+            await YDOC_MANAGER.document_exists(document_id)
+            and len(await YDOC_MANAGER.get_users(document_id)) == 0
+        ):
+            log.info(f"Cleaning up document {document_id} as no users are left")
+            await YDOC_MANAGER.clear_document(document_id)
+
+    except Exception as e:
+        log.error(f"Error in yjs_document_leave: {e}")
+
+
+@sio.on("ydoc:awareness:update")
+async def yjs_awareness_update(sid, data):
+    """Handle awareness updates (cursors, selections, etc.)"""
+    try:
+        document_id = data["document_id"]
+        user_id = data.get("user_id", sid)
+        update = data["update"]
+
+        # Broadcast awareness update to all other users in the document
+        await sio.emit(
+            "ydoc:awareness:update",
+            {"document_id": document_id, "user_id": user_id, "update": update},
+            room=f"doc_{document_id}",
+            skip_sid=sid,
+        )
+
+    except Exception as e:
+        log.error(f"Error in yjs_awareness_update: {e}")
 
 
 @sio.event
@@ -295,7 +630,7 @@ async def disconnect(sid):
         if len(USER_POOL[user_id]) == 0:
             del USER_POOL[user_id]
 
-        await sio.emit("user-list", {"user_ids": list(USER_POOL.keys())})
+        await YDOC_MANAGER.remove_user_from_all_documents(sid)
     else:
         pass
         # print(f"Unknown session ID {sid} disconnected")
@@ -388,30 +723,3 @@ def get_event_call(request_info):
 
 
 get_event_caller = get_event_call
-
-
-def get_user_id_from_session_pool(sid):
-    user = SESSION_POOL.get(sid)
-    if user:
-        return user["id"]
-    return None
-
-
-def get_user_ids_from_room(room):
-    active_session_ids = sio.manager.get_participants(
-        namespace="/",
-        room=room,
-    )
-
-    active_user_ids = list(
-        set(
-            [SESSION_POOL.get(session_id[0])["id"] for session_id in active_session_ids]
-        )
-    )
-    return active_user_ids
-
-
-def get_active_status_by_user_id(user_id):
-    if user_id in USER_POOL:
-        return True
-    return False

backend/open_webui/socket/utils.py

@@ -1,16 +1,30 @@
 import json
 import uuid
 from open_webui.utils.redis import get_redis_connection
+from open_webui.env import REDIS_KEY_PREFIX
+from typing import Optional, List, Tuple
+import pycrdt as Y
 
 
 class RedisLock:
-    def __init__(self, redis_url, lock_name, timeout_secs, redis_sentinels=[]):
+    def __init__(
+        self,
+        redis_url,
+        lock_name,
+        timeout_secs,
+        redis_sentinels=[],
+        redis_cluster=False,
+    ):
+
         self.lock_name = lock_name
         self.lock_id = str(uuid.uuid4())
         self.timeout_secs = timeout_secs
         self.lock_obtained = False
         self.redis = get_redis_connection(
-            redis_url, redis_sentinels, decode_responses=True
+            redis_url,
+            redis_sentinels,
+            redis_cluster=redis_cluster,
+            decode_responses=True,
         )
 
     def aquire_lock(self):
@@ -33,10 +47,13 @@ class RedisLock:
 
 
 class RedisDict:
-    def __init__(self, name, redis_url, redis_sentinels=[]):
+    def __init__(self, name, redis_url, redis_sentinels=[], redis_cluster=False):
         self.name = name
         self.redis = get_redis_connection(
-            redis_url, redis_sentinels, decode_responses=True
+            redis_url,
+            redis_sentinels,
+            redis_cluster=redis_cluster,
+            decode_responses=True,
         )
 
     def __setitem__(self, key, value):
@@ -89,3 +106,109 @@ class RedisDict:
         if key not in self:
             self[key] = default
         return self[key]
+
+
+class YdocManager:
+    def __init__(
+        self,
+        redis=None,
+        redis_key_prefix: str = f"{REDIS_KEY_PREFIX}:ydoc:documents",
+    ):
+        self._updates = {}
+        self._users = {}
+        self._redis = redis
+        self._redis_key_prefix = redis_key_prefix
+
+    async def append_to_updates(self, document_id: str, update: bytes):
+        document_id = document_id.replace(":", "_")
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:updates"
+            await self._redis.rpush(redis_key, json.dumps(list(update)))
+        else:
+            if document_id not in self._updates:
+                self._updates[document_id] = []
+            self._updates[document_id].append(update)
+
+    async def get_updates(self, document_id: str) -> List[bytes]:
+        document_id = document_id.replace(":", "_")
+
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:updates"
+            updates = await self._redis.lrange(redis_key, 0, -1)
+            return [bytes(json.loads(update)) for update in updates]
+        else:
+            return self._updates.get(document_id, [])
+
+    async def document_exists(self, document_id: str) -> bool:
+        document_id = document_id.replace(":", "_")
+
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:updates"
+            return await self._redis.exists(redis_key) > 0
+        else:
+            return document_id in self._updates
+
+    async def get_users(self, document_id: str) -> List[str]:
+        document_id = document_id.replace(":", "_")
+
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:users"
+            users = await self._redis.smembers(redis_key)
+            return list(users)
+        else:
+            return self._users.get(document_id, [])
+
+    async def add_user(self, document_id: str, user_id: str):
+        document_id = document_id.replace(":", "_")
+
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:users"
+            await self._redis.sadd(redis_key, user_id)
+        else:
+            if document_id not in self._users:
+                self._users[document_id] = set()
+            self._users[document_id].add(user_id)
+
+    async def remove_user(self, document_id: str, user_id: str):
+        document_id = document_id.replace(":", "_")
+
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:users"
+            await self._redis.srem(redis_key, user_id)
+        else:
+            if document_id in self._users and user_id in self._users[document_id]:
+                self._users[document_id].remove(user_id)
+
+    async def remove_user_from_all_documents(self, user_id: str):
+        if self._redis:
+            keys = await self._redis.keys(f"{self._redis_key_prefix}:*")
+            for key in keys:
+                if key.endswith(":users"):
+                    await self._redis.srem(key, user_id)
+
+                    document_id = key.split(":")[-2]
+                    if len(await self.get_users(document_id)) == 0:
+                        await self.clear_document(document_id)
+
+        else:
+            for document_id in list(self._users.keys()):
+                if user_id in self._users[document_id]:
+                    self._users[document_id].remove(user_id)
+                    if not self._users[document_id]:
+                        del self._users[document_id]
+
+                        await self.clear_document(document_id)
+
+    async def clear_document(self, document_id: str):
+        document_id = document_id.replace(":", "_")
+
+        if self._redis:
+            redis_key = f"{self._redis_key_prefix}:{document_id}:updates"
+            await self._redis.delete(redis_key)
+            redis_users_key = f"{self._redis_key_prefix}:{document_id}:users"
+            await self._redis.delete(redis_users_key)
+        else:
+            if document_id in self._updates:
+                del self._updates[document_id]
+            if document_id in self._users:
+                del self._users[document_id]

backend/open_webui/tasks.py

@@ -2,26 +2,99 @@
 import asyncio
 from typing import Dict
 from uuid import uuid4
+import json
+import logging
+from redis.asyncio import Redis
+from fastapi import Request
+from typing import Dict, List, Optional
+
+from open_webui.env import SRC_LOG_LEVELS, REDIS_KEY_PREFIX
+
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
 # A dictionary to keep track of active tasks
 tasks: Dict[str, asyncio.Task] = {}
-chat_tasks = {}
+item_tasks = {}
 
 
-def cleanup_task(task_id: str, id=None):
+REDIS_TASKS_KEY = f"{REDIS_KEY_PREFIX}:tasks"
+REDIS_ITEM_TASKS_KEY = f"{REDIS_KEY_PREFIX}:tasks:item"
+REDIS_PUBSUB_CHANNEL = f"{REDIS_KEY_PREFIX}:tasks:commands"
+
+
+async def redis_task_command_listener(app):
+    redis: Redis = app.state.redis
+    pubsub = redis.pubsub()
+    await pubsub.subscribe(REDIS_PUBSUB_CHANNEL)
+
+    async for message in pubsub.listen():
+        if message["type"] != "message":
+            continue
+        try:
+            command = json.loads(message["data"])
+            if command.get("action") == "stop":
+                task_id = command.get("task_id")
+                local_task = tasks.get(task_id)
+                if local_task:
+                    local_task.cancel()
+        except Exception as e:
+            log.exception(f"Error handling distributed task command: {e}")
+
+
+### ------------------------------
+### REDIS-ENABLED HANDLERS
+### ------------------------------
+
+
+async def redis_save_task(redis: Redis, task_id: str, item_id: Optional[str]):
+    pipe = redis.pipeline()
+    pipe.hset(REDIS_TASKS_KEY, task_id, item_id or "")
+    if item_id:
+        pipe.sadd(f"{REDIS_ITEM_TASKS_KEY}:{item_id}", task_id)
+    await pipe.execute()
+
+
+async def redis_cleanup_task(redis: Redis, task_id: str, item_id: Optional[str]):
+    pipe = redis.pipeline()
+    pipe.hdel(REDIS_TASKS_KEY, task_id)
+    if item_id:
+        pipe.srem(f"{REDIS_ITEM_TASKS_KEY}:{item_id}", task_id)
+        if (await pipe.scard(f"{REDIS_ITEM_TASKS_KEY}:{item_id}").execute())[-1] == 0:
+            pipe.delete(f"{REDIS_ITEM_TASKS_KEY}:{item_id}")  # Remove if empty set
+    await pipe.execute()
+
+
+async def redis_list_tasks(redis: Redis) -> List[str]:
+    return list(await redis.hkeys(REDIS_TASKS_KEY))
+
+
+async def redis_list_item_tasks(redis: Redis, item_id: str) -> List[str]:
+    return list(await redis.smembers(f"{REDIS_ITEM_TASKS_KEY}:{item_id}"))
+
+
+async def redis_send_command(redis: Redis, command: dict):
+    await redis.publish(REDIS_PUBSUB_CHANNEL, json.dumps(command))
+
+
+async def cleanup_task(redis, task_id: str, id=None):
     """
     Remove a completed or canceled task from the global `tasks` dictionary.
     """
+    if redis:
+        await redis_cleanup_task(redis, task_id, id)
+
     tasks.pop(task_id, None)  # Remove the task if it exists
 
-    # If an ID is provided, remove the task from the chat_tasks dictionary
-    if id and task_id in chat_tasks.get(id, []):
-        chat_tasks[id].remove(task_id)
-        if not chat_tasks[id]:  # If no tasks left for this ID, remove the entry
-            chat_tasks.pop(id, None)
+    # If an ID is provided, remove the task from the item_tasks dictionary
+    if id and task_id in item_tasks.get(id, []):
+        item_tasks[id].remove(task_id)
+        if not item_tasks[id]:  # If no tasks left for this ID, remove the entry
+            item_tasks.pop(id, None)
 
 
-def create_task(coroutine, id=None):
+async def create_task(redis, coroutine, id=None):
     """
     Create a new asyncio task and add it to the global task dictionary.
     """
@@ -29,44 +102,58 @@ def create_task(coroutine, id=None):
     task = asyncio.create_task(coroutine)  # Create the task
 
     # Add a done callback for cleanup
-    task.add_done_callback(lambda t: cleanup_task(task_id, id))
+    task.add_done_callback(
+        lambda t: asyncio.create_task(cleanup_task(redis, task_id, id))
+    )
     tasks[task_id] = task
 
     # If an ID is provided, associate the task with that ID
-    if chat_tasks.get(id):
-        chat_tasks[id].append(task_id)
+    if item_tasks.get(id):
+        item_tasks[id].append(task_id)
     else:
-        chat_tasks[id] = [task_id]
+        item_tasks[id] = [task_id]
+
+    if redis:
+        await redis_save_task(redis, task_id, id)
 
     return task_id, task
 
 
-def get_task(task_id: str):
-    """
-    Retrieve a task by its task ID.
-    """
-    return tasks.get(task_id)
-
-
-def list_tasks():
+async def list_tasks(redis):
     """
     List all currently active task IDs.
     """
+    if redis:
+        return await redis_list_tasks(redis)
     return list(tasks.keys())
 
 
-def list_task_ids_by_chat_id(id):
+async def list_task_ids_by_item_id(redis, id):
     """
     List all tasks associated with a specific ID.
     """
-    return chat_tasks.get(id, [])
+    if redis:
+        return await redis_list_item_tasks(redis, id)
+    return item_tasks.get(id, [])
 
 
-async def stop_task(task_id: str):
+async def stop_task(redis, task_id: str):
     """
     Cancel a running task and remove it from the global task list.
     """
-    task = tasks.get(task_id)
+    if redis:
+        # PUBSUB: All instances check if they have this task, and stop if so.
+        await redis_send_command(
+            redis,
+            {
+                "action": "stop",
+                "task_id": task_id,
+            },
+        )
+        # Optionally check if task_id still in Redis a few moments later for feedback?
+        return {"status": True, "message": f"Stop signal sent for {task_id}"}
+
+    task = tasks.pop(task_id)
     if not task:
         raise ValueError(f"Task with ID {task_id} not found.")
 
@@ -75,7 +162,22 @@ async def stop_task(task_id: str):
         await task  # Wait for the task to handle the cancellation
     except asyncio.CancelledError:
         # Task successfully canceled
-        tasks.pop(task_id, None)  # Remove it from the dictionary
         return {"status": True, "message": f"Task {task_id} successfully stopped."}
 
     return {"status": False, "message": f"Failed to stop task {task_id}."}
+
+
+async def stop_item_tasks(redis: Redis, item_id: str):
+    """
+    Stop all tasks associated with a specific item ID.
+    """
+    task_ids = await list_task_ids_by_item_id(redis, item_id)
+    if not task_ids:
+        return {"status": True, "message": f"No tasks found for item {item_id}."}
+
+    for task_id in task_ids:
+        result = await stop_task(redis, task_id)
+        if not result["status"]:
+            return result  # Return the first failure
+
+    return {"status": True, "message": f"All tasks for item {item_id} stopped."}

backend/open_webui/test/util/test_redis.py

@@ -0,0 +1,793 @@
+import pytest
+from unittest.mock import Mock, patch, AsyncMock
+import redis
+from open_webui.utils.redis import (
+    SentinelRedisProxy,
+    parse_redis_service_url,
+    get_redis_connection,
+    get_sentinels_from_env,
+    MAX_RETRY_COUNT,
+)
+import inspect
+
+
+class TestSentinelRedisProxy:
+    """Test Redis Sentinel failover functionality"""
+
+    def test_parse_redis_service_url_valid(self):
+        """Test parsing valid Redis service URL"""
+        url = "redis://user:pass@mymaster:6379/0"
+        result = parse_redis_service_url(url)
+
+        assert result["username"] == "user"
+        assert result["password"] == "pass"
+        assert result["service"] == "mymaster"
+        assert result["port"] == 6379
+        assert result["db"] == 0
+
+    def test_parse_redis_service_url_defaults(self):
+        """Test parsing Redis service URL with defaults"""
+        url = "redis://mymaster"
+        result = parse_redis_service_url(url)
+
+        assert result["username"] is None
+        assert result["password"] is None
+        assert result["service"] == "mymaster"
+        assert result["port"] == 6379
+        assert result["db"] == 0
+
+    def test_parse_redis_service_url_invalid_scheme(self):
+        """Test parsing invalid URL scheme"""
+        with pytest.raises(ValueError, match="Invalid Redis URL scheme"):
+            parse_redis_service_url("http://invalid")
+
+    def test_get_sentinels_from_env(self):
+        """Test parsing sentinel hosts from environment"""
+        hosts = "sentinel1,sentinel2,sentinel3"
+        port = "26379"
+
+        result = get_sentinels_from_env(hosts, port)
+        expected = [("sentinel1", 26379), ("sentinel2", 26379), ("sentinel3", 26379)]
+
+        assert result == expected
+
+    def test_get_sentinels_from_env_empty(self):
+        """Test empty sentinel hosts"""
+        result = get_sentinels_from_env(None, "26379")
+        assert result == []
+
+    @patch("redis.sentinel.Sentinel")
+    def test_sentinel_redis_proxy_sync_success(self, mock_sentinel_class):
+        """Test successful sync operation with SentinelRedisProxy"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_master.get.return_value = "test_value"
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test attribute access
+        get_method = proxy.__getattr__("get")
+        result = get_method("test_key")
+
+        assert result == "test_value"
+        mock_sentinel.master_for.assert_called_with("mymaster")
+        mock_master.get.assert_called_with("test_key")
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_sentinel_redis_proxy_async_success(self, mock_sentinel_class):
+        """Test successful async operation with SentinelRedisProxy"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_master.get = AsyncMock(return_value="test_value")
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test async attribute access
+        get_method = proxy.__getattr__("get")
+        result = await get_method("test_key")
+
+        assert result == "test_value"
+        mock_sentinel.master_for.assert_called_with("mymaster")
+        mock_master.get.assert_called_with("test_key")
+
+    @patch("redis.sentinel.Sentinel")
+    def test_sentinel_redis_proxy_failover_retry(self, mock_sentinel_class):
+        """Test retry mechanism during failover"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # First call fails, second succeeds
+        mock_master.get.side_effect = [
+            redis.exceptions.ConnectionError("Master down"),
+            "test_value",
+        ]
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        get_method = proxy.__getattr__("get")
+        result = get_method("test_key")
+
+        assert result == "test_value"
+        assert mock_master.get.call_count == 2
+
+    @patch("redis.sentinel.Sentinel")
+    def test_sentinel_redis_proxy_max_retries_exceeded(self, mock_sentinel_class):
+        """Test failure after max retries exceeded"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # All calls fail
+        mock_master.get.side_effect = redis.exceptions.ConnectionError("Master down")
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        get_method = proxy.__getattr__("get")
+
+        with pytest.raises(redis.exceptions.ConnectionError):
+            get_method("test_key")
+
+        assert mock_master.get.call_count == MAX_RETRY_COUNT
+
+    @patch("redis.sentinel.Sentinel")
+    def test_sentinel_redis_proxy_readonly_error_retry(self, mock_sentinel_class):
+        """Test retry on ReadOnlyError"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # First call gets ReadOnlyError (old master), second succeeds (new master)
+        mock_master.get.side_effect = [
+            redis.exceptions.ReadOnlyError("Read only"),
+            "test_value",
+        ]
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        get_method = proxy.__getattr__("get")
+        result = get_method("test_key")
+
+        assert result == "test_value"
+        assert mock_master.get.call_count == 2
+
+    @patch("redis.sentinel.Sentinel")
+    def test_sentinel_redis_proxy_factory_methods(self, mock_sentinel_class):
+        """Test factory methods are passed through directly"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_pipeline = Mock()
+        mock_master.pipeline.return_value = mock_pipeline
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Factory methods should be passed through without wrapping
+        pipeline_method = proxy.__getattr__("pipeline")
+        result = pipeline_method()
+
+        assert result == mock_pipeline
+        mock_master.pipeline.assert_called_once()
+
+    @patch("redis.sentinel.Sentinel")
+    @patch("redis.from_url")
+    def test_get_redis_connection_with_sentinel(
+        self, mock_from_url, mock_sentinel_class
+    ):
+        """Test getting Redis connection with Sentinel"""
+        mock_sentinel = Mock()
+        mock_sentinel_class.return_value = mock_sentinel
+
+        sentinels = [("sentinel1", 26379), ("sentinel2", 26379)]
+        redis_url = "redis://user:pass@mymaster:6379/0"
+
+        result = get_redis_connection(
+            redis_url=redis_url, redis_sentinels=sentinels, async_mode=False
+        )
+
+        assert isinstance(result, SentinelRedisProxy)
+        mock_sentinel_class.assert_called_once()
+        mock_from_url.assert_not_called()
+
+    @patch("redis.Redis.from_url")
+    def test_get_redis_connection_without_sentinel(self, mock_from_url):
+        """Test getting Redis connection without Sentinel"""
+        mock_redis = Mock()
+        mock_from_url.return_value = mock_redis
+
+        redis_url = "redis://localhost:6379/0"
+
+        result = get_redis_connection(
+            redis_url=redis_url, redis_sentinels=None, async_mode=False
+        )
+
+        assert result == mock_redis
+        mock_from_url.assert_called_once_with(redis_url, decode_responses=True)
+
+    @patch("redis.asyncio.from_url")
+    def test_get_redis_connection_without_sentinel_async(self, mock_from_url):
+        """Test getting async Redis connection without Sentinel"""
+        mock_redis = Mock()
+        mock_from_url.return_value = mock_redis
+
+        redis_url = "redis://localhost:6379/0"
+
+        result = get_redis_connection(
+            redis_url=redis_url, redis_sentinels=None, async_mode=True
+        )
+
+        assert result == mock_redis
+        mock_from_url.assert_called_once_with(redis_url, decode_responses=True)
+
+
+class TestSentinelRedisProxyCommands:
+    """Test Redis commands through SentinelRedisProxy"""
+
+    @patch("redis.sentinel.Sentinel")
+    def test_hash_commands_sync(self, mock_sentinel_class):
+        """Test Redis hash commands in sync mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock hash command responses
+        mock_master.hset.return_value = 1
+        mock_master.hget.return_value = "test_value"
+        mock_master.hgetall.return_value = {"key1": "value1", "key2": "value2"}
+        mock_master.hdel.return_value = 1
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test hset
+        hset_method = proxy.__getattr__("hset")
+        result = hset_method("test_hash", "field1", "value1")
+        assert result == 1
+        mock_master.hset.assert_called_with("test_hash", "field1", "value1")
+
+        # Test hget
+        hget_method = proxy.__getattr__("hget")
+        result = hget_method("test_hash", "field1")
+        assert result == "test_value"
+        mock_master.hget.assert_called_with("test_hash", "field1")
+
+        # Test hgetall
+        hgetall_method = proxy.__getattr__("hgetall")
+        result = hgetall_method("test_hash")
+        assert result == {"key1": "value1", "key2": "value2"}
+        mock_master.hgetall.assert_called_with("test_hash")
+
+        # Test hdel
+        hdel_method = proxy.__getattr__("hdel")
+        result = hdel_method("test_hash", "field1")
+        assert result == 1
+        mock_master.hdel.assert_called_with("test_hash", "field1")
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_hash_commands_async(self, mock_sentinel_class):
+        """Test Redis hash commands in async mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock async hash command responses
+        mock_master.hset = AsyncMock(return_value=1)
+        mock_master.hget = AsyncMock(return_value="test_value")
+        mock_master.hgetall = AsyncMock(
+            return_value={"key1": "value1", "key2": "value2"}
+        )
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test hset
+        hset_method = proxy.__getattr__("hset")
+        result = await hset_method("test_hash", "field1", "value1")
+        assert result == 1
+        mock_master.hset.assert_called_with("test_hash", "field1", "value1")
+
+        # Test hget
+        hget_method = proxy.__getattr__("hget")
+        result = await hget_method("test_hash", "field1")
+        assert result == "test_value"
+        mock_master.hget.assert_called_with("test_hash", "field1")
+
+        # Test hgetall
+        hgetall_method = proxy.__getattr__("hgetall")
+        result = await hgetall_method("test_hash")
+        assert result == {"key1": "value1", "key2": "value2"}
+        mock_master.hgetall.assert_called_with("test_hash")
+
+    @patch("redis.sentinel.Sentinel")
+    def test_string_commands_sync(self, mock_sentinel_class):
+        """Test Redis string commands in sync mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock string command responses
+        mock_master.set.return_value = True
+        mock_master.get.return_value = "test_value"
+        mock_master.delete.return_value = 1
+        mock_master.exists.return_value = True
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test set
+        set_method = proxy.__getattr__("set")
+        result = set_method("test_key", "test_value")
+        assert result is True
+        mock_master.set.assert_called_with("test_key", "test_value")
+
+        # Test get
+        get_method = proxy.__getattr__("get")
+        result = get_method("test_key")
+        assert result == "test_value"
+        mock_master.get.assert_called_with("test_key")
+
+        # Test delete
+        delete_method = proxy.__getattr__("delete")
+        result = delete_method("test_key")
+        assert result == 1
+        mock_master.delete.assert_called_with("test_key")
+
+        # Test exists
+        exists_method = proxy.__getattr__("exists")
+        result = exists_method("test_key")
+        assert result is True
+        mock_master.exists.assert_called_with("test_key")
+
+    @patch("redis.sentinel.Sentinel")
+    def test_list_commands_sync(self, mock_sentinel_class):
+        """Test Redis list commands in sync mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock list command responses
+        mock_master.lpush.return_value = 1
+        mock_master.rpop.return_value = "test_value"
+        mock_master.llen.return_value = 5
+        mock_master.lrange.return_value = ["item1", "item2", "item3"]
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test lpush
+        lpush_method = proxy.__getattr__("lpush")
+        result = lpush_method("test_list", "item1")
+        assert result == 1
+        mock_master.lpush.assert_called_with("test_list", "item1")
+
+        # Test rpop
+        rpop_method = proxy.__getattr__("rpop")
+        result = rpop_method("test_list")
+        assert result == "test_value"
+        mock_master.rpop.assert_called_with("test_list")
+
+        # Test llen
+        llen_method = proxy.__getattr__("llen")
+        result = llen_method("test_list")
+        assert result == 5
+        mock_master.llen.assert_called_with("test_list")
+
+        # Test lrange
+        lrange_method = proxy.__getattr__("lrange")
+        result = lrange_method("test_list", 0, -1)
+        assert result == ["item1", "item2", "item3"]
+        mock_master.lrange.assert_called_with("test_list", 0, -1)
+
+    @patch("redis.sentinel.Sentinel")
+    def test_pubsub_commands_sync(self, mock_sentinel_class):
+        """Test Redis pubsub commands in sync mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_pubsub = Mock()
+
+        # Mock pubsub responses
+        mock_master.pubsub.return_value = mock_pubsub
+        mock_master.publish.return_value = 1
+        mock_pubsub.subscribe.return_value = None
+        mock_pubsub.get_message.return_value = {"type": "message", "data": "test_data"}
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test pubsub (factory method - should pass through)
+        pubsub_method = proxy.__getattr__("pubsub")
+        result = pubsub_method()
+        assert result == mock_pubsub
+        mock_master.pubsub.assert_called_once()
+
+        # Test publish
+        publish_method = proxy.__getattr__("publish")
+        result = publish_method("test_channel", "test_message")
+        assert result == 1
+        mock_master.publish.assert_called_with("test_channel", "test_message")
+
+    @patch("redis.sentinel.Sentinel")
+    def test_pipeline_commands_sync(self, mock_sentinel_class):
+        """Test Redis pipeline commands in sync mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_pipeline = Mock()
+
+        # Mock pipeline responses
+        mock_master.pipeline.return_value = mock_pipeline
+        mock_pipeline.set.return_value = mock_pipeline
+        mock_pipeline.get.return_value = mock_pipeline
+        mock_pipeline.execute.return_value = [True, "test_value"]
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test pipeline (factory method - should pass through)
+        pipeline_method = proxy.__getattr__("pipeline")
+        result = pipeline_method()
+        assert result == mock_pipeline
+        mock_master.pipeline.assert_called_once()
+
+    @patch("redis.sentinel.Sentinel")
+    def test_commands_with_failover_retry(self, mock_sentinel_class):
+        """Test Redis commands with failover retry mechanism"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # First call fails with connection error, second succeeds
+        mock_master.hget.side_effect = [
+            redis.exceptions.ConnectionError("Connection failed"),
+            "recovered_value",
+        ]
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test hget with retry
+        hget_method = proxy.__getattr__("hget")
+        result = hget_method("test_hash", "field1")
+
+        assert result == "recovered_value"
+        assert mock_master.hget.call_count == 2
+
+        # Verify both calls were made with same parameters
+        expected_calls = [(("test_hash", "field1"),), (("test_hash", "field1"),)]
+        actual_calls = [call.args for call in mock_master.hget.call_args_list]
+        assert actual_calls == expected_calls
+
+    @patch("redis.sentinel.Sentinel")
+    def test_commands_with_readonly_error_retry(self, mock_sentinel_class):
+        """Test Redis commands with ReadOnlyError retry mechanism"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # First call fails with ReadOnlyError, second succeeds
+        mock_master.hset.side_effect = [
+            redis.exceptions.ReadOnlyError(
+                "READONLY You can't write against a read only replica"
+            ),
+            1,
+        ]
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=False)
+
+        # Test hset with retry
+        hset_method = proxy.__getattr__("hset")
+        result = hset_method("test_hash", "field1", "value1")
+
+        assert result == 1
+        assert mock_master.hset.call_count == 2
+
+        # Verify both calls were made with same parameters
+        expected_calls = [
+            (("test_hash", "field1", "value1"),),
+            (("test_hash", "field1", "value1"),),
+        ]
+        actual_calls = [call.args for call in mock_master.hset.call_args_list]
+        assert actual_calls == expected_calls
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_async_commands_with_failover_retry(self, mock_sentinel_class):
+        """Test async Redis commands with failover retry mechanism"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # First call fails with connection error, second succeeds
+        mock_master.hget = AsyncMock(
+            side_effect=[
+                redis.exceptions.ConnectionError("Connection failed"),
+                "recovered_value",
+            ]
+        )
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test async hget with retry
+        hget_method = proxy.__getattr__("hget")
+        result = await hget_method("test_hash", "field1")
+
+        assert result == "recovered_value"
+        assert mock_master.hget.call_count == 2
+
+        # Verify both calls were made with same parameters
+        expected_calls = [(("test_hash", "field1"),), (("test_hash", "field1"),)]
+        actual_calls = [call.args for call in mock_master.hget.call_args_list]
+        assert actual_calls == expected_calls
+
+
+class TestSentinelRedisProxyFactoryMethods:
+    """Test Redis factory methods in async mode - these are special cases that remain sync"""
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_pubsub_factory_method_async(self, mock_sentinel_class):
+        """Test pubsub factory method in async mode - should pass through without wrapping"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_pubsub = Mock()
+
+        # Mock pubsub factory method
+        mock_master.pubsub.return_value = mock_pubsub
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test pubsub factory method - should NOT be wrapped as async
+        pubsub_method = proxy.__getattr__("pubsub")
+        result = pubsub_method()
+
+        assert result == mock_pubsub
+        mock_master.pubsub.assert_called_once()
+
+        # Verify it's not wrapped as async (no await needed)
+        assert not inspect.iscoroutine(result)
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_pipeline_factory_method_async(self, mock_sentinel_class):
+        """Test pipeline factory method in async mode - should pass through without wrapping"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_pipeline = Mock()
+
+        # Mock pipeline factory method
+        mock_master.pipeline.return_value = mock_pipeline
+        mock_pipeline.set.return_value = mock_pipeline
+        mock_pipeline.get.return_value = mock_pipeline
+        mock_pipeline.execute.return_value = [True, "test_value"]
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test pipeline factory method - should NOT be wrapped as async
+        pipeline_method = proxy.__getattr__("pipeline")
+        result = pipeline_method()
+
+        assert result == mock_pipeline
+        mock_master.pipeline.assert_called_once()
+
+        # Verify it's not wrapped as async (no await needed)
+        assert not inspect.iscoroutine(result)
+
+        # Test pipeline usage (these should also be sync)
+        pipeline_result = result.set("key", "value").get("key").execute()
+        assert pipeline_result == [True, "test_value"]
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_factory_methods_vs_regular_commands_async(self, mock_sentinel_class):
+        """Test that factory methods behave differently from regular commands in async mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock both factory method and regular command
+        mock_pubsub = Mock()
+        mock_master.pubsub.return_value = mock_pubsub
+        mock_master.get = AsyncMock(return_value="test_value")
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test factory method - should NOT be wrapped
+        pubsub_method = proxy.__getattr__("pubsub")
+        pubsub_result = pubsub_method()
+
+        # Test regular command - should be wrapped as async
+        get_method = proxy.__getattr__("get")
+        get_result = get_method("test_key")
+
+        # Factory method returns directly
+        assert pubsub_result == mock_pubsub
+        assert not inspect.iscoroutine(pubsub_result)
+
+        # Regular command returns coroutine
+        assert inspect.iscoroutine(get_result)
+
+        # Regular command needs await
+        actual_value = await get_result
+        assert actual_value == "test_value"
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_factory_methods_with_failover_async(self, mock_sentinel_class):
+        """Test factory methods with failover in async mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # First call fails, second succeeds
+        mock_pubsub = Mock()
+        mock_master.pubsub.side_effect = [
+            redis.exceptions.ConnectionError("Connection failed"),
+            mock_pubsub,
+        ]
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test pubsub factory method with failover
+        pubsub_method = proxy.__getattr__("pubsub")
+        result = pubsub_method()
+
+        assert result == mock_pubsub
+        assert mock_master.pubsub.call_count == 2  # Retry happened
+
+        # Verify it's still not wrapped as async after retry
+        assert not inspect.iscoroutine(result)
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_monitor_factory_method_async(self, mock_sentinel_class):
+        """Test monitor factory method in async mode - should pass through without wrapping"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_monitor = Mock()
+
+        # Mock monitor factory method
+        mock_master.monitor.return_value = mock_monitor
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test monitor factory method - should NOT be wrapped as async
+        monitor_method = proxy.__getattr__("monitor")
+        result = monitor_method()
+
+        assert result == mock_monitor
+        mock_master.monitor.assert_called_once()
+
+        # Verify it's not wrapped as async (no await needed)
+        assert not inspect.iscoroutine(result)
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_client_factory_method_async(self, mock_sentinel_class):
+        """Test client factory method in async mode - should pass through without wrapping"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_client = Mock()
+
+        # Mock client factory method
+        mock_master.client.return_value = mock_client
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test client factory method - should NOT be wrapped as async
+        client_method = proxy.__getattr__("client")
+        result = client_method()
+
+        assert result == mock_client
+        mock_master.client.assert_called_once()
+
+        # Verify it's not wrapped as async (no await needed)
+        assert not inspect.iscoroutine(result)
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_transaction_factory_method_async(self, mock_sentinel_class):
+        """Test transaction factory method in async mode - should pass through without wrapping"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+        mock_transaction = Mock()
+
+        # Mock transaction factory method
+        mock_master.transaction.return_value = mock_transaction
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test transaction factory method - should NOT be wrapped as async
+        transaction_method = proxy.__getattr__("transaction")
+        result = transaction_method()
+
+        assert result == mock_transaction
+        mock_master.transaction.assert_called_once()
+
+        # Verify it's not wrapped as async (no await needed)
+        assert not inspect.iscoroutine(result)
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_all_factory_methods_async(self, mock_sentinel_class):
+        """Test all factory methods in async mode - comprehensive test"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock all factory methods
+        mock_objects = {
+            "pipeline": Mock(),
+            "pubsub": Mock(),
+            "monitor": Mock(),
+            "client": Mock(),
+            "transaction": Mock(),
+        }
+
+        for method_name, mock_obj in mock_objects.items():
+            setattr(mock_master, method_name, Mock(return_value=mock_obj))
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Test all factory methods
+        for method_name, expected_obj in mock_objects.items():
+            method = proxy.__getattr__(method_name)
+            result = method()
+
+            assert result == expected_obj
+            assert not inspect.iscoroutine(result)
+            getattr(mock_master, method_name).assert_called_once()
+
+            # Reset mock for next iteration
+            getattr(mock_master, method_name).reset_mock()
+
+    @patch("redis.sentinel.Sentinel")
+    @pytest.mark.asyncio
+    async def test_mixed_factory_and_regular_commands_async(self, mock_sentinel_class):
+        """Test using both factory methods and regular commands in async mode"""
+        mock_sentinel = Mock()
+        mock_master = Mock()
+
+        # Mock pipeline factory and regular commands
+        mock_pipeline = Mock()
+        mock_master.pipeline.return_value = mock_pipeline
+        mock_pipeline.set.return_value = mock_pipeline
+        mock_pipeline.get.return_value = mock_pipeline
+        mock_pipeline.execute.return_value = [True, "pipeline_value"]
+
+        mock_master.get = AsyncMock(return_value="regular_value")
+
+        mock_sentinel.master_for.return_value = mock_master
+
+        proxy = SentinelRedisProxy(mock_sentinel, "mymaster", async_mode=True)
+
+        # Use factory method (sync)
+        pipeline = proxy.__getattr__("pipeline")()
+        pipeline_result = pipeline.set("key1", "value1").get("key1").execute()
+
+        # Use regular command (async)
+        get_method = proxy.__getattr__("get")
+        regular_result = await get_method("key2")
+
+        # Verify both work correctly
+        assert pipeline_result == [True, "pipeline_value"]
+        assert regular_result == "regular_value"
+
+        # Verify calls
+        mock_master.pipeline.assert_called_once()
+        mock_master.get.assert_called_with("key2")

backend/open_webui/utils/access_control.py

@@ -60,7 +60,7 @@ def get_permissions(
 
     # Combine permissions from all user groups
     for group in user_groups:
-        group_permissions = group.permissions
+        group_permissions = group.permissions or {}
         permissions = combine_permissions(permissions, group_permissions)
 
     # Ensure all fields from default_permissions are present and filled in

backend/open_webui/utils/auth.py

@@ -8,6 +8,12 @@ import requests
 import os
 
 
+from cryptography.hazmat.primitives.ciphers.aead import AESGCM
+from cryptography.hazmat.primitives.asymmetric import ed25519
+from cryptography.hazmat.primitives import serialization
+import json
+
+
 from datetime import datetime, timedelta
 import pytz
 from pytz import UTC
@@ -18,11 +24,16 @@ from opentelemetry import trace
 from open_webui.models.users import Users
 
 from open_webui.constants import ERROR_MESSAGES
+
 from open_webui.env import (
+    OFFLINE_MODE,
+    LICENSE_BLOB,
+    pk,
     WEBUI_SECRET_KEY,
     TRUSTED_SIGNATURE_KEY,
     STATIC_DIR,
     SRC_LOG_LEVELS,
+    WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
 )
 
 from fastapi import BackgroundTasks, Depends, HTTPException, Request, Response, status
@@ -73,33 +84,73 @@ def override_static(path: str, content: str):
 
 
 def get_license_data(app, key):
-    if key:
-        try:
-            res = requests.post(
-                "https://api.openwebui.com/api/v1/license/",
-                json={"key": key, "version": "1"},
-                timeout=5,
+    def data_handler(data):
+        for k, v in data.items():
+            if k == "resources":
+                for p, c in v.items():
+                    globals().get("override_static", lambda a, b: None)(p, c)
+            elif k == "count":
+                setattr(app.state, "USER_COUNT", v)
+            elif k == "name":
+                setattr(app.state, "WEBUI_NAME", v)
+            elif k == "metadata":
+                setattr(app.state, "LICENSE_METADATA", v)
+
+    def handler(u):
+        res = requests.post(
+            f"{u}/api/v1/license/",
+            json={"key": key, "version": "1"},
+            timeout=5,
+        )
+
+        if getattr(res, "ok", False):
+            payload = getattr(res, "json", lambda: {})()
+            data_handler(payload)
+            return True
+        else:
+            log.error(
+                f"License: retrieval issue: {getattr(res, 'text', 'unknown error')}"
             )
 
-            if getattr(res, "ok", False):
-                payload = getattr(res, "json", lambda: {})()
-                for k, v in payload.items():
-                    if k == "resources":
-                        for p, c in v.items():
-                            globals().get("override_static", lambda a, b: None)(p, c)
-                    elif k == "count":
-                        setattr(app.state, "USER_COUNT", v)
-                    elif k == "name":
-                        setattr(app.state, "WEBUI_NAME", v)
-                    elif k == "metadata":
-                        setattr(app.state, "LICENSE_METADATA", v)
-                return True
-            else:
-                log.error(
-                    f"License: retrieval issue: {getattr(res, 'text', 'unknown error')}"
-                )
+    if key:
+        us = [
+            "https://api.openwebui.com",
+            "https://licenses.api.openwebui.com",
+        ]
+        try:
+            for u in us:
+                if handler(u):
+                    return True
         except Exception as ex:
             log.exception(f"License: Uncaught Exception: {ex}")
+
+    try:
+        if LICENSE_BLOB:
+            nl = 12
+            kb = hashlib.sha256((key.replace("-", "").upper()).encode()).digest()
+
+            def nt(b):
+                return b[:nl], b[nl:]
+
+            lb = base64.b64decode(LICENSE_BLOB)
+            ln, lt = nt(lb)
+
+            aesgcm = AESGCM(kb)
+            p = json.loads(aesgcm.decrypt(ln, lt, None))
+            pk.verify(base64.b64decode(p["s"]), p["p"].encode())
+
+            pb = base64.b64decode(p["p"])
+            pn, pt = nt(pb)
+
+            data = json.loads(aesgcm.decrypt(pn, pt, None).decode())
+            if not data.get("exp") and data.get("exp") < datetime.now().date():
+                return False
+
+            data_handler(data)
+            return True
+    except Exception as e:
+        log.error(f"License: {e}")
+
     return False
 
 
@@ -157,6 +208,7 @@ def get_http_authorization_cred(auth_header: Optional[str]):
 
 def get_current_user(
     request: Request,
+    response: Response,
     background_tasks: BackgroundTasks,
     auth_token: HTTPAuthorizationCredentials = Depends(bearer_security),
 ):
@@ -169,7 +221,7 @@ def get_current_user(
         token = request.cookies.get("token")
 
     if token is None:
-        raise HTTPException(status_code=403, detail="Not authenticated")
+        raise HTTPException(status_code=401, detail="Not authenticated")
 
     # auth by api key
     if token.startswith("sk-"):
@@ -225,6 +277,21 @@ def get_current_user(
                 detail=ERROR_MESSAGES.INVALID_TOKEN,
             )
         else:
+            if WEBUI_AUTH_TRUSTED_EMAIL_HEADER:
+                trusted_email = request.headers.get(
+                    WEBUI_AUTH_TRUSTED_EMAIL_HEADER, ""
+                ).lower()
+                if trusted_email and user.email != trusted_email:
+                    # Delete the token cookie
+                    response.delete_cookie("token")
+                    # Delete OAuth token if present
+                    if request.cookies.get("oauth_id_token"):
+                        response.delete_cookie("oauth_id_token")
+                    raise HTTPException(
+                        status_code=status.HTTP_401_UNAUTHORIZED,
+                        detail="User mismatch. Please sign in again.",
+                    )
+
             # Add user info to current span
             current_span = trace.get_current_span()
             if current_span:

backend/open_webui/utils/chat.py

@@ -320,12 +320,7 @@ async def chat_completed(request: Request, form_data: dict, user: Any):
     extra_params = {
         "__event_emitter__": get_event_emitter(metadata),
         "__event_call__": get_event_call(metadata),
-        "__user__": {
-            "id": user.id,
-            "email": user.email,
-            "name": user.name,
-            "role": user.role,
-        },
+        "__user__": user.model_dump() if isinstance(user, UserModel) else {},
         "__metadata__": metadata,
         "__request__": request,
         "__model__": model,
@@ -424,12 +419,7 @@ async def chat_action(request: Request, action_id: str, form_data: dict, user: A
                     params[key] = value
 
             if "__user__" in sig.parameters:
-                __user__ = {
-                    "id": user.id,
-                    "email": user.email,
-                    "name": user.name,
-                    "role": user.role,
-                }
+                __user__ = user.model_dump() if isinstance(user, UserModel) else {}
 
                 try:
                     if hasattr(function_module, "UserValves"):

backend/open_webui/utils/embeddings.py

@@ -0,0 +1,90 @@
+import random
+import logging
+import sys
+
+from fastapi import Request
+from open_webui.models.users import UserModel
+from open_webui.models.models import Models
+from open_webui.utils.models import check_model_access
+from open_webui.env import SRC_LOG_LEVELS, GLOBAL_LOG_LEVEL, BYPASS_MODEL_ACCESS_CONTROL
+
+from open_webui.routers.openai import embeddings as openai_embeddings
+from open_webui.routers.ollama import (
+    embeddings as ollama_embeddings,
+    GenerateEmbeddingsForm,
+)
+
+
+from open_webui.utils.payload import convert_embedding_payload_openai_to_ollama
+from open_webui.utils.response import convert_embedding_response_ollama_to_openai
+
+logging.basicConfig(stream=sys.stdout, level=GLOBAL_LOG_LEVEL)
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["MAIN"])
+
+
+async def generate_embeddings(
+    request: Request,
+    form_data: dict,
+    user: UserModel,
+    bypass_filter: bool = False,
+):
+    """
+    Dispatch and handle embeddings generation based on the model type (OpenAI, Ollama).
+
+    Args:
+        request (Request): The FastAPI request context.
+        form_data (dict): The input data sent to the endpoint.
+        user (UserModel): The authenticated user.
+        bypass_filter (bool): If True, disables access filtering (default False).
+
+    Returns:
+        dict: The embeddings response, following OpenAI API compatibility.
+    """
+    if BYPASS_MODEL_ACCESS_CONTROL:
+        bypass_filter = True
+
+    # Attach extra metadata from request.state if present
+    if hasattr(request.state, "metadata"):
+        if "metadata" not in form_data:
+            form_data["metadata"] = request.state.metadata
+        else:
+            form_data["metadata"] = {
+                **form_data["metadata"],
+                **request.state.metadata,
+            }
+
+    # If "direct" flag present, use only that model
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS
+
+    model_id = form_data.get("model")
+    if model_id not in models:
+        raise Exception("Model not found")
+    model = models[model_id]
+
+    # Access filtering
+    if not getattr(request.state, "direct", False):
+        if not bypass_filter and user.role == "user":
+            check_model_access(user, model)
+
+    # Ollama backend
+    if model.get("owned_by") == "ollama":
+        ollama_payload = convert_embedding_payload_openai_to_ollama(form_data)
+        response = await ollama_embeddings(
+            request=request,
+            form_data=GenerateEmbeddingsForm(**ollama_payload),
+            user=user,
+        )
+        return convert_embedding_response_ollama_to_openai(response)
+
+    # Default: OpenAI or compatible backend
+    return await openai_embeddings(
+        request=request,
+        form_data=form_data,
+        user=user,
+    )

backend/open_webui/utils/logger.py

@@ -4,12 +4,15 @@ import sys
 from typing import TYPE_CHECKING
 
 from loguru import logger
-
+from opentelemetry import trace
 from open_webui.env import (
+    AUDIT_UVICORN_LOGGER_NAMES,
     AUDIT_LOG_FILE_ROTATION_SIZE,
     AUDIT_LOG_LEVEL,
     AUDIT_LOGS_FILE_PATH,
     GLOBAL_LOG_LEVEL,
+    ENABLE_OTEL,
+    ENABLE_OTEL_LOGS,
 )
 
 
@@ -26,13 +29,16 @@ def stdout_format(record: "Record") -> str:
     Returns:
     str: A formatted log string intended for stdout.
     """
-    record["extra"]["extra_json"] = json.dumps(record["extra"])
+    if record["extra"]:
+        record["extra"]["extra_json"] = json.dumps(record["extra"])
+        extra_format = " - {extra[extra_json]}"
+    else:
+        extra_format = ""
     return (
         "<green>{time:YYYY-MM-DD HH:mm:ss.SSS}</green> | "
         "<level>{level: <8}</level> | "
         "<cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - "
-        "<level>{message}</level> - {extra[extra_json]}"
-        "\n{exception}"
+        "<level>{message}</level>" + extra_format + "\n{exception}"
     )
 
 
@@ -58,9 +64,24 @@ class InterceptHandler(logging.Handler):
             frame = frame.f_back
             depth += 1
 
-        logger.opt(depth=depth, exception=record.exc_info).log(
-            level, record.getMessage()
-        )
+        logger.opt(depth=depth, exception=record.exc_info).bind(
+            **self._get_extras()
+        ).log(level, record.getMessage())
+        if ENABLE_OTEL and ENABLE_OTEL_LOGS:
+            from open_webui.utils.telemetry.logs import otel_handler
+
+            otel_handler.emit(record)
+
+    def _get_extras(self):
+        if not ENABLE_OTEL:
+            return {}
+
+        extras = {}
+        context = trace.get_current_span().get_span_context()
+        if context.is_valid:
+            extras["trace_id"] = trace.format_trace_id(context.trace_id)
+            extras["span_id"] = trace.format_span_id(context.span_id)
+        return extras
 
 
 def file_format(record: "Record"):
@@ -111,7 +132,6 @@ def start_logger():
         format=stdout_format,
         filter=lambda record: "auditable" not in record["extra"],
     )
-
     if AUDIT_LOG_LEVEL != "NONE":
         try:
             logger.add(
@@ -128,11 +148,13 @@ def start_logger():
     logging.basicConfig(
         handlers=[InterceptHandler()], level=GLOBAL_LOG_LEVEL, force=True
     )
+
     for uvicorn_logger_name in ["uvicorn", "uvicorn.error"]:
         uvicorn_logger = logging.getLogger(uvicorn_logger_name)
         uvicorn_logger.setLevel(GLOBAL_LOG_LEVEL)
         uvicorn_logger.handlers = []
-    for uvicorn_logger_name in ["uvicorn.access"]:
+
+    for uvicorn_logger_name in AUDIT_UVICORN_LOGGER_NAMES:
         uvicorn_logger = logging.getLogger(uvicorn_logger_name)
         uvicorn_logger.setLevel(GLOBAL_LOG_LEVEL)
         uvicorn_logger.handlers = [InterceptHandler()]

backend/open_webui/utils/misc.py

@@ -208,6 +208,7 @@ def openai_chat_message_template(model: str):
 def openai_chat_chunk_message_template(
     model: str,
     content: Optional[str] = None,
+    reasoning_content: Optional[str] = None,
     tool_calls: Optional[list[dict]] = None,
     usage: Optional[dict] = None,
 ) -> dict:
@@ -220,10 +221,13 @@ def openai_chat_chunk_message_template(
     if content:
         template["choices"][0]["delta"]["content"] = content
 
+    if reasoning_content:
+        template["choices"][0]["delta"]["reasoning_content"] = reasoning_content
+
     if tool_calls:
         template["choices"][0]["delta"]["tool_calls"] = tool_calls
 
-    if not content and not tool_calls:
+    if not content and not reasoning_content and not tool_calls:
         template["choices"][0]["finish_reason"] = "stop"
 
     if usage:
@@ -234,6 +238,7 @@ def openai_chat_chunk_message_template(
 def openai_chat_completion_message_template(
     model: str,
     message: Optional[str] = None,
+    reasoning_content: Optional[str] = None,
     tool_calls: Optional[list[dict]] = None,
     usage: Optional[dict] = None,
 ) -> dict:
@@ -241,8 +246,9 @@ def openai_chat_completion_message_template(
     template["object"] = "chat.completion"
     if message is not None:
         template["choices"][0]["message"] = {
-            "content": message,
             "role": "assistant",
+            "content": message,
+            **({"reasoning_content": reasoning_content} if reasoning_content else {}),
             **({"tool_calls": tool_calls} if tool_calls else {}),
         }
 

backend/open_webui/utils/models.py

@@ -76,8 +76,19 @@ async def get_all_base_models(request: Request, user: UserModel = None):
     return function_models + openai_models + ollama_models
 
 
-async def get_all_models(request, user: UserModel = None):
-    models = await get_all_base_models(request, user=user)
+async def get_all_models(request, refresh: bool = False, user: UserModel = None):
+    if (
+        request.app.state.MODELS
+        and request.app.state.BASE_MODELS
+        and (request.app.state.config.ENABLE_BASE_MODELS_CACHE and not refresh)
+    ):
+        base_models = request.app.state.BASE_MODELS
+    else:
+        base_models = await get_all_base_models(request, user=user)
+        request.app.state.BASE_MODELS = base_models
+
+    # deep copy the base models to avoid modifying the original list
+    models = [model.copy() for model in base_models]
 
     # If there are no models, return an empty list
     if len(models) == 0:
@@ -137,6 +148,7 @@ async def get_all_models(request, user: UserModel = None):
     custom_models = Models.get_all_models()
     for custom_model in custom_models:
         if custom_model.base_model_id is None:
+            # Applied directly to a base model
             for model in models:
                 if custom_model.id == model["id"] or (
                     model.get("owned_by") == "ollama"

backend/open_webui/utils/oauth.py

@@ -27,6 +27,7 @@ from open_webui.config import (
     ENABLE_OAUTH_GROUP_CREATION,
     OAUTH_BLOCKED_GROUPS,
     OAUTH_ROLES_CLAIM,
+    OAUTH_SUB_CLAIM,
     OAUTH_GROUPS_CLAIM,
     OAUTH_EMAIL_CLAIM,
     OAUTH_PICTURE_CLAIM,
@@ -65,6 +66,7 @@ auth_manager_config.ENABLE_OAUTH_GROUP_MANAGEMENT = ENABLE_OAUTH_GROUP_MANAGEMEN
 auth_manager_config.ENABLE_OAUTH_GROUP_CREATION = ENABLE_OAUTH_GROUP_CREATION
 auth_manager_config.OAUTH_BLOCKED_GROUPS = OAUTH_BLOCKED_GROUPS
 auth_manager_config.OAUTH_ROLES_CLAIM = OAUTH_ROLES_CLAIM
+auth_manager_config.OAUTH_SUB_CLAIM = OAUTH_SUB_CLAIM
 auth_manager_config.OAUTH_GROUPS_CLAIM = OAUTH_GROUPS_CLAIM
 auth_manager_config.OAUTH_EMAIL_CLAIM = OAUTH_EMAIL_CLAIM
 auth_manager_config.OAUTH_PICTURE_CLAIM = OAUTH_PICTURE_CLAIM
@@ -88,11 +90,12 @@ class OAuthManager:
         return self.oauth.create_client(provider_name)
 
     def get_user_role(self, user, user_data):
-        if user and Users.get_num_users() == 1:
+        user_count = Users.get_num_users()
+        if user and user_count == 1:
             # If the user is the only user, assign the role "admin" - actually repairs role for single user on login
             log.debug("Assigning the only user the admin role")
             return "admin"
-        if not user and Users.get_num_users() == 0:
+        if not user and user_count == 0:
             # If there are no users, assign the role "admin", as the first user will be an admin
             log.debug("Assigning the first user the admin role")
             return "admin"
@@ -358,11 +361,18 @@ class OAuthManager:
             log.warning(f"OAuth callback failed, user data is missing: {token}")
             raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
 
-        sub = user_data.get(OAUTH_PROVIDERS[provider].get("sub_claim", "sub"))
+        if auth_manager_config.OAUTH_SUB_CLAIM:
+            sub = user_data.get(auth_manager_config.OAUTH_SUB_CLAIM)
+        else:
+            # Fallback to the default sub claim if not configured
+            sub = user_data.get(OAUTH_PROVIDERS[provider].get("sub_claim", "sub"))
+
         if not sub:
             log.warning(f"OAuth callback failed, sub is missing: {user_data}")
             raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
+
         provider_sub = f"{provider}@{sub}"
+
         email_claim = auth_manager_config.OAUTH_EMAIL_CLAIM
         email = user_data.get(email_claim, "")
         # We currently mandate that email addresses are provided
@@ -449,8 +459,6 @@ class OAuthManager:
                         log.debug(f"Updated profile picture for user {user.email}")
 
         if not user:
-            user_count = Users.get_num_users()
-
             # If the user does not exist, check if signups are enabled
             if auth_manager_config.ENABLE_OAUTH_SIGNUP:
                 # Check if an existing user with the same email already exists
@@ -521,7 +529,7 @@ class OAuthManager:
         response.set_cookie(
             key="token",
             value=jwt_token,
-            httponly=True,  # Ensures the cookie is not accessible via JavaScript
+            httponly=False,  # Required for frontend access
             samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
             secure=WEBUI_AUTH_COOKIE_SECURE,
         )
@@ -537,9 +545,9 @@ class OAuthManager:
             )
         # Redirect back to the frontend with the JWT token
 
-        redirect_base_url = request.app.state.config.WEBUI_URL or request.base_url
+        redirect_base_url = str(request.app.state.config.WEBUI_URL or request.base_url)
         if redirect_base_url.endswith("/"):
             redirect_base_url = redirect_base_url[:-1]
-        redirect_url = f"{redirect_base_url}/auth#token={jwt_token}"
+        redirect_url = f"{redirect_base_url}/auth"
 
         return RedirectResponse(url=redirect_url, headers=response.headers)

backend/open_webui/utils/payload.py

@@ -10,9 +10,8 @@ import json
 
 # inplace function: form_data is modified
 def apply_model_system_prompt_to_body(
-    params: dict, form_data: dict, metadata: Optional[dict] = None, user=None
+    system: Optional[str], form_data: dict, metadata: Optional[dict] = None, user=None
 ) -> dict:
-    system = params.get("system", None)
     if not system:
         return form_data
 
@@ -58,10 +57,46 @@ def apply_model_params_to_body(
     return form_data
 
 
+def remove_open_webui_params(params: dict) -> dict:
+    """
+    Removes OpenWebUI specific parameters from the provided dictionary.
+
+    Args:
+        params (dict): The dictionary containing parameters.
+
+    Returns:
+        dict: The modified dictionary with OpenWebUI parameters removed.
+    """
+    open_webui_params = {
+        "stream_response": bool,
+        "stream_delta_chunk_size": int,
+        "function_calling": str,
+        "system": str,
+    }
+
+    for key in list(params.keys()):
+        if key in open_webui_params:
+            del params[key]
+
+    return params
+
+
 # inplace function: form_data is modified
 def apply_model_params_to_body_openai(params: dict, form_data: dict) -> dict:
+    params = remove_open_webui_params(params)
+
     custom_params = params.pop("custom_params", {})
     if custom_params:
+        # Attempt to parse custom_params if they are strings
+        for key, value in custom_params.items():
+            if isinstance(value, str):
+                try:
+                    # Attempt to parse the string as JSON
+                    custom_params[key] = json.loads(value)
+                except json.JSONDecodeError:
+                    # If it fails, keep the original string
+                    pass
+
         # If there are custom parameters, we need to apply them first
         params = deep_update(params, custom_params)
 
@@ -82,8 +117,20 @@ def apply_model_params_to_body_openai(params: dict, form_data: dict) -> dict:
 
 
 def apply_model_params_to_body_ollama(params: dict, form_data: dict) -> dict:
+    params = remove_open_webui_params(params)
+
     custom_params = params.pop("custom_params", {})
     if custom_params:
+        # Attempt to parse custom_params if they are strings
+        for key, value in custom_params.items():
+            if isinstance(value, str):
+                try:
+                    # Attempt to parse the string as JSON
+                    custom_params[key] = json.loads(value)
+                except json.JSONDecodeError:
+                    # If it fails, keep the original string
+                    pass
+
         # If there are custom parameters, we need to apply them first
         params = deep_update(params, custom_params)
 
@@ -129,16 +176,32 @@ def apply_model_params_to_body_ollama(params: dict, form_data: dict) -> dict:
         "num_thread": int,
     }
 
-    # Extract keep_alive from options if it exists
-    if "options" in form_data and "keep_alive" in form_data["options"]:
-        form_data["keep_alive"] = form_data["options"]["keep_alive"]
-        del form_data["options"]["keep_alive"]
+    def parse_json(value: str) -> dict:
+        """
+        Parses a JSON string into a dictionary, handling potential JSONDecodeError.
+        """
+        try:
+            return json.loads(value)
+        except Exception as e:
+            return value
 
-    if "options" in form_data and "format" in form_data["options"]:
-        form_data["format"] = form_data["options"]["format"]
-        del form_data["options"]["format"]
+    ollama_root_params = {
+        "format": lambda x: parse_json(x),
+        "keep_alive": lambda x: parse_json(x),
+        "think": bool,
+    }
 
-    return apply_model_params_to_body(params, form_data, mappings)
+    for key, value in ollama_root_params.items():
+        if (param := params.get(key, None)) is not None:
+            # Copy the parameter to new name then delete it, to prevent Ollama warning of invalid option provided
+            form_data[key] = value(param)
+            del params[key]
+
+    # Unlike OpenAI, Ollama does not support params directly in the body
+    form_data["options"] = apply_model_params_to_body(
+        params, (form_data.get("options", {}) or {}), mappings
+    )
+    return form_data
 
 
 def convert_messages_openai_to_ollama(messages: list[dict]) -> list[dict]:
@@ -233,36 +296,48 @@ def convert_payload_openai_to_ollama(openai_payload: dict) -> dict:
         openai_payload.get("messages")
     )
     ollama_payload["stream"] = openai_payload.get("stream", False)
-
     if "tools" in openai_payload:
         ollama_payload["tools"] = openai_payload["tools"]
 
-    if "format" in openai_payload:
-        ollama_payload["format"] = openai_payload["format"]
-
     # If there are advanced parameters in the payload, format them in Ollama's options field
     if openai_payload.get("options"):
         ollama_payload["options"] = openai_payload["options"]
         ollama_options = openai_payload["options"]
 
+        def parse_json(value: str) -> dict:
+            """
+            Parses a JSON string into a dictionary, handling potential JSONDecodeError.
+            """
+            try:
+                return json.loads(value)
+            except Exception as e:
+                return value
+
+        ollama_root_params = {
+            "format": lambda x: parse_json(x),
+            "keep_alive": lambda x: parse_json(x),
+            "think": bool,
+        }
+
+        # Ollama's options field can contain parameters that should be at the root level.
+        for key, value in ollama_root_params.items():
+            if (param := ollama_options.get(key, None)) is not None:
+                # Copy the parameter to new name then delete it, to prevent Ollama warning of invalid option provided
+                ollama_payload[key] = value(param)
+                del ollama_options[key]
+
         # Re-Mapping OpenAI's `max_tokens` -> Ollama's `num_predict`
         if "max_tokens" in ollama_options:
             ollama_options["num_predict"] = ollama_options["max_tokens"]
-            del ollama_options[
-                "max_tokens"
-            ]  # To prevent Ollama warning of invalid option provided
+            del ollama_options["max_tokens"]
 
         # Ollama lacks a "system" prompt option. It has to be provided as a direct parameter, so we copy it down.
+        # Comment: Not sure why this is needed, but we'll keep it for compatibility.
         if "system" in ollama_options:
             ollama_payload["system"] = ollama_options["system"]
-            del ollama_options[
-                "system"
-            ]  # To prevent Ollama warning of invalid option provided
+            del ollama_options["system"]
 
-        # Extract keep_alive from options if it exists
-        if "keep_alive" in ollama_options:
-            ollama_payload["keep_alive"] = ollama_options["keep_alive"]
-            del ollama_options["keep_alive"]
+        ollama_payload["options"] = ollama_options
 
     # If there is the "stop" parameter in the openai_payload, remap it to the ollama_payload.options
     if "stop" in openai_payload:
@@ -283,3 +358,32 @@ def convert_payload_openai_to_ollama(openai_payload: dict) -> dict:
             ollama_payload["format"] = format
 
     return ollama_payload
+
+
+def convert_embedding_payload_openai_to_ollama(openai_payload: dict) -> dict:
+    """
+    Convert an embeddings request payload from OpenAI format to Ollama format.
+
+    Args:
+        openai_payload (dict): The original payload designed for OpenAI API usage.
+
+    Returns:
+        dict: A payload compatible with the Ollama API embeddings endpoint.
+    """
+    ollama_payload = {"model": openai_payload.get("model")}
+    input_value = openai_payload.get("input")
+
+    # Ollama expects 'input' as a list, and 'prompt' as a single string.
+    if isinstance(input_value, list):
+        ollama_payload["input"] = input_value
+        ollama_payload["prompt"] = "\n".join(str(x) for x in input_value)
+    else:
+        ollama_payload["input"] = [input_value]
+        ollama_payload["prompt"] = str(input_value)
+
+    # Optionally forward other fields if present
+    for optional_key in ("options", "truncate", "keep_alive"):
+        if optional_key in openai_payload:
+            ollama_payload[optional_key] = openai_payload[optional_key]
+
+    return ollama_payload

backend/open_webui/utils/redis.py

@@ -1,13 +1,103 @@
-import socketio
-import redis
-from redis import asyncio as aioredis
+import inspect
 from urllib.parse import urlparse
 
+import logging
+
+import redis
+
+from open_webui.env import REDIS_SENTINEL_MAX_RETRY_COUNT
+
+log = logging.getLogger(__name__)
+
+
+_CONNECTION_CACHE = {}
+
+
+class SentinelRedisProxy:
+    def __init__(self, sentinel, service, *, async_mode: bool = True, **kw):
+        self._sentinel = sentinel
+        self._service = service
+        self._kw = kw
+        self._async_mode = async_mode
+
+    def _master(self):
+        return self._sentinel.master_for(self._service, **self._kw)
+
+    def __getattr__(self, item):
+        master = self._master()
+        orig_attr = getattr(master, item)
+
+        if not callable(orig_attr):
+            return orig_attr
+
+        FACTORY_METHODS = {"pipeline", "pubsub", "monitor", "client", "transaction"}
+        if item in FACTORY_METHODS:
+            return orig_attr
+
+        if self._async_mode:
+
+            async def _wrapped(*args, **kwargs):
+                for i in range(REDIS_SENTINEL_MAX_RETRY_COUNT):
+                    try:
+                        method = getattr(self._master(), item)
+                        result = method(*args, **kwargs)
+                        if inspect.iscoroutine(result):
+                            return await result
+                        return result
+                    except (
+                        redis.exceptions.ConnectionError,
+                        redis.exceptions.ReadOnlyError,
+                    ) as e:
+                        if i < REDIS_SENTINEL_MAX_RETRY_COUNT - 1:
+                            log.debug(
+                                "Redis sentinel fail-over (%s). Retry %s/%s",
+                                type(e).__name__,
+                                i + 1,
+                                REDIS_SENTINEL_MAX_RETRY_COUNT,
+                            )
+                            continue
+                        log.error(
+                            "Redis operation failed after %s retries: %s",
+                            REDIS_SENTINEL_MAX_RETRY_COUNT,
+                            e,
+                        )
+                        raise e from e
+
+            return _wrapped
+
+        else:
+
+            def _wrapped(*args, **kwargs):
+                for i in range(REDIS_SENTINEL_MAX_RETRY_COUNT):
+                    try:
+                        method = getattr(self._master(), item)
+                        return method(*args, **kwargs)
+                    except (
+                        redis.exceptions.ConnectionError,
+                        redis.exceptions.ReadOnlyError,
+                    ) as e:
+                        if i < REDIS_SENTINEL_MAX_RETRY_COUNT - 1:
+                            log.debug(
+                                "Redis sentinel fail-over (%s). Retry %s/%s",
+                                type(e).__name__,
+                                i + 1,
+                                REDIS_SENTINEL_MAX_RETRY_COUNT,
+                            )
+                            continue
+                        log.error(
+                            "Redis operation failed after %s retries: %s",
+                            REDIS_SENTINEL_MAX_RETRY_COUNT,
+                            e,
+                        )
+                        raise e from e
+
+            return _wrapped
+
 
 def parse_redis_service_url(redis_url):
     parsed_url = urlparse(redis_url)
-    if parsed_url.scheme != "redis":
-        raise ValueError("Invalid Redis URL scheme. Must be 'redis'.")
+    if parsed_url.scheme != "redis" and parsed_url.scheme != "rediss":
+        raise ValueError("Invalid Redis URL scheme. Must be 'redis' or 'rediss'.")
 
     return {
         "username": parsed_url.username or None,
@@ -18,23 +108,84 @@ def parse_redis_service_url(redis_url):
     }
 
 
-def get_redis_connection(redis_url, redis_sentinels, decode_responses=True):
-    if redis_sentinels:
-        redis_config = parse_redis_service_url(redis_url)
-        sentinel = redis.sentinel.Sentinel(
-            redis_sentinels,
-            port=redis_config["port"],
-            db=redis_config["db"],
-            username=redis_config["username"],
-            password=redis_config["password"],
-            decode_responses=decode_responses,
-        )
+def get_redis_connection(
+    redis_url,
+    redis_sentinels,
+    redis_cluster=False,
+    async_mode=False,
+    decode_responses=True,
+):
 
-        # Get a master connection from Sentinel
-        return sentinel.master_for(redis_config["service"])
+    cache_key = (
+        redis_url,
+        tuple(redis_sentinels) if redis_sentinels else (),
+        async_mode,
+        decode_responses,
+    )
+
+    if cache_key in _CONNECTION_CACHE:
+        return _CONNECTION_CACHE[cache_key]
+
+    connection = None
+
+    if async_mode:
+        import redis.asyncio as redis
+
+        # If using sentinel in async mode
+        if redis_sentinels:
+            redis_config = parse_redis_service_url(redis_url)
+            sentinel = redis.sentinel.Sentinel(
+                redis_sentinels,
+                port=redis_config["port"],
+                db=redis_config["db"],
+                username=redis_config["username"],
+                password=redis_config["password"],
+                decode_responses=decode_responses,
+            )
+            connection = SentinelRedisProxy(
+                sentinel,
+                redis_config["service"],
+                async_mode=async_mode,
+            )
+        elif redis_cluster:
+            if not redis_url:
+                raise ValueError("Redis URL must be provided for cluster mode.")
+            return redis.cluster.RedisCluster.from_url(
+                redis_url, decode_responses=decode_responses
+            )
+        elif redis_url:
+            connection = redis.from_url(redis_url, decode_responses=decode_responses)
     else:
-        # Standard Redis connection
-        return redis.Redis.from_url(redis_url, decode_responses=decode_responses)
+        import redis
+
+        if redis_sentinels:
+            redis_config = parse_redis_service_url(redis_url)
+            sentinel = redis.sentinel.Sentinel(
+                redis_sentinels,
+                port=redis_config["port"],
+                db=redis_config["db"],
+                username=redis_config["username"],
+                password=redis_config["password"],
+                decode_responses=decode_responses,
+            )
+            connection = SentinelRedisProxy(
+                sentinel,
+                redis_config["service"],
+                async_mode=async_mode,
+            )
+        elif redis_cluster:
+            if not redis_url:
+                raise ValueError("Redis URL must be provided for cluster mode.")
+            return redis.cluster.RedisCluster.from_url(
+                redis_url, decode_responses=decode_responses
+            )
+        elif redis_url:
+            connection = redis.Redis.from_url(
+                redis_url, decode_responses=decode_responses
+            )
+
+    _CONNECTION_CACHE[cache_key] = connection
+    return connection
 
 
 def get_sentinels_from_env(sentinel_hosts_env, sentinel_port_env):

backend/open_webui/utils/response.py

@@ -6,18 +6,17 @@ from open_webui.utils.misc import (
 )
 
 
-def convert_ollama_tool_call_to_openai(tool_calls: dict) -> dict:
+def convert_ollama_tool_call_to_openai(tool_calls: list) -> list:
     openai_tool_calls = []
     for tool_call in tool_calls:
+        function = tool_call.get("function", {})
         openai_tool_call = {
-            "index": tool_call.get("index", 0),
+            "index": tool_call.get("index", function.get("index", 0)),
             "id": tool_call.get("id", f"call_{str(uuid4())}"),
             "type": "function",
             "function": {
-                "name": tool_call.get("function", {}).get("name", ""),
-                "arguments": json.dumps(
-                    tool_call.get("function", {}).get("arguments", {})
-                ),
+                "name": function.get("name", ""),
+                "arguments": json.dumps(function.get("arguments", {})),
             },
         }
         openai_tool_calls.append(openai_tool_call)
@@ -83,6 +82,7 @@ def convert_ollama_usage_to_openai(data: dict) -> dict:
 def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
     model = ollama_response.get("model", "ollama")
     message_content = ollama_response.get("message", {}).get("content", "")
+    reasoning_content = ollama_response.get("message", {}).get("thinking", None)
     tool_calls = ollama_response.get("message", {}).get("tool_calls", None)
     openai_tool_calls = None
 
@@ -94,7 +94,7 @@ def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
     usage = convert_ollama_usage_to_openai(data)
 
     response = openai_chat_completion_message_template(
-        model, message_content, openai_tool_calls, usage
+        model, message_content, reasoning_content, openai_tool_calls, usage
     )
     return response
 
@@ -105,6 +105,7 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)
 
         model = data.get("model", "ollama")
         message_content = data.get("message", {}).get("content", None)
+        reasoning_content = data.get("message", {}).get("thinking", None)
         tool_calls = data.get("message", {}).get("tool_calls", None)
         openai_tool_calls = None
 
@@ -118,10 +119,71 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)
             usage = convert_ollama_usage_to_openai(data)
 
         data = openai_chat_chunk_message_template(
-            model, message_content, openai_tool_calls, usage
+            model, message_content, reasoning_content, openai_tool_calls, usage
         )
 
         line = f"data: {json.dumps(data)}\n\n"
         yield line
 
     yield "data: [DONE]\n\n"
+
+
+def convert_embedding_response_ollama_to_openai(response) -> dict:
+    """
+    Convert the response from Ollama embeddings endpoint to the OpenAI-compatible format.
+
+    Args:
+        response (dict): The response from the Ollama API,
+            e.g. {"embedding": [...], "model": "..."}
+            or {"embeddings": [{"embedding": [...], "index": 0}, ...], "model": "..."}
+
+    Returns:
+        dict: Response adapted to OpenAI's embeddings API format.
+            e.g. {
+                "object": "list",
+                "data": [
+                    {"object": "embedding", "embedding": [...], "index": 0},
+                    ...
+                ],
+                "model": "...",
+            }
+    """
+    # Ollama batch-style output
+    if isinstance(response, dict) and "embeddings" in response:
+        openai_data = []
+        for i, emb in enumerate(response["embeddings"]):
+            openai_data.append(
+                {
+                    "object": "embedding",
+                    "embedding": emb.get("embedding"),
+                    "index": emb.get("index", i),
+                }
+            )
+        return {
+            "object": "list",
+            "data": openai_data,
+            "model": response.get("model"),
+        }
+    # Ollama single output
+    elif isinstance(response, dict) and "embedding" in response:
+        return {
+            "object": "list",
+            "data": [
+                {
+                    "object": "embedding",
+                    "embedding": response["embedding"],
+                    "index": 0,
+                }
+            ],
+            "model": response.get("model"),
+        }
+    # Already OpenAI-compatible?
+    elif (
+        isinstance(response, dict)
+        and "data" in response
+        and isinstance(response["data"], list)
+    ):
+        return response
+
+    # Fallback: return as is if unrecognized
+    return response

backend/open_webui/utils/task.py

@@ -207,6 +207,24 @@ def title_generation_template(
     return template
 
 
+def follow_up_generation_template(
+    template: str, messages: list[dict], user: Optional[dict] = None
+) -> str:
+    prompt = get_last_user_message(messages)
+    template = replace_prompt_variable(template, prompt)
+    template = replace_messages_variable(template, messages)
+
+    template = prompt_template(
+        template,
+        **(
+            {"user_name": user.get("name"), "user_location": user.get("location")}
+            if user
+            else {}
+        ),
+    )
+    return template
+
+
 def tags_generation_template(
     template: str, messages: list[dict], user: Optional[dict] = None
 ) -> str:

backend/open_webui/utils/telemetry/exporters.py

@@ -1,31 +0,0 @@
-import threading
-
-from opentelemetry.sdk.trace import ReadableSpan
-from opentelemetry.sdk.trace.export import BatchSpanProcessor
-
-
-class LazyBatchSpanProcessor(BatchSpanProcessor):
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.done = True
-        with self.condition:
-            self.condition.notify_all()
-        self.worker_thread.join()
-        self.done = False
-        self.worker_thread = None
-
-    def on_end(self, span: ReadableSpan) -> None:
-        if self.worker_thread is None:
-            self.worker_thread = threading.Thread(
-                name=self.__class__.__name__, target=self.worker, daemon=True
-            )
-            self.worker_thread.start()
-        super().on_end(span)
-
-    def shutdown(self) -> None:
-        self.done = True
-        with self.condition:
-            self.condition.notify_all()
-        if self.worker_thread:
-            self.worker_thread.join()
-        self.span_exporter.shutdown()

backend/open_webui/utils/telemetry/logs.py

@@ -0,0 +1,53 @@
+import logging
+from base64 import b64encode
+from opentelemetry.sdk._logs import (
+    LoggingHandler,
+    LoggerProvider,
+)
+from opentelemetry.exporter.otlp.proto.grpc._log_exporter import OTLPLogExporter
+from opentelemetry.exporter.otlp.proto.http._log_exporter import (
+    OTLPLogExporter as HttpOTLPLogExporter,
+)
+from opentelemetry.sdk._logs.export import BatchLogRecordProcessor
+from opentelemetry._logs import set_logger_provider
+from opentelemetry.sdk.resources import SERVICE_NAME, Resource
+from open_webui.env import (
+    OTEL_SERVICE_NAME,
+    OTEL_LOGS_EXPORTER_OTLP_ENDPOINT,
+    OTEL_LOGS_EXPORTER_OTLP_INSECURE,
+    OTEL_LOGS_BASIC_AUTH_USERNAME,
+    OTEL_LOGS_BASIC_AUTH_PASSWORD,
+    OTEL_LOGS_OTLP_SPAN_EXPORTER,
+)
+
+
+def setup_logging():
+    headers = []
+    if OTEL_LOGS_BASIC_AUTH_USERNAME and OTEL_LOGS_BASIC_AUTH_PASSWORD:
+        auth_string = f"{OTEL_LOGS_BASIC_AUTH_USERNAME}:{OTEL_LOGS_BASIC_AUTH_PASSWORD}"
+        auth_header = b64encode(auth_string.encode()).decode()
+        headers = [("authorization", f"Basic {auth_header}")]
+    resource = Resource.create(attributes={SERVICE_NAME: OTEL_SERVICE_NAME})
+
+    if OTEL_LOGS_OTLP_SPAN_EXPORTER == "http":
+        exporter = HttpOTLPLogExporter(
+            endpoint=OTEL_LOGS_EXPORTER_OTLP_ENDPOINT,
+            headers=headers,
+        )
+    else:
+        exporter = OTLPLogExporter(
+            endpoint=OTEL_LOGS_EXPORTER_OTLP_ENDPOINT,
+            insecure=OTEL_LOGS_EXPORTER_OTLP_INSECURE,
+            headers=headers,
+        )
+    logger_provider = LoggerProvider(resource=resource)
+    set_logger_provider(logger_provider)
+
+    logger_provider.add_log_record_processor(BatchLogRecordProcessor(exporter))
+
+    otel_handler = LoggingHandler(logger_provider=logger_provider)
+
+    return otel_handler
+
+
+otel_handler = setup_logging()

backend/open_webui/utils/telemetry/metrics.py

@@ -0,0 +1,182 @@
+"""OpenTelemetry metrics bootstrap for Open WebUI.
+
+This module initialises a MeterProvider that sends metrics to an OTLP
+collector. The collector is responsible for exposing a Prometheus
+`/metrics` endpoint – WebUI does **not** expose it directly.
+
+Metrics collected:
+
+* http.server.requests (counter)
+* http.server.duration (histogram, milliseconds)
+
+Attributes used: http.method, http.route, http.status_code
+
+If you wish to add more attributes (e.g. user-agent) you can, but beware of
+high-cardinality label sets.
+"""
+
+from __future__ import annotations
+
+import time
+from typing import Dict, List, Sequence, Any
+from base64 import b64encode
+
+from fastapi import FastAPI, Request
+from opentelemetry import metrics
+from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import (
+    OTLPMetricExporter,
+)
+
+from opentelemetry.exporter.otlp.proto.http.metric_exporter import (
+    OTLPMetricExporter as OTLPHttpMetricExporter,
+)
+from opentelemetry.sdk.metrics import MeterProvider
+from opentelemetry.sdk.metrics.view import View
+from opentelemetry.sdk.metrics.export import (
+    PeriodicExportingMetricReader,
+)
+from opentelemetry.sdk.resources import Resource
+
+from open_webui.env import (
+    OTEL_SERVICE_NAME,
+    OTEL_METRICS_EXPORTER_OTLP_ENDPOINT,
+    OTEL_METRICS_BASIC_AUTH_USERNAME,
+    OTEL_METRICS_BASIC_AUTH_PASSWORD,
+    OTEL_METRICS_OTLP_SPAN_EXPORTER,
+    OTEL_METRICS_EXPORTER_OTLP_INSECURE,
+)
+from open_webui.socket.main import get_active_user_ids
+from open_webui.models.users import Users
+
+_EXPORT_INTERVAL_MILLIS = 10_000  # 10 seconds
+
+
+def _build_meter_provider(resource: Resource) -> MeterProvider:
+    """Return a configured MeterProvider."""
+    headers = []
+    if OTEL_METRICS_BASIC_AUTH_USERNAME and OTEL_METRICS_BASIC_AUTH_PASSWORD:
+        auth_string = (
+            f"{OTEL_METRICS_BASIC_AUTH_USERNAME}:{OTEL_METRICS_BASIC_AUTH_PASSWORD}"
+        )
+        auth_header = b64encode(auth_string.encode()).decode()
+        headers = [("authorization", f"Basic {auth_header}")]
+
+    # Periodic reader pushes metrics over OTLP/gRPC to collector
+    if OTEL_METRICS_OTLP_SPAN_EXPORTER == "http":
+        readers: List[PeriodicExportingMetricReader] = [
+            PeriodicExportingMetricReader(
+                OTLPHttpMetricExporter(
+                    endpoint=OTEL_METRICS_EXPORTER_OTLP_ENDPOINT, headers=headers
+                ),
+                export_interval_millis=_EXPORT_INTERVAL_MILLIS,
+            )
+        ]
+    else:
+        readers: List[PeriodicExportingMetricReader] = [
+            PeriodicExportingMetricReader(
+                OTLPMetricExporter(
+                    endpoint=OTEL_METRICS_EXPORTER_OTLP_ENDPOINT,
+                    insecure=OTEL_METRICS_EXPORTER_OTLP_INSECURE,
+                    headers=headers,
+                ),
+                export_interval_millis=_EXPORT_INTERVAL_MILLIS,
+            )
+        ]
+
+    # Optional view to limit cardinality: drop user-agent etc.
+    views: List[View] = [
+        View(
+            instrument_name="http.server.duration",
+            attribute_keys=["http.method", "http.route", "http.status_code"],
+        ),
+        View(
+            instrument_name="http.server.requests",
+            attribute_keys=["http.method", "http.route", "http.status_code"],
+        ),
+        View(
+            instrument_name="webui.users.total",
+        ),
+        View(
+            instrument_name="webui.users.active",
+        ),
+    ]
+
+    provider = MeterProvider(
+        resource=resource,
+        metric_readers=list(readers),
+        views=views,
+    )
+    return provider
+
+
+def setup_metrics(app: FastAPI, resource: Resource) -> None:
+    """Attach OTel metrics middleware to *app* and initialise provider."""
+
+    metrics.set_meter_provider(_build_meter_provider(resource))
+    meter = metrics.get_meter(__name__)
+
+    # Instruments
+    request_counter = meter.create_counter(
+        name="http.server.requests",
+        description="Total HTTP requests",
+        unit="1",
+    )
+    duration_histogram = meter.create_histogram(
+        name="http.server.duration",
+        description="HTTP request duration",
+        unit="ms",
+    )
+
+    def observe_active_users(
+        options: metrics.CallbackOptions,
+    ) -> Sequence[metrics.Observation]:
+        return [
+            metrics.Observation(
+                value=len(get_active_user_ids()),
+            )
+        ]
+
+    def observe_total_registered_users(
+        options: metrics.CallbackOptions,
+    ) -> Sequence[metrics.Observation]:
+        return [
+            metrics.Observation(
+                value=len(Users.get_users()["users"]),
+            )
+        ]
+
+    meter.create_observable_gauge(
+        name="webui.users.total",
+        description="Total number of registered users",
+        unit="users",
+        callbacks=[observe_total_registered_users],
+    )
+
+    meter.create_observable_gauge(
+        name="webui.users.active",
+        description="Number of currently active users",
+        unit="users",
+        callbacks=[observe_active_users],
+    )
+
+    # FastAPI middleware
+    @app.middleware("http")
+    async def _metrics_middleware(request: Request, call_next):
+        start_time = time.perf_counter()
+        response = await call_next(request)
+        elapsed_ms = (time.perf_counter() - start_time) * 1000.0
+
+        # Route template e.g. "/items/{item_id}" instead of real path.
+        route = request.scope.get("route")
+        route_path = getattr(route, "path", request.url.path)
+
+        attrs: Dict[str, str | int] = {
+            "http.method": request.method,
+            "http.route": route_path,
+            "http.status_code": response.status_code,
+        }
+
+        request_counter.add(1, attrs)
+        duration_histogram.record(elapsed_ms, attrs)
+
+        return response

backend/open_webui/utils/telemetry/setup.py

@@ -1,23 +1,56 @@
 from fastapi import FastAPI
 from opentelemetry import trace
+
 from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
+from opentelemetry.exporter.otlp.proto.http.trace_exporter import (
+    OTLPSpanExporter as HttpOTLPSpanExporter,
+)
 from opentelemetry.sdk.resources import SERVICE_NAME, Resource
 from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
 from sqlalchemy import Engine
+from base64 import b64encode
 
-from open_webui.utils.telemetry.exporters import LazyBatchSpanProcessor
 from open_webui.utils.telemetry.instrumentors import Instrumentor
-from open_webui.env import OTEL_SERVICE_NAME, OTEL_EXPORTER_OTLP_ENDPOINT
+from open_webui.utils.telemetry.metrics import setup_metrics
+from open_webui.env import (
+    OTEL_SERVICE_NAME,
+    OTEL_EXPORTER_OTLP_ENDPOINT,
+    OTEL_EXPORTER_OTLP_INSECURE,
+    ENABLE_OTEL_METRICS,
+    OTEL_BASIC_AUTH_USERNAME,
+    OTEL_BASIC_AUTH_PASSWORD,
+    OTEL_OTLP_SPAN_EXPORTER,
+)
 
 
 def setup(app: FastAPI, db_engine: Engine):
     # set up trace
-    trace.set_tracer_provider(
-        TracerProvider(
-            resource=Resource.create(attributes={SERVICE_NAME: OTEL_SERVICE_NAME})
-        )
-    )
+    resource = Resource.create(attributes={SERVICE_NAME: OTEL_SERVICE_NAME})
+    trace.set_tracer_provider(TracerProvider(resource=resource))
+
+    # Add basic auth header only if both username and password are not empty
+    headers = []
+    if OTEL_BASIC_AUTH_USERNAME and OTEL_BASIC_AUTH_PASSWORD:
+        auth_string = f"{OTEL_BASIC_AUTH_USERNAME}:{OTEL_BASIC_AUTH_PASSWORD}"
+        auth_header = b64encode(auth_string.encode()).decode()
+        headers = [("authorization", f"Basic {auth_header}")]
+
     # otlp export
-    exporter = OTLPSpanExporter(endpoint=OTEL_EXPORTER_OTLP_ENDPOINT)
-    trace.get_tracer_provider().add_span_processor(LazyBatchSpanProcessor(exporter))
+    if OTEL_OTLP_SPAN_EXPORTER == "http":
+        exporter = HttpOTLPSpanExporter(
+            endpoint=OTEL_EXPORTER_OTLP_ENDPOINT,
+            headers=headers,
+        )
+    else:
+        exporter = OTLPSpanExporter(
+            endpoint=OTEL_EXPORTER_OTLP_ENDPOINT,
+            insecure=OTEL_EXPORTER_OTLP_INSECURE,
+            headers=headers,
+        )
+    trace.get_tracer_provider().add_span_processor(BatchSpanProcessor(exporter))
     Instrumentor(app=app, db_engine=db_engine).instrument()
+
+    # set up metrics only if enabled
+    if ENABLE_OTEL_METRICS:
+        setup_metrics(app, resource)

backend/open_webui/utils/tools.py

@@ -101,9 +101,6 @@ def get_tools(
 
                     def make_tool_function(function_name, token, tool_server_data):
                         async def tool_function(**kwargs):
-                            print(
-                                f"Executing tool function {function_name} with params: {kwargs}"
-                            )
                             return await execute_tool_server(
                                 token=token,
                                 url=tool_server_data["url"],
@@ -380,7 +377,6 @@ def convert_openapi_to_tool_payload(openapi_spec):
         for method, operation in methods.items():
             if operation.get("operationId"):
                 tool = {
-                    "type": "function",
                     "name": operation.get("operationId"),
                     "description": operation.get(
                         "description",
@@ -402,10 +398,16 @@ def convert_openapi_to_tool_payload(openapi_spec):
                         description += (
                             f". Possible values: {', '.join(param_schema.get('enum'))}"
                         )
-                    tool["parameters"]["properties"][param_name] = {
+                    param_property = {
                         "type": param_schema.get("type"),
                         "description": description,
                     }
+
+                    # Include items property for array types (required by OpenAI)
+                    if param_schema.get("type") == "array" and "items" in param_schema:
+                        param_property["items"] = param_schema["items"]
+
+                    tool["parameters"]["properties"][param_name] = param_property
                     if param.get("required"):
                         tool["parameters"]["required"].append(param_name)
 
@@ -479,7 +481,7 @@ async def get_tool_server_data(token: str, url: str) -> Dict[str, Any]:
         "specs": convert_openapi_to_tool_payload(res),
     }
 
-    log.info("Fetched data:", data)
+    log.info(f"Fetched data: {data}")
     return data
 
 
@@ -492,15 +494,7 @@ async def get_tool_servers_data(
         if server.get("config", {}).get("enable"):
             # Path (to OpenAPI spec URL) can be either a full URL or a path to append to the base URL
             openapi_path = server.get("path", "openapi.json")
-            if "://" in openapi_path:
-                # If it contains "://", it's a full URL
-                full_url = openapi_path
-            else:
-                if not openapi_path.startswith("/"):
-                    # Ensure the path starts with a slash
-                    openapi_path = f"/{openapi_path}"
-
-                full_url = f"{server.get('url')}{openapi_path}"
+            full_url = get_tool_server_url(server.get("url"), openapi_path)
 
             info = server.get("info", {})
 
@@ -531,6 +525,8 @@ async def get_tool_servers_data(
         openapi_data = response.get("openapi", {})
 
         if info and isinstance(openapi_data, dict):
+            openapi_data["info"] = openapi_data.get("info", {})
+
             if "name" in info:
                 openapi_data["info"]["title"] = info.get("name", "Tool Server")
 
@@ -644,5 +640,18 @@ async def execute_tool_server(
 
     except Exception as err:
         error = str(err)
-        log.exception("API Request Error:", error)
+        log.exception(f"API Request Error: {error}")
         return {"error": error}
+
+
+def get_tool_server_url(url: Optional[str], path: str) -> str:
+    """
+    Build the full URL for a tool server, given a base url and a path.
+    """
+    if "://" in path:
+        # If it contains "://", it's a full URL
+        return path
+    if not path.startswith("/"):
+        # Ensure the path starts with a slash
+        path = f"/{path}"
+    return f"{url}{path}"

backend/requirements.txt

@@ -1,19 +1,20 @@
 fastapi==0.115.7
-uvicorn[standard]==0.34.0
-pydantic==2.10.6
+uvicorn[standard]==0.35.0
+pydantic==2.11.7
 python-multipart==0.0.20
 
 python-socketio==5.13.0
 python-jose==3.4.0
 passlib[bcrypt]==1.7.4
+cryptography
 
-requests==2.32.3
-aiohttp==3.11.11
+requests==2.32.4
+aiohttp==3.12.15
 async-timeout
 aiocache
 aiofiles
 starlette-compress==1.6.0
-
+httpx[socks,http2,zstd,cli,brotli]==0.28.1
 
 sqlalchemy==2.0.38
 alembic==1.14.0
@@ -26,11 +27,13 @@ bcrypt==4.3.0
 
 pymongo
 redis
-boto3==1.35.53
+boto3==1.40.5
 
 argon2-cffi==23.1.0
 APScheduler==3.10.4
 
+pycrdt==0.12.25
+
 RestrictedPython==8.0
 
 loguru==0.7.3
@@ -39,26 +42,30 @@ asgiref==3.8.1
 # AI libraries
 openai
 anthropic
-google-genai==1.15.0
+google-genai==1.28.0
 google-generativeai==0.8.5
 tiktoken
 
-langchain==0.3.24
-langchain-community==0.3.23
+langchain==0.3.26
+langchain-community==0.3.26
 
-fake-useragent==2.1.0
+fake-useragent==2.2.0
 chromadb==0.6.3
+posthog==5.4.0
 pymilvus==2.5.0
-qdrant-client~=1.12.0
+qdrant-client==1.14.3
 opensearch-py==2.8.0
 playwright==1.49.1 # Caution: version must match docker-compose.playwright.yaml
 elasticsearch==9.0.1
 pinecone==6.0.2
+oracledb==3.2.0
 
+av==14.0.1 # Caution: Set due to FATAL FIPS SELFTEST FAILURE, see discussion https://github.com/open-webui/open-webui/discussions/15720
 transformers
 sentence-transformers==4.1.0
 accelerate
 colbert-ai==0.2.21
+pyarrow==20.0.0
 einops==0.8.1
 
 
@@ -67,22 +74,22 @@ pypdf==4.3.1
 fpdf2==2.8.2
 pymdown-extensions==10.14.2
 docx2txt==0.8
-python-pptx==1.0.0
+python-pptx==1.0.2
 unstructured==0.16.17
 nltk==3.9.1
-Markdown==3.7
+Markdown==3.8.2
 pypandoc==1.15
 pandas==2.2.3
 openpyxl==3.1.5
 pyxlsb==1.0.10
 xlrd==2.0.1
-validators==0.34.0
+validators==0.35.0
 psutil
 sentencepiece
 soundfile==0.13.1
-azure-ai-documentintelligence==1.0.0
+azure-ai-documentintelligence==1.0.2
 
-pillow==11.1.0
+pillow==11.3.0
 opencv-python-headless==4.11.0.86
 rapidocr-onnxruntime==1.4.4
 rank-bm25==0.2.2
@@ -92,16 +99,15 @@ onnxruntime==1.20.1
 faster-whisper==1.1.1
 
 PyJWT[crypto]==2.10.1
-authlib==1.4.1
+authlib==1.6.1
 
 black==25.1.0
 langfuse==2.44.0
-youtube-transcript-api==1.0.3
+youtube-transcript-api==1.1.0
 pytube==15.0.0
 
-extract_msg
 pydub
-duckduckgo-search==8.0.2
+ddgs==9.0.0
 
 ## Google Drive
 google-api-python-client
@@ -116,7 +122,7 @@ pytest-docker~=3.1.1
 googleapis-common-protos==1.63.2
 google-cloud-storage==2.19.0
 
-azure-identity==1.21.0
+azure-identity==1.23.0
 azure-storage-blob==12.24.1
 
 
@@ -130,14 +136,14 @@ firecrawl-py==1.12.0
 tencentcloud-sdk-python==3.0.1336
 
 ## Trace
-opentelemetry-api==1.32.1
-opentelemetry-sdk==1.32.1
-opentelemetry-exporter-otlp==1.32.1
-opentelemetry-instrumentation==0.53b1
-opentelemetry-instrumentation-fastapi==0.53b1
-opentelemetry-instrumentation-sqlalchemy==0.53b1
-opentelemetry-instrumentation-redis==0.53b1
-opentelemetry-instrumentation-requests==0.53b1
-opentelemetry-instrumentation-logging==0.53b1
-opentelemetry-instrumentation-httpx==0.53b1
-opentelemetry-instrumentation-aiohttp-client==0.53b1
+opentelemetry-api==1.36.0
+opentelemetry-sdk==1.36.0
+opentelemetry-exporter-otlp==1.36.0
+opentelemetry-instrumentation==0.57b0
+opentelemetry-instrumentation-fastapi==0.57b0
+opentelemetry-instrumentation-sqlalchemy==0.57b0
+opentelemetry-instrumentation-redis==0.57b0
+opentelemetry-instrumentation-requests==0.57b0
+opentelemetry-instrumentation-logging==0.57b0
+opentelemetry-instrumentation-httpx==0.57b0
+opentelemetry-instrumentation-aiohttp-client==0.57b0