Merge pull request #15448 from open-webui/dev

0.6.16

.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/ISSUE_TEMPLATE/bug_report.yaml

@@ -89,9 +89,20 @@ body:
           required: true
         - label: I have included the Docker container logs.
           required: true
-        - label: I have listed steps to reproduce the bug in detail.
+        - label: I have **provided every relevant configuration, setting, and environment variable used in my setup.**
+          required: true
+        - label: I have clearly **listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup** (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc).
+          required: true
+        - label: |
+            I have documented **step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation**. My steps:
+            - Start with the initial platform/version/OS and dependencies used,
+            - Specify exact install/launch/configure commands,
+            - List URLs visited, user input (incl. example values/emails/passwords if needed),
+            - Describe all options and toggles enabled or changed,
+            - Include any files or environmental changes,
+            - Identify the expected and actual result at each stage,
+            - Ensure any reasonably skilled user can follow and hit the same issue.
           required: true
-
   - type: textarea
     id: expected-behavior
     attributes:
@@ -112,15 +123,25 @@ body:
     id: reproduction-steps
     attributes:
       label: Steps to Reproduce
-      description: Providing clear, step-by-step instructions helps us reproduce and fix the issue faster. If we can't reproduce it, we can't fix it.
+      description: |
+        Please provide a **very detailed, step-by-step guide** to reproduce the issue. Your instructions should be so clear and precise that anyone can follow them without guesswork. Include every relevant detail—settings, configuration options, exact commands used, values entered, and any prerequisites or environment variables.  
+        **If full reproduction steps and all relevant settings are not provided, your issue may not be addressed.**
+
       placeholder: |
-        1. Go to '...'
-        2. Click on '...'
-        3. Scroll down to '...'
-        4. See the error message '...'
+        Example (include every detail):
+        1. Start with a clean Ubuntu 22.04 install.
+        2. Install Docker v24.0.5 and start the service.
+        3. Clone the Open WebUI repo (git clone ...).
+        4. Use the Docker Compose file without modifications.
+        5. Open browser Chrome 115.0 in incognito mode.
+        6. Go to http://localhost:8080 and log in with user "test@example.com".
+        7. Set the language to "English" and theme to "Dark".
+        8. Attempt to connect to Ollama at "http://localhost:11434".
+        9. Observe that the error message "Connection refused" appears at the top right.
+
+        Please list each step carefully and include all relevant configuration, settings, and options.
     validations:
       required: true
-
   - type: textarea
     id: logs-screenshots
     attributes:

.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/release-pypi.yml

@@ -17,6 +17,10 @@ jobs:
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - name: Install Git
+        run: sudo apt-get update && sudo apt-get install -y git
       - uses: actions/setup-node@v4
         with:
           node-version: 22

.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 +5,303 @@ 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).
 
+## [0.6.16] - 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
+
+- 🧩 **Custom Advanced Model Parameters**: You can now add your own tailor-made advanced parameters to any model, empowering you to fine-tune behavior and unlock greater flexibility beyond just the built-in options—accelerate your experimentation.
+- 🪧 **Datalab Marker API Content Extraction Support**: Seamlessly extract content from files and documents using the Datalab Marker API directly in your workflows, enabling more robust structured data extraction for RAG and document processing with just a simple engine switch in the UI.
+- ⚡ **Parallelized Base Model Fetching**: Experience noticeably faster startup and model refresh times—base model data now loads in parallel, drastically shortening delays in busy or large-scale deployments.
+- 🧠 **Efficient Function Loading and Caching**: Functions are now only reloaded if their content changes, preventing unnecessary duplicate loads, saving bandwidth, and boosting performance.
+- 🌍 **Localization & Translation Enhancements**: Improved and expanded Simplified, Traditional Chinese, and Russian translations, providing smoother, more accurate, and context-aware experiences for global users.
+
+### Fixed
+
+- 💬 **Stable Message Input Box**: Fixed an issue where the message input box would shift unexpectedly (especially on mobile or with screen reader support), ensuring a smooth and reliable typing experience for every user.
+- 🔊 **Reliable Read Aloud (Text-to-Speech)**: Read aloud now works seamlessly across messages, so users depending on TTS for accessibility or multitasking will experience uninterrupted and clear voice playback.
+- 🖼 **Image Preview and Download Restored**: Fixed problems with image preview and downloads, ensuring frictionless creation, previewing, and downloading of images in your chats—no more interruptions in creative or documentation workflows.
+- 📱 **Improved Mobile Styling for Workspace Capabilities**: Capabilities management is now readable and easy-to-use even on mobile devices, empowering admins and users to manage access quickly on the go.
+- 🔁 **/api/v1/retrieval/query/collection Endpoint Reliability**: Queries to retrieval collections now return the expected results, bolstering the reliability of your knowledge workflows and citation-ready responses.
+
+### Removed
+
+- 🧹 **Duplicate CSS Elements**: Streamlined the UI by removing redundant CSS, reducing clutter and improving load times for a smoother visual experience.
+
+## [0.6.11] - 2025-05-27
+
+### Added
+
+- 🟢 **Ollama Model Status Indicator in Model Selector**: Instantly see which Ollama models are currently loaded with a clear indicator in the model selector, helping you stay organized and optimize local model usage.
+- 🗑️ **Unload Ollama Model Directly from Model Selector**: Easily release memory and resources by unloading any loaded Ollama model right in the model selector—streamline hardware management without switching pages.
+- 🗣️ **User-Configurable Speech-to-Text Language Setting**: Improve transcription accuracy by letting individual users explicitly set their preferred STT language in their settings—ideal for multilingual teams and clear audio capture.
+- ⚡ **Granular Audio Playback Speed Control**: Instead of just presets, you can now choose granular audio speed using a numeric input, giving you complete control over playback pace in transcriptions and media reviews.
+- 📦 **GZip, Brotli, ZStd Compression Middleware**: Enjoy significantly faster page loads and reduced bandwidth usage with new server-side compression—giving users a snappier, more efficient experience.
+- 🏷️ **Configurable Weight for BM25 in Hybrid Search**: Fine-tune search relevance by adjusting the weight for BM25 inside hybrid search from the UI, letting you tailor knowledge search results to your workflow.
+- 🧪 **Bypass File Creation with CTRL + SHIFT + V**: When “Paste Large Text as File” is enabled, use CTRL + SHIFT + V to skip the file creation dialog and instantly upload text as a file—perfect for rapid document prep.
+- 🌐 **Bypass Web Loader in Web Search**: Choose to bypass web content loading and use snippets directly in web search for faster, more reliable results when page loads are slow or blocked.
+- 🚀 **Environment Variable: WEBUI_AUTH_TRUSTED_GROUPS_HEADER**: Now sync and manage user groups directly via trusted HTTP header, unlocking smoother single sign-on and identity integrations for organizations.
+- 🏢 **Workspace Models Visibility Controls**: You can now hide workspace-level models from both the model selector and shared environments—keep your team focused and reduce clutter from rarely-used endpoints.
+- 🛡️ **Copy Model Link**: You can now copy a direct link to any model—including those hidden from the selector—making sharing and onboarding others more seamless.
+- 🔗 **Load Function Directly from URL**: Simplify custom function management—just paste any GitHub function URL into Open WebUI and import new functions in seconds.
+- ⚙️ **Custom Name/Description for External Tool Servers**: Personalize and clarify external tool servers by assigning custom names and descriptions, making it easier to manage integrations in large-scale workspaces.
+- 🌍 **Custom OpenAPI JSON URL Support for Tool Servers**: Supports specifying any custom OpenAPI JSON URL, unlocking more flexible integration with any backend for tool calls.
+- 📊 **Source Field Now Displays in Non-Streaming Responses with Attachments**: When files or knowledge are attached, the "source" field now appears for all responses, even in non-streaming mode—enabling improved citation workflow.
+- 🎛 **Pinned Chats**: Reduced payload size on pinned chat requests—leading to faster load times and less data usage, especially on busy warehouses.
+- 🛠 **Import/Export Default Prompt Suggestions**: Enjoy one-click import/export of prompt suggestions, making it much easier to share, reuse, and manage best practices across teams or deployments.
+- 🍰 **Banners Now Sortable from Admin Settings**: Quickly re-order or prioritize banners, letting you highlight the most critical info for your team.
+- 🛠 **Advanced Chat Parameters—Clearer Ollama Support Labels**: Parameters and advanced settings now explicitly indicate if they are Ollama-specific, reducing confusion and improving setup accuracy.
+- 🤏 **Scroll Bar Thumb Improved for Better Visibility**: Enhanced scrollbar styling makes navigation more accessible and visually intuitive.
+- 🗄️ **Modal Redesign for Archived and User Chat Listings**: Clean, modern modal interface for browsing archived and user-specific chats makes locating conversations faster and more pleasant.
+- 📝 **Add/Edit Memory Modal UX**: Memory modals are now larger and have resizable input fields, supporting easier editing of long or complex memory content.
+- 🏆 **Translation & Localization Enhancements**: Major upgrades to Chinese (Simplified & Traditional), Korean, Russian, German, Danish, Finnish—not just fixing typos, but consistency, tone, and terminology for a more natural native-language experience.
+- ⚡ **General Backend Stability & Security Enhancements**: Various backend refinements ensure a more resilient, reliable, and secure platform for smoother operation and peace of mind.
+
+### Fixed
+
+- 🖼️ **Image Generation with Allowed File Extensions Now Works Reliably**: Ensure seamless image generation even when strict file extension rules are set—no more blocked creative workflows due to technical hiccups.
+- 🗂 **Remove Leading Dot for File Extension Check**: Fixed an issue where file validation failed because of a leading dot, making file uploads and knowledge management more robust.
+- 🏷️ **Correct Local/External Model Classification**: The platform now accurately distinguishes between local and external models—preventing local models from showing up as external (and vice versa)—ensuring seamless setup, clarity, and management of your AI model endpoints.
+- 📄 **External Document Loader Now Functions as Intended**: External document loaders are reliably invoked, ensuring smoother knowledge ingestion from external sources—expanding your RAG and knowledge workflows.
+- 🎯 **Correct Handling of Toggle Filters**: Toggle filters are now robustly managed, preventing accidental auto-activation and ensuring user preferences are always respected.
+- 🗃 **S3 Tagging Character Restrictions Fixed**: Tags for files in S3 now automatically meet Amazon’s allowed character set, avoiding upload errors and ensuring cross-cloud compatibility.
+- 🛡️ **Authentication Now Uses Password Hash When Duplicate Emails Exist**: Ensures account security and prevents access issues if duplicate emails are present in your system.
+
+### Changed
+
+- 🧩 **Admin Settings: OAuth Redirects Now Use WEBUI_URL**: The OAuth redirect URL is now based on the explicitly set WEBUI_URL, ensuring single sign-on and identity provider integrations always send users to the correct frontend.
+
+### Removed
+
+- 💡 **Duplicate/Typo Component Removals**: Obsolete components have been cleaned up, reducing confusion and improving overall code quality for the team.
+- 🚫 **Streaming Upsert in Pinecone Removed**: Removed streaming upsert references for better compatibility and future-proofing with latest Pinecone SDK updates.
+
+## [0.6.10] - 2025-05-19
+
+### Added
+
+- 🧩 **Experimental Azure OpenAI Support**: Instantly connect to Azure OpenAI endpoints by simply pasting your Azure OpenAI URL into the model connections—bringing flexible, enterprise-grade AI integration directly to your workflow.
+- 💧 **Watermark AI Responses**: Easily add a visible watermark to AI-generated responses for clear content provenance and compliance with EU AI regulations—perfect for regulated environments and transparent communication.
+- 🔍 **Enhanced Search Experience with Dedicated Modal**: Enjoy a modern, powerful search UI in a dedicated modal (open with Ctrl/Cmd + K) accessible from anywhere—quickly find chats, models, or content and boost your productivity.
+- 🔲 **"Toggle" Filter Type for Chat Input**: Add interactive toggle filters (e.g. Web Search, Image, Code Interpreter) right into the chat input—giving you one-click control to activate features and simplifying the chat configuration process.
+- 🧰 **Granular Model Capabilities Editor**: Define detailed capabilities and feature access for each AI model directly from the model editor—enabling tailored behavior, modular control, and a more customized AI environment for every team or use case.
+- 🌐 **Flexible Local and External Connection Support**: Now you can designate any AI connection—whether OpenAI, Ollama, or others—as local or external, enabling seamless setup for on-premise, self-hosted, or cloud configurations and giving you maximum control and flexibility.
+- 🗂️ **Allowed File Extensions for RAG**: Gain full control over your Retrieval-Augmented Generation (RAG) workflows by specifying exactly which file extensions are permitted for upload, improving security and relevance of indexed documents.
+- 🔊 **Enhanced Audio Transcription Logic**: Experience smoother, more reliable audio transcription—very long audio files are now automatically split and processed in segments, preventing errors and ensuring even challenging files are transcribed seamlessly, all part of a broader stability upgrade for robust media workflows.
+- 🦾 **External Document Loader Support**: Enhance knowledge base building by integrating documents using external loaders from a wide range of data sources, expanding what your AI can read and process.
+- 📝 **Preview Button for Code Artifacts**: Instantly jump from an HTML code block to its associated artifacts page with the click of a new preview button—speeding up review and streamlining analysis.
+- 🦻 **Screen Reader Support for Response Messages**: All chat responses are now fully compatible with screen readers, making the platform more inclusive and accessible for everyone.
+- 🧑‍💼 **Customizable Pending User Overlay**: You can now tailor the overlay title and content shown to pending users, ensuring onboarding messaging is perfectly aligned with your organization’s tone and requirements.
+- 🔐 **Option to Disable LDAP Certificate Validation**: You can now disable LDAP certificate validation for maximum flexibility in diverse IT environments—making integrations and troubleshooting much easier.
+- 🎯 **Workspace Search by Username or Email**: Easily search across workspace pages using any username or email address, streamlining user and resource management.
+- 🎨 **High Contrast & Dark Mode Enhancements**: Further improved placeholder, input, suggestion, toast, and model selector contrasts—including a dedicated placeholder dark mode—for more comfortable viewing in all lighting conditions.
+- 🛡️ **Refined Security for Pipelines & Model Uploads**: Strengthened safeguards against path traversal vulnerabilities during uploads—ensuring your platform’s document and model management remains secure.
+- 🌏 **Major Localization Upgrades**: Comprehensive translation updates and improvements across Korean, Bulgarian, Catalan, Japanese, Italian, Traditional Chinese, and Spanish—including more accurate AI terminology for clarity; your experience is now more natural, inclusive, and professional everywhere.
+- 🦾 **General Backend Stability & Security**: Multiple backend improvements (including file upload, command navigation, and logging refactorings) deliver increased resilience, better error handling, and a more robust platform for all users.
+
+### Fixed
+
+- ✅ **Evaluation Feedback Endpoint Reliability**: Addressed issues with feedback submission endpoints to ensure seamless user feedback collection on model responses.
+- 🫰 **Model List State Fixes**: Resolved issues where model status toggles in the workspace page might inadvertently switch or confuse state, making the management of active/inactive models more dependable.
+- ✍️ **Admin Signup Logic**: Admin self-signup experience and validation flow is smoother and more robust.
+- 🔁 **Signout Redirect Flow Improved**: Logging out now redirects more reliably, reducing confusion and making session management seamless.
+
+## [0.6.9] - 2025-05-10
+
+### Added
+
+- 📝 **Edit Attached Images/Files in Messages**: You can now easily edit your sent messages by removing attached files—streamlining document management, correcting mistakes on the fly, and keeping your chats clutter-free.
+- 🚨 **Clear Alerts for Private Task Models**: When interacting with private task models, the UI now clearly alerts you—making it easier to understand resource availability and access, reducing confusion during workflow setup.
+
+### Fixed
+
+- 🛡️ **Confirm Dialog Focus Trap Reliability**: The focus now stays correctly within confirmation dialogs, ensuring keyboard navigation and accessibility is seamless and preventing accidental operations—especially helpful during critical or rapid workflows.
+- 💬 **Temporary Chat Admin Controls & Session Cleanliness**: Admins are now able to properly enable temporary chat mode without errors, and previous session prompts or tool selections no longer carry over—delivering a fresh, predictable, and consistent temporary chat experience every time.
+- 🤖 **External Reranker Integration Functionality Restored**: External reranker integrations now work correctly, allowing you to fully leverage advanced ranking services for sharper, more relevant search results in your RAG and knowledge base workflows.
+
+## [0.6.8] - 2025-05-10
+
+### Added
+
+- 🏆 **External Reranker Support for Knowledge Base Search**: Supercharge your Retrieval-Augmented Generation (RAG) workflows with the new External Reranker integration; easily plug in advanced reranking services via the UI to deliver sharper and more relevant search results, accelerating research and insight discovery.
+- 📤 **Unstylized PDF Export Option (Reduced File Size)**: When exporting chat transcripts or documents, you can now choose an unstylized PDF export for snappier downloads, minimal file size, and clean data archiving—perfect for large-scale storage or sharing.
+- 📝 **Vazirmatn Font for Persian & Arabic**: Arabic and Persian users will now see their text beautifully rendered with the specialized Vazirmatn font for an improved localized reading experience.
+- 🏷️ **SharePoint Tenant ID Support for OneDrive**: You can now specify a SharePoint tenant ID in OneDrive settings for seamless authentication and granular enterprise integration.
+- 👤 **Refresh OAuth Profile Picture**: Your OAuth profile picture now updates in real-time, ensuring your presence and avatar always match your latest identity across integrated platforms.
+- 🔧 **Milvus Configuration Improvements**: Configure index and metric types for Milvus directly within settings; take full control of your vector database for more accurate and robust AI search experiences.
+- 🛡️ **S3 Tagging Toggle for Compatibility**: Optional S3 tagging via an environment toggle grants full compatibility with all storage backends—including those that don’t support tagging like Cloudflare R2—ensuring error-free attachment and document management.
+- 👨‍🦯 **Icon Button Accessibility Improvements**: Key interactive icon-buttons now include aria-labels and ARIA descriptions, so screen readers provide precise guidance about what action each button performs for improved accessibility.
+- ♿ **Enhanced Accessibility with Modal Focus Trap**: Modal dialogs and pop-ups now feature a focus trap and improved ARIA roles, ensuring seamless navigation and screen reader support—making the interface friendlier for everyone, including keyboard and assistive tech users.
+- 🏃 **Improved Admin User List Loading Indicator**: The user list loading experience is now clearer and more responsive in the admin panel.
+- 🧑‍🤝‍🧑 **Larger Admin User List Page Size**: Admins can now manage up to 30 users per page in the admin interface, drastically reducing pagination and making large user teams easier and faster to manage.
+- 🌠 **Default Code Interpreter Prompt Clarified**: The built-in code interpreter prompt is now more explicit, preventing AI from wrapping code in Markdown blocks when not needed—ensuring properly formatted code runs as intended every time.
+- 🧾 **Improved Default Title Generation Prompt Template**: Title generation now uses a robust template for reliable JSON output, improving chat organization and searchability.
+- 🔗 **Support Jupyter Notebooks with Non-Root Base URLs**: Notebook-based code execution now supports non-root deployed Jupyter servers, granting full flexibility for hybrid or multi-user setups.
+- 📰 **UI Scrollbar Always Visible for Overflow Tools**: When available tools overflow the display, the scrollbar is now always visible and there’s a handy "show all" toggle, making navigation of large toolsets snappier and more intuitive.
+- 🛠️ **General Backend Refactoring for Stability**: Multiple under-the-hood improvements have been made across backend components, ensuring smoother performance, fewer errors, and a more reliable overall experience for all users.
+- 🚀 **Optimized Web Search for Faster Results**: Web search speed and performance have been significantly enhanced, delivering answers and sources in record time to accelerate your research-heavy workflows.
+- 💡 **More Supported Languages**: Expanded language support ensures an even wider range of users can enjoy an intuitive and natural interface in their native tongue.
+
+### Fixed
+
+- 🏃‍♂️ **Exhausting Workers in Nginx Reverse Proxy Due to Websocket Fix**: Websocket sessions are now fully compatible behind Nginx, eliminating worker exhaustion and restoring 24/7 reliability for real-time chats even in complex deployments.
+- 🎤 **Audio Transcription Issue with OpenAI Resolved**: OpenAI-based audio transcription now handles WebM and newer formats without error, ensuring seamless voice-to-text workflows every time.
+- 👉 **Message Input RTL Issue Fixed**: The chat message input now displays correctly for right-to-left languages, creating a flawless typing and reading experience for Arabic, Hebrew, and more.
+- 🀄 **Katex: Proper Rendering of Chinese Characters Next to Math**: Math formulas now render perfectly even when directly adjacent to Chinese (CJK) characters, improving visual clarity for multilingual teams and cross-language documents.
+- 🔂 **Duplicate Web Search URLs Eliminated**: Search results now reliably filter out URL duplicates, so your knowledge and search citations are always clean, trimmed, and easy to review.
+- 📄 **Markdown Rendering Fixed in Knowledge Bases**: Markdown is now displayed correctly within knowledge bases, enabling better formatting and clarity of information-rich files.
+- 🗂️ **LDAP Import/Loading Issue Resolved**: LDAP user imports process correctly, ensuring smooth onboarding and access without interruption.
+- 🌎 **Pinecone Batch Operations and Async Safety**: All Pinecone operations (batch insert, upsert, delete) now run efficiently and safely in an async environment, boosting performance and preventing slowdowns in large-scale RAG jobs.
+
 ## [0.6.7] - 2025-05-07
 
 ### Added

Caddyfile.localhost

@@ -1,64 +0,0 @@
-# Run with
-#    caddy run --envfile ./example.env --config ./Caddyfile.localhost
-#
-# This is configured for
-#    - Automatic HTTPS (even for localhost)
-#    - Reverse Proxying to Ollama API Base URL (http://localhost:11434/api)
-#    - CORS
-#    - HTTP Basic Auth API Tokens (uncomment basicauth section)
-
-
-# CORS Preflight (OPTIONS) + Request (GET, POST, PATCH, PUT, DELETE)
-(cors-api) {
-	@match-cors-api-preflight method OPTIONS
-	handle @match-cors-api-preflight {
-		header {
-			Access-Control-Allow-Origin "{http.request.header.origin}"
-			Access-Control-Allow-Methods "GET, POST, PUT, PATCH, DELETE, OPTIONS"
-			Access-Control-Allow-Headers "Origin, Accept, Authorization, Content-Type, X-Requested-With"
-			Access-Control-Allow-Credentials "true"
-			Access-Control-Max-Age "3600"
-			defer
-		}
-		respond "" 204
-	}
-
-	@match-cors-api-request {
-		not {
-			header Origin "{http.request.scheme}://{http.request.host}"
-		}
-		header Origin "{http.request.header.origin}"
-	}
-	handle @match-cors-api-request {
-		header {
-			Access-Control-Allow-Origin "{http.request.header.origin}"
-			Access-Control-Allow-Methods "GET, POST, PUT, PATCH, DELETE, OPTIONS"
-			Access-Control-Allow-Headers "Origin, Accept, Authorization, Content-Type, X-Requested-With"
-			Access-Control-Allow-Credentials "true"
-			Access-Control-Max-Age "3600"
-			defer
-		}
-	}
-}
-
-# replace localhost with example.com or whatever
-localhost {
-	## HTTP Basic Auth
-	## (uncomment to enable)
-	# basicauth {
-	# 	# see .example.env for how to generate tokens
-	# 	{env.OLLAMA_API_ID} {env.OLLAMA_API_TOKEN_DIGEST}
-	# }
-
-	handle /api/* {
-		# Comment to disable CORS
-		import cors-api
-
-		reverse_proxy localhost:11434
-	}
-
-	# Same-Origin Static Web Server
-	file_server {
-		root ./build/
-	}
-}

Dockerfile

@@ -26,6 +26,9 @@ ARG BUILD_HASH
 
 WORKDIR /app
 
+# to store git revision in build
+RUN apk add --no-cache git
+
 COPY package.json package-lock.json ./
 RUN npm ci
 

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]  
@@ -73,7 +75,17 @@ Want to learn more about Open WebUI's features? Check out our [Open WebUI docume
       </a>
     </td>
     <td>
-      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://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>
+      <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>
@@ -171,6 +183,8 @@ 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/open_webui/config.py

@@ -13,12 +13,15 @@ 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,
@@ -109,54 +112,7 @@ if os.path.exists(f"{DATA_DIR}/config.json"):
 
 DEFAULT_CONFIG = {
     "version": 0,
-    "ui": {
-        "default_locale": "",
-        "prompt_suggestions": [
-            {
-                "title": [
-                    "Help me study",
-                    "vocabulary for a college entrance exam",
-                ],
-                "content": "Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option.",
-            },
-            {
-                "title": [
-                    "Give me ideas",
-                    "for what to do with my kids' art",
-                ],
-                "content": "What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter.",
-            },
-            {
-                "title": ["Tell me a fun fact", "about the Roman Empire"],
-                "content": "Tell me a random fun fact about the Roman Empire",
-            },
-            {
-                "title": [
-                    "Show me a code snippet",
-                    "of a website's sticky header",
-                ],
-                "content": "Show me a code snippet of a website's sticky header in CSS and JavaScript.",
-            },
-            {
-                "title": [
-                    "Explain options trading",
-                    "if I'm familiar with buying and selling stocks",
-                ],
-                "content": "Explain options trading in simple terms if I'm familiar with buying and selling stocks.",
-            },
-            {
-                "title": ["Overcome procrastination", "give me tips"],
-                "content": "Could you start by asking me about instances when I procrastinate the most and then give me some suggestions to overcome it?",
-            },
-            {
-                "title": [
-                    "Grammar check",
-                    "rewrite it for better readability ",
-                ],
-                "content": 'Check the following sentence for grammar and clarity: "[sentence]". Rewrite it for better readability while maintaining its original meaning.',
-            },
-        ],
-    },
+    "ui": {},
 }
 
 
@@ -258,11 +214,16 @@ class PersistentConfig(Generic[T]):
 class AppConfig:
     _state: dict[str, PersistentConfig]
     _redis: Optional[redis.Redis] = 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_key_prefix: str = "open-webui",
     ):
         super().__setattr__("_state", {})
+        super().__setattr__("_redis_key_prefix", redis_key_prefix)
         if redis_url:
             super().__setattr__(
                 "_redis",
@@ -277,7 +238,7 @@ 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):
@@ -286,7 +247,7 @@ 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:
@@ -394,6 +355,24 @@ 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",
@@ -460,6 +439,12 @@ 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_CODE_CHALLENGE_METHOD = PersistentConfig(
     "OAUTH_CODE_CHALLENGE_METHOD",
     "oauth.oidc.code_challenge_method",
@@ -552,18 +537,31 @@ OAUTH_ALLOWED_DOMAINS = PersistentConfig(
     ],
 )
 
+OAUTH_UPDATE_PICTURE_ON_LOGIN = PersistentConfig(
+    "OAUTH_UPDATE_PICTURE_ON_LOGIN",
+    "oauth.update_picture_on_login",
+    os.environ.get("OAUTH_UPDATE_PICTURE_ON_LOGIN", "False").lower() == "true",
+)
+
 
 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,
             )
 
@@ -578,27 +576,32 @@ 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,
@@ -607,7 +610,14 @@ 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,
             )
 
@@ -623,9 +633,12 @@ 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,
+                **(
+                    {"timeout": int(OAUTH_TIMEOUT.value)} if OAUTH_TIMEOUT.value else {}
+                ),
             }
 
             if (
@@ -761,9 +774,10 @@ S3_BUCKET_NAME = os.environ.get("S3_BUCKET_NAME", None)
 S3_KEY_PREFIX = os.environ.get("S3_KEY_PREFIX", None)
 S3_ENDPOINT_URL = os.environ.get("S3_ENDPOINT_URL", None)
 S3_USE_ACCELERATE_ENDPOINT = (
-    os.environ.get("S3_USE_ACCELERATE_ENDPOINT", "False").lower() == "true"
+    os.environ.get("S3_USE_ACCELERATE_ENDPOINT", "false").lower() == "true"
 )
 S3_ADDRESSING_STYLE = os.environ.get("S3_ADDRESSING_STYLE", None)
+S3_ENABLE_TAGGING = os.getenv("S3_ENABLE_TAGGING", "false").lower() == "true"
 
 GCS_BUCKET_NAME = os.environ.get("GCS_BUCKET_NAME", None)
 GOOGLE_APPLICATION_CREDENTIALS_JSON = os.environ.get(
@@ -917,6 +931,18 @@ 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
 ####################################
@@ -941,9 +967,7 @@ 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(
@@ -1029,6 +1053,26 @@ DEFAULT_USER_ROLE = PersistentConfig(
     os.getenv("DEFAULT_USER_ROLE", "pending"),
 )
 
+PENDING_USER_OVERLAY_TITLE = PersistentConfig(
+    "PENDING_USER_OVERLAY_TITLE",
+    "ui.pending_user_overlay_title",
+    os.environ.get("PENDING_USER_OVERLAY_TITLE", ""),
+)
+
+PENDING_USER_OVERLAY_CONTENT = PersistentConfig(
+    "PENDING_USER_OVERLAY_CONTENT",
+    "ui.pending_user_overlay_content",
+    os.environ.get("PENDING_USER_OVERLAY_CONTENT", ""),
+)
+
+
+RESPONSE_WATERMARK = PersistentConfig(
+    "RESPONSE_WATERMARK",
+    "ui.watermark",
+    os.environ.get("RESPONSE_WATERMARK", ""),
+)
+
+
 USER_PERMISSIONS_WORKSPACE_MODELS_ACCESS = (
     os.environ.get("USER_PERMISSIONS_WORKSPACE_MODELS_ACCESS", "False").lower()
     == "true"
@@ -1081,6 +1125,10 @@ USER_PERMISSIONS_CHAT_CONTROLS = (
     os.environ.get("USER_PERMISSIONS_CHAT_CONTROLS", "True").lower() == "true"
 )
 
+USER_PERMISSIONS_CHAT_SYSTEM_PROMPT = (
+    os.environ.get("USER_PERMISSIONS_CHAT_SYSTEM_PROMPT", "True").lower() == "true"
+)
+
 USER_PERMISSIONS_CHAT_FILE_UPLOAD = (
     os.environ.get("USER_PERMISSIONS_CHAT_FILE_UPLOAD", "True").lower() == "true"
 )
@@ -1166,6 +1214,7 @@ DEFAULT_USER_PERMISSIONS = {
     },
     "chat": {
         "controls": USER_PERMISSIONS_CHAT_CONTROLS,
+        "system_prompt": USER_PERMISSIONS_CHAT_SYSTEM_PROMPT,
         "file_upload": USER_PERMISSIONS_CHAT_FILE_UPLOAD,
         "delete": USER_PERMISSIONS_CHAT_DELETE,
         "edit": USER_PERMISSIONS_CHAT_EDIT,
@@ -1255,13 +1304,16 @@ ENABLE_USER_WEBHOOKS = PersistentConfig(
 )
 
 # FastAPI / AnyIO settings
-THREAD_POOL_SIZE = int(os.getenv("THREAD_POOL_SIZE", "0"))
+THREAD_POOL_SIZE = os.getenv("THREAD_POOL_SIZE", None)
 
-
-def validate_cors_origins(origins):
-    for origin in origins:
-        if origin != "*":
-            validate_cors_origin(origin)
+if THREAD_POOL_SIZE is not None and isinstance(THREAD_POOL_SIZE, str):
+    try:
+        THREAD_POOL_SIZE = int(THREAD_POOL_SIZE)
+    except ValueError:
+        log.warning(
+            f"THREAD_POOL_SIZE is not a valid integer: {THREAD_POOL_SIZE}. Defaulting to None."
+        )
+        THREAD_POOL_SIZE = None
 
 
 def validate_cors_origin(origin):
@@ -1283,16 +1335,17 @@ 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:
+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):
@@ -1357,6 +1410,9 @@ Generate a concise, 3-5 word title with an emoji summarizing the chat history.
 - Use emojis that enhance understanding of the topic, but avoid quotation marks or special formatting.
 - Write the title in the chat's primary language; default to English if multilingual.
 - Prioritize accuracy over excessive creativity; keep it clear and simple.
+- Your entire response must consist solely of the JSON object, without any introductory or concluding text.
+- The output must be a single, raw JSON object, without any markdown code fences or other encapsulating text.
+- Ensure no conversational text, affirmations, or explanations precede or follow the raw JSON output, as this will cause direct parsing failure.
 ### Output:
 JSON format: { "title": "your concise title here" }
 ### Examples:
@@ -1421,6 +1477,35 @@ 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",
@@ -1699,7 +1784,8 @@ DEFAULT_CODE_INTERPRETER_PROMPT = """
 1. **Code Interpreter**: `<code_interpreter type="code" lang="python"></code_interpreter>`
    - You have access to a Python shell that runs directly in the user's browser, enabling fast execution of code for analysis, calculations, or problem-solving.  Use it in this response.
    - The Python code you write can incorporate a wide array of libraries, handle data manipulation or visualization, perform API calls for web-related tasks, or tackle virtually any computational challenge. Use this flexibility to **think outside the box, craft elegant solutions, and harness Python's full potential**.
-   - To use it, **you must enclose your code within `<code_interpreter type="code" lang="python">` XML tags** and stop right away. If you don't, the code won't execute. Do NOT use triple backticks.
+   - To use it, **you must enclose your code within `<code_interpreter type="code" lang="python">` XML tags** and stop right away. If you don't, the code won't execute. 
+   - When writing code in the code_interpreter XML tag, Do NOT use the triple backticks code block for markdown formatting, example: ```py # python code ``` will cause an error because it is markdown formatting, it is not python code.
    - When coding, **always aim to print meaningful outputs** (e.g., results, tables, summaries, or visuals) to better interpret and verify the findings. Avoid relying on implicit outputs; prioritize explicit and clear print statements so the results are effectively communicated to the user.  
    - After obtaining the printed output, **always provide a concise analysis, interpretation, or next steps to help the user understand the findings or refine the outcome further.**  
    - If the results are unclear, unexpected, or require validation, refine the code and execute it again as needed. Always aim to deliver meaningful insights from the results, iterating if necessary.  
@@ -1746,12 +1832,22 @@ MILVUS_URI = os.environ.get("MILVUS_URI", f"{DATA_DIR}/vector_db/milvus.db")
 MILVUS_DB = os.environ.get("MILVUS_DB", "default")
 MILVUS_TOKEN = os.environ.get("MILVUS_TOKEN", None)
 
+MILVUS_INDEX_TYPE = os.environ.get("MILVUS_INDEX_TYPE", "HNSW")
+MILVUS_METRIC_TYPE = os.environ.get("MILVUS_METRIC_TYPE", "COSINE")
+MILVUS_HNSW_M = int(os.environ.get("MILVUS_HNSW_M", "16"))
+MILVUS_HNSW_EFCONSTRUCTION = int(os.environ.get("MILVUS_HNSW_EFCONSTRUCTION", "100"))
+MILVUS_IVF_FLAT_NLIST = int(os.environ.get("MILVUS_IVF_FLAT_NLIST", "128"))
+
 # Qdrant
 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"))
+ENABLE_QDRANT_MULTITENANCY_MODE = (
+    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")
@@ -1783,6 +1879,52 @@ 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)
@@ -1833,6 +1975,11 @@ ONEDRIVE_SHAREPOINT_URL = PersistentConfig(
     os.environ.get("ONEDRIVE_SHAREPOINT_URL", ""),
 )
 
+ONEDRIVE_SHAREPOINT_TENANT_ID = PersistentConfig(
+    "ONEDRIVE_SHAREPOINT_TENANT_ID",
+    "onedrive.sharepoint_tenant_id",
+    os.environ.get("ONEDRIVE_SHAREPOINT_TENANT_ID", ""),
+)
 
 # RAG Content Extraction
 CONTENT_EXTRACTION_ENGINE = PersistentConfig(
@@ -1841,6 +1988,73 @@ CONTENT_EXTRACTION_ENGINE = PersistentConfig(
     os.environ.get("CONTENT_EXTRACTION_ENGINE", "").lower(),
 )
 
+DATALAB_MARKER_API_KEY = PersistentConfig(
+    "DATALAB_MARKER_API_KEY",
+    "rag.datalab_marker_api_key",
+    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_USE_LLM = PersistentConfig(
+    "DATALAB_MARKER_USE_LLM",
+    "rag.DATALAB_MARKER_USE_LLM",
+    os.environ.get("DATALAB_MARKER_USE_LLM", "false").lower() == "true",
+)
+
+DATALAB_MARKER_SKIP_CACHE = PersistentConfig(
+    "DATALAB_MARKER_SKIP_CACHE",
+    "rag.datalab_marker_skip_cache",
+    os.environ.get("DATALAB_MARKER_SKIP_CACHE", "false").lower() == "true",
+)
+
+DATALAB_MARKER_FORCE_OCR = PersistentConfig(
+    "DATALAB_MARKER_FORCE_OCR",
+    "rag.datalab_marker_force_ocr",
+    os.environ.get("DATALAB_MARKER_FORCE_OCR", "false").lower() == "true",
+)
+
+DATALAB_MARKER_PAGINATE = PersistentConfig(
+    "DATALAB_MARKER_PAGINATE",
+    "rag.datalab_marker_paginate",
+    os.environ.get("DATALAB_MARKER_PAGINATE", "false").lower() == "true",
+)
+
+DATALAB_MARKER_STRIP_EXISTING_OCR = PersistentConfig(
+    "DATALAB_MARKER_STRIP_EXISTING_OCR",
+    "rag.datalab_marker_strip_existing_ocr",
+    os.environ.get("DATALAB_MARKER_STRIP_EXISTING_OCR", "false").lower() == "true",
+)
+
+DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = PersistentConfig(
+    "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION",
+    "rag.datalab_marker_disable_image_extraction",
+    os.environ.get("DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION", "false").lower()
+    == "true",
+)
+
+DATALAB_MARKER_OUTPUT_FORMAT = PersistentConfig(
+    "DATALAB_MARKER_OUTPUT_FORMAT",
+    "rag.datalab_marker_output_format",
+    os.environ.get("DATALAB_MARKER_OUTPUT_FORMAT", "markdown"),
+)
+
+EXTERNAL_DOCUMENT_LOADER_URL = PersistentConfig(
+    "EXTERNAL_DOCUMENT_LOADER_URL",
+    "rag.external_document_loader_url",
+    os.environ.get("EXTERNAL_DOCUMENT_LOADER_URL", ""),
+)
+
+EXTERNAL_DOCUMENT_LOADER_API_KEY = PersistentConfig(
+    "EXTERNAL_DOCUMENT_LOADER_API_KEY",
+    "rag.external_document_loader_api_key",
+    os.environ.get("EXTERNAL_DOCUMENT_LOADER_API_KEY", ""),
+)
+
 TIKA_SERVER_URL = PersistentConfig(
     "TIKA_SERVER_URL",
     "rag.tika_server_url",
@@ -1865,6 +2079,46 @@ DOCLING_OCR_LANG = PersistentConfig(
     os.getenv("DOCLING_OCR_LANG", "eng,fra,deu,spa"),
 )
 
+DOCLING_DO_PICTURE_DESCRIPTION = PersistentConfig(
+    "DOCLING_DO_PICTURE_DESCRIPTION",
+    "rag.docling_do_picture_description",
+    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",
@@ -1903,6 +2157,11 @@ RAG_RELEVANCE_THRESHOLD = PersistentConfig(
     "rag.relevance_threshold",
     float(os.environ.get("RAG_RELEVANCE_THRESHOLD", "0.0")),
 )
+RAG_HYBRID_BM25_WEIGHT = PersistentConfig(
+    "RAG_HYBRID_BM25_WEIGHT",
+    "rag.hybrid_bm25_weight",
+    float(os.environ.get("RAG_HYBRID_BM25_WEIGHT", "0.5")),
+)
 
 ENABLE_RAG_HYBRID_SEARCH = PersistentConfig(
     "ENABLE_RAG_HYBRID_SEARCH",
@@ -1936,6 +2195,37 @@ 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",
+    [
+        ext.strip()
+        for ext in os.environ.get("RAG_ALLOWED_FILE_EXTENSIONS", "").split(",")
+        if ext.strip()
+    ],
+)
+
 RAG_EMBEDDING_ENGINE = PersistentConfig(
     "RAG_EMBEDDING_ENGINE",
     "rag.embedding_engine",
@@ -1981,6 +2271,12 @@ RAG_EMBEDDING_PREFIX_FIELD_NAME = os.environ.get(
     "RAG_EMBEDDING_PREFIX_FIELD_NAME", None
 )
 
+RAG_RERANKING_ENGINE = PersistentConfig(
+    "RAG_RERANKING_ENGINE",
+    "rag.reranking_engine",
+    os.environ.get("RAG_RERANKING_ENGINE", ""),
+)
+
 RAG_RERANKING_MODEL = PersistentConfig(
     "RAG_RERANKING_MODEL",
     "rag.reranking_model",
@@ -1989,6 +2285,7 @@ RAG_RERANKING_MODEL = PersistentConfig(
 if RAG_RERANKING_MODEL.value != "":
     log.info(f"Reranking model set: {RAG_RERANKING_MODEL.value}")
 
+
 RAG_RERANKING_MODEL_AUTO_UPDATE = (
     not OFFLINE_MODE
     and os.environ.get("RAG_RERANKING_MODEL_AUTO_UPDATE", "True").lower() == "true"
@@ -1998,6 +2295,18 @@ RAG_RERANKING_MODEL_TRUST_REMOTE_CODE = (
     os.environ.get("RAG_RERANKING_MODEL_TRUST_REMOTE_CODE", "True").lower() == "true"
 )
 
+RAG_EXTERNAL_RERANKER_URL = PersistentConfig(
+    "RAG_EXTERNAL_RERANKER_URL",
+    "rag.external_reranker_url",
+    os.environ.get("RAG_EXTERNAL_RERANKER_URL", ""),
+)
+
+RAG_EXTERNAL_RERANKER_API_KEY = PersistentConfig(
+    "RAG_EXTERNAL_RERANKER_API_KEY",
+    "rag.external_reranker_api_key",
+    os.environ.get("RAG_EXTERNAL_RERANKER_API_KEY", ""),
+)
+
 
 RAG_TEXT_SPLITTER = PersistentConfig(
     "RAG_TEXT_SPLITTER",
@@ -2070,6 +2379,22 @@ 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",
@@ -2123,6 +2448,12 @@ BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL = PersistentConfig(
 )
 
 
+BYPASS_WEB_SEARCH_WEB_LOADER = PersistentConfig(
+    "BYPASS_WEB_SEARCH_WEB_LOADER",
+    "rag.web.search.bypass_web_loader",
+    os.getenv("BYPASS_WEB_SEARCH_WEB_LOADER", "False").lower() == "true",
+)
+
 WEB_SEARCH_RESULT_COUNT = PersistentConfig(
     "WEB_SEARCH_RESULT_COUNT",
     "rag.web.search.result_count",
@@ -2148,6 +2479,7 @@ WEB_SEARCH_CONCURRENT_REQUESTS = PersistentConfig(
     int(os.getenv("WEB_SEARCH_CONCURRENT_REQUESTS", "10")),
 )
 
+
 WEB_LOADER_ENGINE = PersistentConfig(
     "WEB_LOADER_ENGINE",
     "rag.web.loader.engine",
@@ -2307,6 +2639,18 @@ 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",
@@ -2671,6 +3015,18 @@ 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",
@@ -2836,6 +3192,31 @@ LDAP_CA_CERT_FILE = PersistentConfig(
     os.environ.get("LDAP_CA_CERT_FILE", ""),
 )
 
+LDAP_VALIDATE_CERT = PersistentConfig(
+    "LDAP_VALIDATE_CERT",
+    "ldap.server.validate_cert",
+    os.environ.get("LDAP_VALIDATE_CERT", "True").lower() == "true",
+)
+
 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

@@ -5,6 +5,7 @@ import os
 import pkgutil
 import sys
 import shutil
+from uuid import uuid4
 from pathlib import Path
 
 import markdown
@@ -130,6 +131,7 @@ 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 +199,7 @@ CHANGELOG = changelog_json
 
 SAFE_MODE = os.environ.get("SAFE_MODE", "false").lower() == "true"
 
+
 ####################################
 # ENABLE_FORWARD_USER_INFO_HEADERS
 ####################################
@@ -264,21 +267,43 @@ 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}"
+if DATABASE_CRED:
+    DATABASE_CRED += "@"
+
+
+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']}"
+
 # 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)
 
@@ -323,6 +348,7 @@ ENABLE_REALTIME_CHAT_SAVE = (
 ####################################
 
 REDIS_URL = os.environ.get("REDIS_URL", "")
+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")
 
@@ -349,6 +375,10 @@ WEBUI_AUTH_TRUSTED_EMAIL_HEADER = os.environ.get(
     "WEBUI_AUTH_TRUSTED_EMAIL_HEADER", None
 )
 WEBUI_AUTH_TRUSTED_NAME_HEADER = os.environ.get("WEBUI_AUTH_TRUSTED_NAME_HEADER", None)
+WEBUI_AUTH_TRUSTED_GROUPS_HEADER = os.environ.get(
+    "WEBUI_AUTH_TRUSTED_GROUPS_HEADER", None
+)
+
 
 BYPASS_MODEL_ACCESS_CONTROL = (
     os.environ.get("BYPASS_MODEL_ACCESS_CONTROL", "False").lower() == "true"
@@ -390,10 +420,33 @@ 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"
+)
+
+####################################
+# 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
+
+
+####################################
+# 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)
@@ -500,11 +553,14 @@ 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
@@ -513,6 +569,14 @@ 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:
@@ -533,9 +597,13 @@ 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"
 OTEL_EXPORTER_OTLP_ENDPOINT = os.environ.get(
     "OTEL_EXPORTER_OTLP_ENDPOINT", "http://localhost:4317"
 )
+OTEL_EXPORTER_OTLP_INSECURE = (
+    os.environ.get("OTEL_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", ""
@@ -543,6 +611,14 @@ 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_OTLP_SPAN_EXPORTER = os.environ.get(
+    "OTEL_OTLP_SPAN_EXPORTER", "grpc"
+).lower()  # grpc or http
+
 
 ####################################
 # TOOLS/FUNCTIONS PIP OPTIONS

backend/open_webui/functions.py

@@ -25,10 +25,14 @@ 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
 
-from open_webui.utils.plugin import load_function_module_by_id
+from open_webui.utils.plugin import (
+    load_function_module_by_id,
+    get_function_module_from_cache,
+)
 from open_webui.utils.tools import get_tools
 from open_webui.utils.access_control import has_access
 
@@ -53,12 +57,7 @@ log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
 
 def get_function_module_by_id(request: Request, pipe_id: str):
-    # Check if function is already loaded
-    if pipe_id not in request.app.state.FUNCTIONS:
-        function_module, _, _ = load_function_module_by_id(pipe_id)
-        request.app.state.FUNCTIONS[pipe_id] = function_module
-    else:
-        function_module = request.app.state.FUNCTIONS[pipe_id]
+    function_module, _, _ = get_function_module_from_cache(request, pipe_id)
 
     if hasattr(function_module, "valves") and hasattr(function_module, "Valves"):
         valves = Functions.get_function_valves_by_id(pipe_id)
@@ -229,12 +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,
     }
@@ -255,8 +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

@@ -62,6 +62,9 @@ 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
@@ -81,20 +84,23 @@ if "sqlite" in SQLALCHEMY_DATABASE_URL:
         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

@@ -43,7 +43,7 @@ class ReconnectingPostgresqlDatabase(CustomReconnectMixin, PostgresqlDatabase):
 
 
 def register_connection(db_url):
-    db = connect(db_url, unquote_password=True)
+    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
@@ -51,7 +51,7 @@ def register_connection(db_url):
         log.info("Connected to PostgreSQL database")
 
         # Get the connection details
-        connection = parse(db_url, unquote_password=True)
+        connection = parse(db_url, unquote_user=True, unquote_password=True)
 
         # Use our custom database class that supports reconnection
         db = ReconnectingPostgresqlDatabase(**connection)

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 (
@@ -33,17 +36,19 @@ 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
+
 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
@@ -52,6 +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,
@@ -82,6 +89,7 @@ from open_webui.routers import (
 
 from open_webui.routers.retrieval import (
     get_embedding_function,
+    get_reranking_function,
     get_ef,
     get_rf,
 )
@@ -103,11 +111,14 @@ from open_webui.config import (
     ENABLE_OPENAI_API,
     ONEDRIVE_CLIENT_ID,
     ONEDRIVE_SHAREPOINT_URL,
+    ONEDRIVE_SHAREPOINT_TENANT_ID,
     OPENAI_API_BASE_URLS,
     OPENAI_API_KEYS,
     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
@@ -151,6 +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,
@@ -187,30 +199,55 @@ from open_webui.config import (
     RAG_EMBEDDING_MODEL,
     RAG_EMBEDDING_MODEL_AUTO_UPDATE,
     RAG_EMBEDDING_MODEL_TRUST_REMOTE_CODE,
+    RAG_RERANKING_ENGINE,
     RAG_RERANKING_MODEL,
+    RAG_EXTERNAL_RERANKER_URL,
+    RAG_EXTERNAL_RERANKER_API_KEY,
     RAG_RERANKING_MODEL_AUTO_UPDATE,
     RAG_RERANKING_MODEL_TRUST_REMOTE_CODE,
     RAG_EMBEDDING_ENGINE,
     RAG_EMBEDDING_BATCH_SIZE,
+    RAG_TOP_K,
+    RAG_TOP_K_RERANKER,
     RAG_RELEVANCE_THRESHOLD,
+    RAG_HYBRID_BM25_WEIGHT,
+    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_SKIP_CACHE,
+    DATALAB_MARKER_FORCE_OCR,
+    DATALAB_MARKER_PAGINATE,
+    DATALAB_MARKER_STRIP_EXISTING_OCR,
+    DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+    DATALAB_MARKER_OUTPUT_FORMAT,
+    DATALAB_MARKER_USE_LLM,
+    EXTERNAL_DOCUMENT_LOADER_URL,
+    EXTERNAL_DOCUMENT_LOADER_API_KEY,
     TIKA_SERVER_URL,
     DOCLING_SERVER_URL,
     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,
-    RAG_TOP_K,
-    RAG_TOP_K_RERANKER,
     RAG_TEXT_SPLITTER,
     TIKTOKEN_ENCODING_NAME,
     PDF_EXTRACT_IMAGES,
@@ -220,6 +257,7 @@ from open_webui.config import (
     ENABLE_WEB_SEARCH,
     WEB_SEARCH_ENGINE,
     BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL,
+    BYPASS_WEB_SEARCH_WEB_LOADER,
     WEB_SEARCH_RESULT_COUNT,
     WEB_SEARCH_CONCURRENT_REQUESTS,
     WEB_SEARCH_TRUST_ENV,
@@ -244,6 +282,8 @@ 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,
@@ -255,6 +295,7 @@ from open_webui.config import (
     GOOGLE_DRIVE_API_KEY,
     ONEDRIVE_CLIENT_ID,
     ONEDRIVE_SHAREPOINT_URL,
+    ONEDRIVE_SHAREPOINT_TENANT_ID,
     ENABLE_RAG_HYBRID_SEARCH,
     ENABLE_RAG_LOCAL_WEB_FETCH,
     ENABLE_WEB_LOADER_SSL_VERIFICATION,
@@ -286,6 +327,8 @@ from open_webui.config import (
     ENABLE_EVALUATION_ARENA_MODELS,
     USER_PERMISSIONS,
     DEFAULT_USER_ROLE,
+    PENDING_USER_OVERLAY_CONTENT,
+    PENDING_USER_OVERLAY_TITLE,
     DEFAULT_PROMPT_SUGGESTIONS,
     DEFAULT_MODELS,
     DEFAULT_ARENA_MODEL,
@@ -312,7 +355,12 @@ from open_webui.config import (
     LDAP_APP_PASSWORD,
     LDAP_USE_TLS,
     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,
@@ -322,6 +370,7 @@ from open_webui.config import (
     DEFAULT_LOCALE,
     OAUTH_PROVIDERS,
     WEBUI_URL,
+    RESPONSE_WATERMARK,
     # Admin
     ENABLE_ADMIN_CHAT_ACCESS,
     ENABLE_ADMIN_EXPORT,
@@ -330,10 +379,12 @@ 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,
@@ -348,6 +399,7 @@ from open_webui.env import (
     AUDIT_LOG_LEVEL,
     CHANGELOG,
     REDIS_URL,
+    REDIS_KEY_PREFIX,
     REDIS_SENTINEL_HOSTS,
     REDIS_SENTINEL_PORT,
     GLOBAL_LOG_LEVEL,
@@ -355,6 +407,7 @@ from open_webui.env import (
     SAFE_MODE,
     SRC_LOG_LEVELS,
     VERSION,
+    INSTANCE_ID,
     WEBUI_BUILD_HASH,
     WEBUI_SECRET_KEY,
     WEBUI_SESSION_COOKIE_SAME_SITE,
@@ -362,12 +415,14 @@ from open_webui.env import (
     WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
     WEBUI_AUTH_TRUSTED_NAME_HEADER,
     WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
+    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,
 )
 
 
@@ -381,6 +436,7 @@ 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
 
@@ -394,9 +450,11 @@ 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
@@ -447,7 +505,9 @@ 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()
 
@@ -459,15 +519,51 @@ async def lifespan(app: FastAPI):
     log.info("Installing external dependencies of functions and tools...")
     install_tool_and_function_dependencies()
 
-    pool_size = THREAD_POOL_SIZE
-    if pool_size and pool_size > 0:
+    app.state.redis = get_redis_connection(
+        redis_url=REDIS_URL,
+        redis_sentinels=get_sentinels_from_env(
+            REDIS_SENTINEL_HOSTS, REDIS_SENTINEL_PORT
+        ),
+        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 = pool_size
+        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",
@@ -479,10 +575,13 @@ 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_key_prefix=REDIS_KEY_PREFIX,
 )
+app.state.redis = None
 
 app.state.WEBUI_NAME = WEBUI_NAME
 app.state.LICENSE_METADATA = None
@@ -543,6 +642,15 @@ app.state.TOOL_SERVERS = []
 
 app.state.config.ENABLE_DIRECT_CONNECTIONS = ENABLE_DIRECT_CONNECTIONS
 
+########################################
+#
+# MODELS
+#
+########################################
+
+app.state.config.ENABLE_BASE_MODELS_CACHE = ENABLE_BASE_MODELS_CACHE
+app.state.BASE_MODELS = []
+
 ########################################
 #
 # WEBUI
@@ -569,6 +677,11 @@ app.state.config.DEFAULT_MODELS = DEFAULT_MODELS
 app.state.config.DEFAULT_PROMPT_SUGGESTIONS = DEFAULT_PROMPT_SUGGESTIONS
 app.state.config.DEFAULT_USER_ROLE = DEFAULT_USER_ROLE
 
+app.state.config.PENDING_USER_OVERLAY_CONTENT = PENDING_USER_OVERLAY_CONTENT
+app.state.config.PENDING_USER_OVERLAY_TITLE = PENDING_USER_OVERLAY_TITLE
+
+app.state.config.RESPONSE_WATERMARK = RESPONSE_WATERMARK
+
 app.state.config.USER_PERMISSIONS = USER_PERMISSIONS
 app.state.config.WEBHOOK_URL = WEBHOOK_URL
 app.state.config.BANNERS = WEBUI_BANNERS
@@ -605,8 +718,14 @@ app.state.config.LDAP_SEARCH_BASE = LDAP_SEARCH_BASE
 app.state.config.LDAP_SEARCH_FILTERS = LDAP_SEARCH_FILTERS
 app.state.config.LDAP_USE_TLS = LDAP_USE_TLS
 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
@@ -614,8 +733,12 @@ app.state.WEBUI_AUTH_SIGNOUT_REDIRECT_URL = WEBUI_AUTH_SIGNOUT_REDIRECT_URL
 app.state.EXTERNAL_PWA_MANIFEST_URL = EXTERNAL_PWA_MANIFEST_URL
 
 app.state.USER_COUNT = None
+
 app.state.TOOLS = {}
+app.state.TOOL_CONTENTS = {}
+
 app.state.FUNCTIONS = {}
+app.state.FUNCTION_CONTENTS = {}
 
 ########################################
 #
@@ -627,8 +750,14 @@ app.state.FUNCTIONS = {}
 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
@@ -637,10 +766,27 @@ app.state.config.ENABLE_RAG_HYBRID_SEARCH = ENABLE_RAG_HYBRID_SEARCH
 app.state.config.ENABLE_WEB_LOADER_SSL_VERIFICATION = ENABLE_WEB_LOADER_SSL_VERIFICATION
 
 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_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
+app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR = DATALAB_MARKER_STRIP_EXISTING_OCR
+app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = (
+    DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
+)
+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
+app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY = EXTERNAL_DOCUMENT_LOADER_API_KEY
 app.state.config.TIKA_SERVER_URL = TIKA_SERVER_URL
 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
@@ -654,12 +800,21 @@ app.state.config.CHUNK_OVERLAP = CHUNK_OVERLAP
 app.state.config.RAG_EMBEDDING_ENGINE = RAG_EMBEDDING_ENGINE
 app.state.config.RAG_EMBEDDING_MODEL = RAG_EMBEDDING_MODEL
 app.state.config.RAG_EMBEDDING_BATCH_SIZE = RAG_EMBEDDING_BATCH_SIZE
+
+app.state.config.RAG_RERANKING_ENGINE = RAG_RERANKING_ENGINE
 app.state.config.RAG_RERANKING_MODEL = RAG_RERANKING_MODEL
+app.state.config.RAG_EXTERNAL_RERANKER_URL = RAG_EXTERNAL_RERANKER_URL
+app.state.config.RAG_EXTERNAL_RERANKER_API_KEY = RAG_EXTERNAL_RERANKER_API_KEY
+
 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
 
@@ -679,6 +834,7 @@ app.state.config.WEB_SEARCH_TRUST_ENV = WEB_SEARCH_TRUST_ENV
 app.state.config.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL = (
     BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL
 )
+app.state.config.BYPASS_WEB_SEARCH_WEB_LOADER = BYPASS_WEB_SEARCH_WEB_LOADER
 
 app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION = ENABLE_GOOGLE_DRIVE_INTEGRATION
 app.state.config.ENABLE_ONEDRIVE_INTEGRATION = ENABLE_ONEDRIVE_INTEGRATION
@@ -706,6 +862,8 @@ 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
@@ -721,6 +879,7 @@ 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
 
@@ -735,7 +894,10 @@ try:
     )
 
     app.state.rf = get_rf(
+        app.state.config.RAG_RERANKING_ENGINE,
         app.state.config.RAG_RERANKING_MODEL,
+        app.state.config.RAG_EXTERNAL_RERANKER_URL,
+        app.state.config.RAG_EXTERNAL_RERANKER_API_KEY,
         RAG_RERANKING_MODEL_AUTO_UPDATE,
     )
 except Exception as e:
@@ -746,18 +908,37 @@ 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,
 )
 
 ########################################
@@ -828,10 +1009,12 @@ 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
@@ -878,6 +1061,7 @@ 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
@@ -885,6 +1069,9 @@ 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
@@ -928,6 +1115,9 @@ class RedirectMiddleware(BaseHTTPMiddleware):
 
 
 # Add the middleware to the app
+if ENABLE_COMPRESSION_MIDDLEWARE:
+    app.add_middleware(CompressMiddleware)
+
 app.add_middleware(RedirectMiddleware)
 app.add_middleware(SecurityHeadersMiddleware)
 
@@ -1043,7 +1233,9 @@ 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:
@@ -1067,7 +1259,7 @@ 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:
@@ -1115,6 +1307,37 @@ 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,
@@ -1155,11 +1378,12 @@ async def chat_completion(
             "chat_id": form_data.pop("chat_id", None),
             "message_id": form_data.pop("id", None),
             "session_id": form_data.pop("session_id", None),
+            "filter_ids": form_data.pop("filter_ids", []),
             "tool_ids": form_data.get("tool_ids", None),
             "tool_servers": form_data.pop("tool_servers", None),
             "files": form_data.get("files", None),
-            "features": form_data.get("features", None),
-            "variables": form_data.get("variables", None),
+            "features": form_data.get("features", {}),
+            "variables": form_data.get("variables", {}),
             "model": model,
             "direct": model_item.get("direct", False),
             **(
@@ -1205,6 +1429,17 @@ async def chat_completion(
             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),
@@ -1255,28 +1490,32 @@ 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}
 
 
@@ -1329,6 +1568,7 @@ async def get_app_config(request: Request):
             "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,
@@ -1372,6 +1612,10 @@ 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": {
@@ -1381,6 +1625,12 @@ async def get_app_config(request: Request):
                 "onedrive": {
                     "client_id": ONEDRIVE_CLIENT_ID.value,
                     "sharepoint_url": ONEDRIVE_SHAREPOINT_URL.value,
+                    "sharepoint_tenant_id": ONEDRIVE_SHAREPOINT_TENANT_ID.value,
+                },
+                "ui": {
+                    "pending_user_overlay_title": app.state.config.PENDING_USER_OVERLAY_TITLE,
+                    "pending_user_overlay_content": app.state.config.PENDING_USER_OVERLAY_CONTENT,
+                    "response_watermark": app.state.config.RESPONSE_WATERMARK,
                 },
                 "license_metadata": app.state.LICENSE_METADATA,
                 **(
@@ -1392,7 +1642,19 @@ async def get_app_config(request: Request):
                 ),
             }
             if user is not None
-            else {}
+            else {
+                **(
+                    {
+                        "metadata": {
+                            "login_footer": app.state.LICENSE_METADATA.get(
+                                "login_footer", ""
+                            )
+                        }
+                    }
+                    if app.state.LICENSE_METADATA
+                    else {}
+                )
+            }
         ),
     }
 
@@ -1424,16 +1686,17 @@ 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:
         timeout = aiohttp.ClientTimeout(total=1)
         async with aiohttp.ClientSession(timeout=timeout, trust_env=True) as session:
             async with session.get(
-                "https://api.github.com/repos/open-webui/open-webui/releases/latest"
+                "https://api.github.com/repos/open-webui/open-webui/releases/latest",
+                ssl=AIOHTTP_CLIENT_SESSION_SSL,
             ) as response:
                 response.raise_for_status()
                 data = await response.json()
@@ -1450,6 +1713,19 @@ 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
 ############################
@@ -1538,7 +1814,20 @@ 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/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/auths.py

@@ -129,12 +129,16 @@ class AuthsTable:
 
     def authenticate_user(self, email: str, password: str) -> Optional[UserModel]:
         log.info(f"authenticate_user: {email}")
+
+        user = Users.get_user_by_email(email)
+        if not user:
+            return None
+
         try:
             with get_db() as db:
-                auth = db.query(Auth).filter_by(email=email, active=True).first()
+                auth = db.query(Auth).filter_by(id=user.id, active=True).first()
                 if auth:
                     if verify_password(password, auth.password):
-                        user = Users.get_user_by_id(auth.id)
                         return user
                     else:
                         return None
@@ -155,8 +159,8 @@ class AuthsTable:
         except Exception:
             return False
 
-    def authenticate_user_by_trusted_header(self, email: str) -> Optional[UserModel]:
-        log.info(f"authenticate_user_by_trusted_header: {email}")
+    def authenticate_user_by_email(self, email: str) -> Optional[UserModel]:
+        log.info(f"authenticate_user_by_email: {email}")
         try:
             with get_db() as db:
                 auth = db.query(Auth).filter_by(email=email, active=True).first()

backend/open_webui/models/chats.py

@@ -12,6 +12,7 @@ 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 +67,14 @@ 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 +121,7 @@ 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 +151,16 @@ 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 +244,10 @@ 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", {})
 
@@ -377,22 +393,47 @@ class ChatTable:
             return False
 
     def get_archived_chat_list_by_user_id(
-        self, user_id: str, skip: int = 0, limit: int = 50
+        self,
+        user_id: str,
+        filter: Optional[dict] = None,
+        skip: int = 0,
+        limit: int = 50,
     ) -> list[ChatModel]:
+
         with get_db() as db:
-            all_chats = (
-                db.query(Chat)
-                .filter_by(user_id=user_id, archived=True)
-                .order_by(Chat.updated_at.desc())
-                # .limit(limit).offset(skip)
-                .all()
-            )
+            query = db.query(Chat).filter_by(user_id=user_id, archived=True)
+
+            if filter:
+                query_key = filter.get("query")
+                if query_key:
+                    query = query.filter(Chat.title.ilike(f"%{query_key}%"))
+
+                order_by = filter.get("order_by")
+                direction = filter.get("direction")
+
+                if order_by and direction and getattr(Chat, order_by):
+                    if direction.lower() == "asc":
+                        query = query.order_by(getattr(Chat, order_by).asc())
+                    elif direction.lower() == "desc":
+                        query = query.order_by(getattr(Chat, order_by).desc())
+                    else:
+                        raise ValueError("Invalid direction for ordering")
+            else:
+                query = query.order_by(Chat.updated_at.desc())
+
+            if skip:
+                query = query.offset(skip)
+            if limit:
+                query = query.limit(limit)
+
+            all_chats = query.all()
             return [ChatModel.model_validate(chat) for chat in all_chats]
 
     def get_chat_list_by_user_id(
         self,
         user_id: str,
         include_archived: bool = False,
+        filter: Optional[dict] = None,
         skip: int = 0,
         limit: int = 50,
     ) -> list[ChatModel]:
@@ -401,7 +442,23 @@ class ChatTable:
             if not include_archived:
                 query = query.filter_by(archived=False)
 
-            query = query.order_by(Chat.updated_at.desc())
+            if filter:
+                query_key = filter.get("query")
+                if query_key:
+                    query = query.filter(Chat.title.ilike(f"%{query_key}%"))
+
+                order_by = filter.get("order_by")
+                direction = filter.get("direction")
+
+                if order_by and direction and getattr(Chat, order_by):
+                    if direction.lower() == "asc":
+                        query = query.order_by(getattr(Chat, order_by).asc())
+                    elif direction.lower() == "desc":
+                        query = query.order_by(getattr(Chat, order_by).desc())
+                    else:
+                        raise ValueError("Invalid direction for ordering")
+            else:
+                query = query.order_by(Chat.updated_at.desc())
 
             if skip:
                 query = query.offset(skip)
@@ -539,10 +596,12 @@ 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(user_id, include_archived, skip, limit)
+            return self.get_chat_list_by_user_id(
+                user_id, include_archived, filter={}, skip=skip, limit=limit
+            )
 
         search_text_words = search_text.split(" ")
 
@@ -571,21 +630,18 @@ class ChatTable:
             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
@@ -620,21 +676,19 @@ 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

@@ -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
@@ -55,6 +57,7 @@ class FolderModel(BaseModel):
 
 class FolderForm(BaseModel):
     name: str
+    data: Optional[dict] = None
     model_config = ConfigDict(extra="allow")
 
 
@@ -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()

backend/open_webui/models/functions.py

@@ -108,6 +108,54 @@ class FunctionsTable:
             log.exception(f"Error creating a new function: {e}")
             return None
 
+    def sync_functions(
+        self, user_id: str, functions: list[FunctionModel]
+    ) -> list[FunctionModel]:
+        # Synchronize functions for a user by updating existing ones, inserting new ones, and removing those that are no longer present.
+        try:
+            with get_db() as db:
+                # Get existing functions
+                existing_functions = db.query(Function).all()
+                existing_ids = {func.id for func in existing_functions}
+
+                # Prepare a set of new function IDs
+                new_function_ids = {func.id for func in functions}
+
+                # Update or insert functions
+                for func in functions:
+                    if func.id in existing_ids:
+                        db.query(Function).filter_by(id=func.id).update(
+                            {
+                                **func.model_dump(),
+                                "user_id": user_id,
+                                "updated_at": int(time.time()),
+                            }
+                        )
+                    else:
+                        new_func = Function(
+                            **{
+                                **func.model_dump(),
+                                "user_id": user_id,
+                                "updated_at": int(time.time()),
+                            }
+                        )
+                        db.add(new_func)
+
+                # Remove functions that are no longer present
+                for func in existing_functions:
+                    if func.id not in new_function_ids:
+                        db.delete(func)
+
+                db.commit()
+
+                return [
+                    FunctionModel.model_validate(func)
+                    for func in db.query(Function).all()
+                ]
+        except Exception as e:
+            log.exception(f"Error syncing functions for user {user_id}: {e}")
+            return []
+
     def get_function_by_id(self, id: str) -> Optional[FunctionModel]:
         try:
             with get_db() as db:

backend/open_webui/models/groups.py

@@ -207,5 +207,73 @@ class GroupTable:
             except Exception:
                 return False
 
+    def create_groups_by_group_names(
+        self, user_id: str, group_names: list[str]
+    ) -> 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()
+                group_ids = [group.id for group in groups]
+
+                # Remove user from groups not in the new list
+                existing_groups = self.get_groups_by_member_id(user_id)
+
+                for group in existing_groups:
+                    if group.id not in group_ids:
+                        group.user_ids.remove(user_id)
+                        db.query(Group).filter_by(id=group.id).update(
+                            {
+                                "user_ids": group.user_ids,
+                                "updated_at": int(time.time()),
+                            }
+                        )
+
+                # Add user to new groups
+                for group in groups:
+                    if user_id not in group.user_ids:
+                        group.user_ids.append(user_id)
+                        db.query(Group).filter_by(id=group.id).update(
+                            {
+                                "user_ids": group.user_ids,
+                                "updated_at": int(time.time()),
+                            }
+                        )
+
+                db.commit()
+                return True
+            except Exception as e:
+                log.exception(e)
+                return False
+
 
 Groups = GroupTable()

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

@@ -95,6 +95,7 @@ class UserRoleUpdateForm(BaseModel):
 
 
 class UserUpdateForm(BaseModel):
+    role: str
     name: str
     email: str
     profile_image_url: str
@@ -369,7 +370,7 @@ 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

@@ -0,0 +1,251 @@
+import os
+import time
+import requests
+import logging
+import json
+from typing import List, Optional
+from langchain_core.documents import Document
+from fastapi import HTTPException, status
+
+log = logging.getLogger(__name__)
+
+
+class DatalabMarkerLoader:
+    def __init__(
+        self,
+        file_path: str,
+        api_key: str,
+        langs: 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,
+        output_format: str = None,
+    ):
+        self.file_path = file_path
+        self.api_key = api_key
+        self.langs = langs
+        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.output_format = output_format
+
+    def _get_mime_type(self, filename: str) -> str:
+        ext = filename.rsplit(".", 1)[-1].lower()
+        mime_map = {
+            "pdf": "application/pdf",
+            "xls": "application/vnd.ms-excel",
+            "xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+            "ods": "application/vnd.oasis.opendocument.spreadsheet",
+            "doc": "application/msword",
+            "docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+            "odt": "application/vnd.oasis.opendocument.text",
+            "ppt": "application/vnd.ms-powerpoint",
+            "pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+            "odp": "application/vnd.oasis.opendocument.presentation",
+            "html": "text/html",
+            "epub": "application/epub+zip",
+            "png": "image/png",
+            "jpeg": "image/jpeg",
+            "jpg": "image/jpeg",
+            "webp": "image/webp",
+            "gif": "image/gif",
+            "tiff": "image/tiff",
+        }
+        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}"
+        headers = {"X-Api-Key": self.api_key}
+        try:
+            response = requests.get(url, headers=headers)
+            response.raise_for_status()
+            result = response.json()
+            log.info(f"Marker API status check for request {request_id}: {result}")
+            return result
+        except requests.HTTPError as e:
+            log.error(f"Error checking Marker request status: {e}")
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY,
+                detail=f"Failed to check Marker request: {e}",
+            )
+        except ValueError as e:
+            log.error(f"Invalid JSON checking Marker request: {e}")
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY, detail=f"Invalid JSON: {e}"
+            )
+
+    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(),
+            "output_format": self.output_format,
+        }
+
+        log.info(
+            f"Datalab Marker POST request parameters: {{'filename': '{filename}', 'mime_type': '{mime_type}', **{form_data}}}"
+        )
+
+        try:
+            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
+                )
+                response.raise_for_status()
+                result = response.json()
+        except FileNotFoundError:
+            raise HTTPException(
+                status.HTTP_404_NOT_FOUND, detail=f"File not found: {self.file_path}"
+            )
+        except requests.HTTPError as e:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Datalab Marker request failed: {e}",
+            )
+        except ValueError as e:
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY, detail=f"Invalid JSON response: {e}"
+            )
+        except Exception as e:
+            raise HTTPException(status.HTTP_500_INTERNAL_SERVER_ERROR, detail=str(e))
+
+        if not result.get("success"):
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Datalab Marker request failed: {result.get('error', 'Unknown error')}",
+            )
+
+        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",
+                    )
+                }
+                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"Final processing failed: {error_msg}",
+            )
+
+        content_key = self.output_format.lower()
+        raw_content = poll_result.get(content_key)
+
+        if content_key == "json":
+            full_text = json.dumps(raw_content, indent=2)
+        elif content_key in {"markdown", "html"}:
+            full_text = str(raw_content).strip()
+        else:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Unsupported output format: {self.output_format}",
+            )
+
+        if not full_text:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail="Datalab 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")
+        output_filename = f"{os.path.splitext(filename)[0]}.{file_ext}"
+        output_path = os.path.join(marker_output_dir, output_filename)
+
+        try:
+            with open(output_path, "w", encoding="utf-8") as f:
+                f.write(full_text)
+            log.info(f"Saved Marker output to: {output_path}")
+        except Exception as e:
+            log.warning(f"Failed to write marker output to disk: {e}")
+
+        metadata = {
+            "source": filename,
+            "output_format": poll_result.get("output_format", self.output_format),
+            "page_count": poll_result.get("page_count", 0),
+            "processed_with_llm": self.use_llm,
+            "request_id": request_id or "",
+        }
+
+        images = poll_result.get("images", {})
+        if images:
+            metadata["image_count"] = len(images)
+            metadata["images"] = json.dumps(list(images.keys()))
+
+        for k, v in metadata.items():
+            if isinstance(v, (dict, list)):
+                metadata[k] = json.dumps(v)
+            elif v is None:
+                metadata[k] = ""
+
+        return [Document(page_content=full_text, metadata=metadata)]

backend/open_webui/retrieval/loaders/external_document.py

@@ -0,0 +1,83 @@
+import requests
+import logging, os
+from typing import Iterator, List, Union
+
+from langchain_core.document_loaders import BaseLoader
+from langchain_core.documents import Document
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class ExternalDocumentLoader(BaseLoader):
+    def __init__(
+        self,
+        file_path,
+        url: str,
+        api_key: str,
+        mime_type=None,
+        **kwargs,
+    ) -> None:
+        self.url = url
+        self.api_key = api_key
+
+        self.file_path = file_path
+        self.mime_type = mime_type
+
+    def load(self) -> List[Document]:
+        with open(self.file_path, "rb") as f:
+            data = f.read()
+
+        headers = {}
+        if self.mime_type is not None:
+            headers["Content-Type"] = self.mime_type
+
+        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]
+
+        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 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")
+
+            else:
+                raise Exception("Error loading document: No content returned")
+        else:
+            raise Exception(
+                f"Error loading document: {response.status_code} {response.text}"
+            )

backend/open_webui/retrieval/loaders/external_web.py

@@ -10,7 +10,7 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
-class ExternalLoader(BaseLoader):
+class ExternalWebLoader(BaseLoader):
     def __init__(
         self,
         web_paths: Union[str, List[str]],
@@ -32,7 +32,7 @@ class ExternalLoader(BaseLoader):
                 response = requests.post(
                     self.external_url,
                     headers={
-                        "User-Agent": "Open WebUI (https://github.com/open-webui/open-webui) RAG Bot",
+                        "User-Agent": "Open WebUI (https://github.com/open-webui/open-webui) External Web Loader",
                         "Authorization": f"Bearer {self.external_api_key}",
                     },
                     json={

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,
@@ -21,7 +22,11 @@ from langchain_community.document_loaders import (
 )
 from langchain_core.documents import Document
 
+from open_webui.retrieval.loaders.external_document import ExternalDocumentLoader
+
 from open_webui.retrieval.loaders.mistral import MistralLoader
+from open_webui.retrieval.loaders.datalab_marker import DatalabMarkerLoader
+
 
 from open_webui.env import SRC_LOG_LEVELS, GLOBAL_LOG_LEVEL
 
@@ -72,7 +77,6 @@ known_source_ext = [
     "swift",
     "vue",
     "svelte",
-    "msg",
     "ex",
     "exs",
     "erl",
@@ -126,14 +130,12 @@ class TikaLoader:
 
 
 class DoclingLoader:
-    def __init__(
-        self, url, file_path=None, mime_type=None, ocr_engine=None, ocr_lang=None
-    ):
+    def __init__(self, url, file_path=None, mime_type=None, params=None):
         self.url = url.rstrip("/")
         self.file_path = file_path
         self.mime_type = mime_type
-        self.ocr_engine = ocr_engine
-        self.ocr_lang = ocr_lang
+
+        self.params = params or {}
 
     def load(self) -> list[Document]:
         with open(self.file_path, "rb") as f:
@@ -145,16 +147,39 @@ class DoclingLoader:
                 )
             }
 
-            params = {
-                "image_export_mode": "placeholder",
-                "table_mode": "accurate",
-            }
+            params = {"image_export_mode": "placeholder", "table_mode": "accurate"}
 
-            if self.ocr_engine and self.ocr_lang:
-                params["ocr_engine"] = self.ocr_engine
-                params["ocr_lang"] = [
-                    lang.strip() for lang in self.ocr_lang.split(",") if lang.strip()
-                ]
+            if self.params:
+                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"] = [
+                        lang.strip()
+                        for lang in self.params.get("ocr_lang").split(",")
+                        if lang.strip()
+                    ]
 
             endpoint = f"{self.url}/v1alpha/convert/file"
             r = requests.post(endpoint, files=files, data=params)
@@ -201,13 +226,27 @@ 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):
         file_ext = filename.split(".")[-1].lower()
 
-        if self.engine == "tika" and self.kwargs.get("TIKA_SERVER_URL"):
+        if (
+            self.engine == "external"
+            and self.kwargs.get("EXTERNAL_DOCUMENT_LOADER_URL")
+            and self.kwargs.get("EXTERNAL_DOCUMENT_LOADER_API_KEY")
+        ):
+            loader = ExternalDocumentLoader(
+                file_path=file_path,
+                url=self.kwargs.get("EXTERNAL_DOCUMENT_LOADER_URL"),
+                api_key=self.kwargs.get("EXTERNAL_DOCUMENT_LOADER_API_KEY"),
+                mime_type=file_content_type,
+            )
+        elif self.engine == "tika" and self.kwargs.get("TIKA_SERVER_URL"):
             if self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:
@@ -217,16 +256,67 @@ class Loader:
                     mime_type=file_content_type,
                     extract_images=self.kwargs.get("PDF_EXTRACT_IMAGES"),
                 )
+        elif (
+            self.engine == "datalab_marker"
+            and self.kwargs.get("DATALAB_MARKER_API_KEY")
+            and file_ext
+            in [
+                "pdf",
+                "xls",
+                "xlsx",
+                "ods",
+                "doc",
+                "docx",
+                "odt",
+                "ppt",
+                "pptx",
+                "odp",
+                "html",
+                "epub",
+                "png",
+                "jpeg",
+                "jpg",
+                "webp",
+                "gif",
+                "tiff",
+            ]
+        ):
+            loader = DatalabMarkerLoader(
+                file_path=file_path,
+                api_key=self.kwargs["DATALAB_MARKER_API_KEY"],
+                langs=self.kwargs.get("DATALAB_MARKER_LANGS"),
+                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),
+                paginate=self.kwargs.get("DATALAB_MARKER_PAGINATE", False),
+                strip_existing_ocr=self.kwargs.get(
+                    "DATALAB_MARKER_STRIP_EXISTING_OCR", False
+                ),
+                disable_image_extraction=self.kwargs.get(
+                    "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION", False
+                ),
+                output_format=self.kwargs.get(
+                    "DATALAB_MARKER_OUTPUT_FORMAT", "markdown"
+                ),
+            )
         elif self.engine == "docling" and self.kwargs.get("DOCLING_SERVER_URL"):
             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,
-                    ocr_engine=self.kwargs.get("DOCLING_OCR_ENGINE"),
-                    ocr_lang=self.kwargs.get("DOCLING_OCR_LANG"),
+                    params=params,
                 )
         elif (
             self.engine == "document_intelligence"
@@ -258,6 +348,15 @@ class Loader:
             loader = MistralLoader(
                 api_key=self.kwargs.get("MISTRAL_OCR_API_KEY"), file_path=file_path
             )
+        elif (
+            self.engine == "external"
+            and self.kwargs.get("MISTRAL_OCR_API_KEY") != ""
+            and file_ext
+            in ["pdf"]  # Mistral OCR currently only supports PDF and images
+        ):
+            loader = MistralLoader(
+                api_key=self.kwargs.get("MISTRAL_OCR_API_KEY"), file_path=file_path
+            )
         else:
             if file_ext == "pdf":
                 loader = PyPDFLoader(
@@ -293,6 +392,8 @@ 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

@@ -1,8 +1,12 @@
 import requests
+import aiohttp
+import asyncio
 import logging
 import os
 import sys
+import time
 from typing import List, Dict, Any
+from contextlib import asynccontextmanager
 
 from langchain_core.documents import Document
 from open_webui.env import SRC_LOG_LEVELS, GLOBAL_LOG_LEVEL
@@ -14,18 +18,37 @@ log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 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"
 
-    def __init__(self, api_key: str, file_path: str):
+    def __init__(
+        self,
+        api_key: str,
+        file_path: str,
+        timeout: int = 300,  # 5 minutes default
+        max_retries: int = 3,
+        enable_debug_logging: bool = False,
+    ):
         """
-        Initializes the loader.
+        Initializes the loader with enhanced features.
 
         Args:
             api_key: Your Mistral API key.
             file_path: The local path to the PDF file to process.
+            timeout: Request timeout in seconds.
+            max_retries: Maximum number of retry attempts.
+            enable_debug_logging: Enable detailed debug logs.
         """
         if not api_key:
             raise ValueError("API key cannot be empty.")
@@ -34,7 +57,46 @@ class MistralLoader:
 
         self.api_key = api_key
         self.file_path = file_path
-        self.headers = {"Authorization": f"Bearer {self.api_key}"}
+        self.timeout = timeout
+        self.max_retries = max_retries
+        self.debug = enable_debug_logging
+
+        # 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",  # Helps API provider track usage
+        }
+
+    def _debug_log(self, message: str, *args) -> None:
+        """
+        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)
 
     def _handle_response(self, response: requests.Response) -> Dict[str, Any]:
         """Checks response status and returns JSON content."""
@@ -54,24 +116,154 @@ class MistralLoader:
             log.error(f"JSON decode error: {json_err} - Response: {response.text}")
             raise  # Re-raise after logging
 
-    def _upload_file(self) -> str:
-        """Uploads the file to Mistral for OCR processing."""
-        log.info("Uploading file to Mistral API")
-        url = f"{self.BASE_API_URL}/files"
-        file_name = os.path.basename(self.file_path)
-
+    async def _handle_response_async(
+        self, response: aiohttp.ClientResponse
+    ) -> Dict[str, Any]:
+        """Async version of response handling with better error info."""
         try:
-            with open(self.file_path, "rb") as f:
-                files = {"file": (file_name, f, "application/pdf")}
-                data = {"purpose": "ocr"}
+            response.raise_for_status()
 
-                upload_headers = self.headers.copy()  # Avoid modifying self.headers
-
-                response = requests.post(
-                    url, headers=upload_headers, files=files, data=data
+            # Check content type
+            content_type = response.headers.get("content-type", "")
+            if "application/json" not in content_type:
+                if response.status == 204:
+                    return {}
+                text = await response.text()
+                raise ValueError(
+                    f"Unexpected content type: {content_type}, body: {text[:200]}..."
                 )
 
-            response_data = self._handle_response(response)
+            return await response.json()
+
+        except aiohttp.ClientResponseError as e:
+            error_text = await response.text() if response else "No response"
+            log.error(f"HTTP {e.status}: {e.message} - Response: {error_text[:500]}")
+            raise
+        except aiohttp.ClientError as e:
+            log.error(f"Client error: {e}")
+            raise
+        except Exception as e:
+            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):
+        """
+        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 Exception as e:
+                if attempt == self.max_retries - 1 or not self._is_retryable_error(e):
+                    raise
+
+                # 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"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):
+        """
+        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 Exception as e:
+                if attempt == self.max_retries - 1 or not self._is_retryable_error(e):
+                    raise
+
+                # PERFORMANCE OPTIMIZATION: Non-blocking exponential backoff
+                wait_time = min((2**attempt) + 0.5, 30)  # Cap at 30 seconds
+                log.warning(
+                    f"Retryable error (attempt {attempt + 1}/{self.max_retries}): {e}. "
+                    f"Retrying in {wait_time}s..."
+                )
+                await asyncio.sleep(wait_time)  # Non-blocking wait
+
+    def _upload_file(self) -> str:
+        """
+        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"
+
+        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": (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.upload_timeout,  # Use specialized upload timeout
+                    stream=False,  # Keep as False for this endpoint
+                )
+
+            return self._handle_response(response)
+
+        try:
+            response_data = self._retry_request_sync(upload_request)
             file_id = response_data.get("id")
             if not file_id:
                 raise ValueError("File ID not found in upload response.")
@@ -81,16 +273,66 @@ class MistralLoader:
             log.error(f"Failed to upload file: {e}")
             raise
 
+    async def _upload_file_async(self, session: aiohttp.ClientSession) -> str:
+        """Async file upload with streaming for better memory efficiency."""
+        url = f"{self.BASE_API_URL}/files"
+
+        async def upload_request():
+            # Create multipart writer for streaming upload
+            writer = aiohttp.MultipartWriter("form-data")
+
+            # Add purpose field
+            purpose_part = writer.append("ocr")
+            purpose_part.set_content_disposition("form-data", name="purpose")
+
+            # Add file part with streaming
+            file_part = writer.append_payload(
+                aiohttp.streams.FilePayload(
+                    self.file_path,
+                    filename=self.file_name,
+                    content_type="application/pdf",
+                )
+            )
+            file_part.set_content_disposition(
+                "form-data", name="file", filename=self.file_name
+            )
+
+            self._debug_log(
+                f"Uploading file: {self.file_name} ({self.file_size:,} bytes)"
+            )
+
+            async with session.post(
+                url,
+                data=writer,
+                headers=self.headers,
+                timeout=aiohttp.ClientTimeout(total=self.upload_timeout),
+            ) as response:
+                return await self._handle_response_async(response)
+
+        response_data = await self._retry_request_async(upload_request)
+
+        file_id = response_data.get("id")
+        if not file_id:
+            raise ValueError("File ID not found in upload response.")
+
+        log.info(f"File uploaded successfully. File ID: {file_id}")
+        return file_id
+
     def _get_signed_url(self, file_id: str) -> str:
-        """Retrieves a temporary signed URL for the uploaded file."""
+        """Retrieves a temporary signed URL for the uploaded file (sync version)."""
         log.info(f"Getting signed URL for file ID: {file_id}")
         url = f"{self.BASE_API_URL}/files/{file_id}/url"
         params = {"expiry": 1}
         signed_url_headers = {**self.headers, "Accept": "application/json"}
 
+        def url_request():
+            response = requests.get(
+                url, headers=signed_url_headers, params=params, timeout=self.url_timeout
+            )
+            return self._handle_response(response)
+
         try:
-            response = requests.get(url, headers=signed_url_headers, params=params)
-            response_data = self._handle_response(response)
+            response_data = self._retry_request_sync(url_request)
             signed_url = response_data.get("url")
             if not signed_url:
                 raise ValueError("Signed URL not found in response.")
@@ -100,8 +342,36 @@ class MistralLoader:
             log.error(f"Failed to get signed URL: {e}")
             raise
 
+    async def _get_signed_url_async(
+        self, session: aiohttp.ClientSession, file_id: str
+    ) -> str:
+        """Async signed URL retrieval."""
+        url = f"{self.BASE_API_URL}/files/{file_id}/url"
+        params = {"expiry": 1}
+
+        headers = {**self.headers, "Accept": "application/json"}
+
+        async def url_request():
+            self._debug_log(f"Getting signed URL for file ID: {file_id}")
+            async with session.get(
+                url,
+                headers=headers,
+                params=params,
+                timeout=aiohttp.ClientTimeout(total=self.url_timeout),
+            ) as response:
+                return await self._handle_response_async(response)
+
+        response_data = await self._retry_request_async(url_request)
+
+        signed_url = response_data.get("url")
+        if not signed_url:
+            raise ValueError("Signed URL not found in response.")
+
+        self._debug_log("Signed URL received successfully")
+        return signed_url
+
     def _process_ocr(self, signed_url: str) -> Dict[str, Any]:
-        """Sends the signed URL to the OCR endpoint for processing."""
+        """Sends the signed URL to the OCR endpoint for processing (sync version)."""
         log.info("Processing OCR via Mistral API")
         url = f"{self.BASE_API_URL}/ocr"
         ocr_headers = {
@@ -118,43 +388,218 @@ class MistralLoader:
             "include_image_base64": False,
         }
 
+        def ocr_request():
+            response = requests.post(
+                url, headers=ocr_headers, json=payload, timeout=self.ocr_timeout
+            )
+            return self._handle_response(response)
+
         try:
-            response = requests.post(url, headers=ocr_headers, json=payload)
-            ocr_response = self._handle_response(response)
+            ocr_response = self._retry_request_sync(ocr_request)
             log.info("OCR processing done.")
-            log.debug("OCR response: %s", ocr_response)
+            self._debug_log("OCR response: %s", ocr_response)
             return ocr_response
         except Exception as e:
             log.error(f"Failed during OCR processing: {e}")
             raise
 
+    async def _process_ocr_async(
+        self, session: aiohttp.ClientSession, signed_url: str
+    ) -> Dict[str, Any]:
+        """Async OCR processing with timing metrics."""
+        url = f"{self.BASE_API_URL}/ocr"
+
+        headers = {
+            **self.headers,
+            "Content-Type": "application/json",
+            "Accept": "application/json",
+        }
+
+        payload = {
+            "model": "mistral-ocr-latest",
+            "document": {
+                "type": "document_url",
+                "document_url": signed_url,
+            },
+            "include_image_base64": False,
+        }
+
+        async def ocr_request():
+            log.info("Starting OCR processing via Mistral API")
+            start_time = time.time()
+
+            async with session.post(
+                url,
+                json=payload,
+                headers=headers,
+                timeout=aiohttp.ClientTimeout(total=self.ocr_timeout),
+            ) as response:
+                ocr_response = await self._handle_response_async(response)
+
+            processing_time = time.time() - start_time
+            log.info(f"OCR processing completed in {processing_time:.2f}s")
+
+            return ocr_response
+
+        return await self._retry_request_async(ocr_request)
+
     def _delete_file(self, file_id: str) -> None:
-        """Deletes the file from Mistral storage."""
+        """Deletes the file from Mistral storage (sync version)."""
         log.info(f"Deleting uploaded file ID: {file_id}")
         url = f"{self.BASE_API_URL}/files/{file_id}"
-        # No specific Accept header needed, default or Authorization is usually sufficient
 
         try:
-            response = requests.delete(url, headers=self.headers)
-            delete_response = self._handle_response(
-                response
-            )  # Check status, ignore response body unless needed
-            log.info(
-                f"File deleted successfully: {delete_response}"
-            )  # Log the response if available
+            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:
             # Log error but don't necessarily halt execution if deletion fails
             log.error(f"Failed to delete file ID {file_id}: {e}")
-            # Depending on requirements, you might choose to raise the error here
+
+    async def _delete_file_async(
+        self, session: aiohttp.ClientSession, file_id: str
+    ) -> None:
+        """Async file deletion with error tolerance."""
+        try:
+
+            async def delete_request():
+                self._debug_log(f"Deleting file ID: {file_id}")
+                async with session.delete(
+                    url=f"{self.BASE_API_URL}/files/{file_id}",
+                    headers=self.headers,
+                    timeout=aiohttp.ClientTimeout(
+                        total=self.cleanup_timeout
+                    ),  # Shorter timeout for cleanup
+                ) as response:
+                    return await self._handle_response_async(response)
+
+            await self._retry_request_async(delete_request)
+            self._debug_log(f"File {file_id} deleted successfully")
+
+        except Exception as e:
+            # Don't fail the entire process if cleanup fails
+            log.warning(f"Failed to delete file ID {file_id}: {e}")
+
+    @asynccontextmanager
+    async def _get_session(self):
+        """Context manager for HTTP session with optimized settings."""
+        connector = aiohttp.TCPConnector(
+            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=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=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 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", "file_name": self.file_name},
+                )
+            ]
+
+        documents = []
+        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 None or page_index is None:
+                skipped_pages += 1
+                self._debug_log(
+                    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(
+                f"Processed {len(documents)} pages, skipped {skipped_pages} empty/invalid pages"
+            )
+
+        if not documents:
+            # Case where pages existed but none had valid markdown/index
+            log.warning(
+                "OCR response contained pages, but none had valid content/index."
+            )
+            return [
+                Document(
+                    page_content="No valid text content found in document",
+                    metadata={
+                        "error": "no_valid_pages",
+                        "total_pages": total_pages,
+                        "file_name": self.file_name,
+                    },
+                )
+            ]
+
+        return documents
 
     def load(self) -> List[Document]:
         """
         Executes the full OCR workflow: upload, get URL, process OCR, delete file.
+        Synchronous version for backward compatibility.
 
         Returns:
             A list of Document objects, one for each page processed.
         """
         file_id = None
+        start_time = time.time()
+
         try:
             # 1. Upload file
             file_id = self._upload_file()
@@ -166,53 +611,30 @@ class MistralLoader:
             ocr_response = self._process_ocr(signed_url)
 
             # 4. Process results
-            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={})]
+            documents = self._process_results(ocr_response)
 
-            documents = []
-            total_pages = len(pages_data)
-            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:
-                    documents.append(
-                        Document(
-                            page_content=page_content,
-                            metadata={
-                                "page": page_index,  # 0-based index from API
-                                "page_label": page_index
-                                + 1,  # 1-based label for convenience
-                                "total_pages": total_pages,
-                                # Add other relevant metadata from page_data if available/needed
-                                # e.g., page_data.get('width'), page_data.get('height')
-                            },
-                        )
-                    )
-                else:
-                    log.warning(
-                        f"Skipping page due to missing 'markdown' or 'index'. Data: {page_data}"
-                    )
-
-            if not documents:
-                # Case where pages existed but none had valid markdown/index
-                log.warning(
-                    "OCR response contained pages, but none had valid content/index."
-                )
-                return [
-                    Document(
-                        page_content="No text content found in valid pages", metadata={}
-                    )
-                ]
+            total_time = time.time() - start_time
+            log.info(
+                f"Sync OCR workflow completed in {total_time:.2f}s, produced {len(documents)} documents"
+            )
 
             return documents
 
         except Exception as e:
-            log.error(f"An error occurred during the loading process: {e}")
-            # Return an empty list or a specific error document on failure
-            return [Document(page_content=f"Error during processing: {e}", metadata={})]
+            total_time = time.time() - start_time
+            log.error(
+                f"An error occurred during the loading process after {total_time:.2f}s: {e}"
+            )
+            # Return an error document on failure
+            return [
+                Document(
+                    page_content=f"Error during processing: {e}",
+                    metadata={
+                        "error": "processing_failed",
+                        "file_name": self.file_name,
+                    },
+                )
+            ]
         finally:
             # 5. Delete file (attempt even if prior steps failed after upload)
             if file_id:
@@ -223,3 +645,124 @@ class MistralLoader:
                     log.error(
                         f"Cleanup error: Could not delete file ID {file_id}. Reason: {del_e}"
                     )
+
+    async def load_async(self) -> List[Document]:
+        """
+        Asynchronous OCR workflow execution with optimized performance.
+
+        Returns:
+            A list of Document objects, one for each page processed.
+        """
+        file_id = None
+        start_time = time.time()
+
+        try:
+            async with self._get_session() as session:
+                # 1. Upload file with streaming
+                file_id = await self._upload_file_async(session)
+
+                # 2. Get signed URL
+                signed_url = await self._get_signed_url_async(session, file_id)
+
+                # 3. Process OCR
+                ocr_response = await self._process_ocr_async(session, signed_url)
+
+                # 4. Process results
+                documents = self._process_results(ocr_response)
+
+                total_time = time.time() - start_time
+                log.info(
+                    f"Async OCR workflow completed in {total_time:.2f}s, produced {len(documents)} documents"
+                )
+
+                return documents
+
+        except Exception as e:
+            total_time = time.time() - start_time
+            log.error(f"Async OCR workflow failed after {total_time:.2f}s: {e}")
+            return [
+                Document(
+                    page_content=f"Error during OCR processing: {e}",
+                    metadata={
+                        "error": "processing_failed",
+                        "file_name": self.file_name,
+                    },
+                )
+            ]
+        finally:
+            # 5. Cleanup - always attempt file deletion
+            if file_id:
+                try:
+                    async with self._get_session() as session:
+                        await self._delete_file_async(session, file_id)
+                except Exception as cleanup_error:
+                    log.error(f"Cleanup failed for file ID {file_id}: {cleanup_error}")
+
+    @staticmethod
+    async def load_multiple_async(
+        loaders: List["MistralLoader"],
+        max_concurrent: int = 5,  # Limit concurrent requests
+    ) -> List[List[Document]]:
+        """
+        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
+        """
+        if not loaders:
+            return []
+
+        log.info(
+            f"Starting concurrent processing of {len(loaders)} files with max {max_concurrent} concurrent"
+        )
+        start_time = time.time()
+
+        # 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
+        processed_results = []
+        for i, result in enumerate(results):
+            if isinstance(result, Exception):
+                log.error(f"File {i} failed: {result}")
+                processed_results.append(
+                    [
+                        Document(
+                            page_content=f"Error processing file: {result}",
+                            metadata={
+                                "error": "batch_processing_failed",
+                                "file_index": i,
+                            },
+                        )
+                    ]
+                )
+            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: "
+            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/base_reranker.py

@@ -0,0 +1,8 @@
+from abc import ABC, abstractmethod
+from typing import Optional, List, Tuple
+
+
+class BaseReranker(ABC):
+    @abstractmethod
+    def predict(self, sentences: List[Tuple[str, str]]) -> Optional[List[float]]:
+        pass

backend/open_webui/retrieval/models/colbert.py

@@ -7,11 +7,13 @@ from colbert.modeling.checkpoint import Checkpoint
 
 from open_webui.env import SRC_LOG_LEVELS
 
+from open_webui.retrieval.models.base_reranker import BaseReranker
+
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
-class ColBERT:
+class ColBERT(BaseReranker):
     def __init__(self, name, **kwargs) -> None:
         log.info("ColBERT: Loading model", name)
         self.device = "cuda" if torch.cuda.is_available() else "cpu"

backend/open_webui/retrieval/models/external.py

@@ -0,0 +1,74 @@
+import logging
+import requests
+from typing import Optional, List, Tuple
+from urllib.parse import quote
+
+
+from open_webui.env import ENABLE_FORWARD_USER_INFO_HEADERS, SRC_LOG_LEVELS
+from open_webui.retrieval.models.base_reranker import BaseReranker
+
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class ExternalReranker(BaseReranker):
+    def __init__(
+        self,
+        api_key: str,
+        url: str = "http://localhost:8080/v1/rerank",
+        model: str = "reranker",
+    ):
+        self.api_key = api_key
+        self.url = url
+        self.model = model
+
+    def predict(
+        self, sentences: List[Tuple[str, str]], user=None
+    ) -> Optional[List[float]]:
+        query = sentences[0][0]
+        docs = [i[1] for i in sentences]
+
+        payload = {
+            "model": self.model,
+            "query": query,
+            "documents": docs,
+            "top_n": len(docs),
+        }
+
+        try:
+            log.info(f"ExternalReranker:predict:model {self.model}")
+            log.info(f"ExternalReranker:predict:query {query}")
+
+            r = requests.post(
+                f"{self.url}",
+                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,
+            )
+
+            r.raise_for_status()
+            data = r.json()
+
+            if "results" in data:
+                sorted_results = sorted(data["results"], key=lambda x: x["index"])
+                return [result["relevance_score"] for result in sorted_results]
+            else:
+                log.error("No results found in external reranking response")
+                return None
+
+        except Exception as e:
+            log.exception(f"Error in external reranking: {e}")
+            return None

backend/open_webui/retrieval/utils.py

@@ -5,19 +5,24 @@ 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
 from langchain_core.documents import Document
 
 from open_webui.config import VECTOR_DB
-from open_webui.retrieval.vector.connector import VECTOR_DB_CLIENT
+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 (
@@ -116,6 +121,7 @@ def query_doc_with_hybrid_search(
     reranking_function,
     k_reranker: int,
     r: float,
+    hybrid_bm25_weight: float,
 ) -> dict:
     try:
         log.debug(f"query_doc_with_hybrid_search:doc {collection_name}")
@@ -131,9 +137,20 @@ def query_doc_with_hybrid_search(
             top_k=k,
         )
 
-        ensemble_retriever = EnsembleRetriever(
-            retrievers=[bm25_retriever, vector_search_retriever], weights=[0.5, 0.5]
-        )
+        if hybrid_bm25_weight <= 0:
+            ensemble_retriever = EnsembleRetriever(
+                retrievers=[vector_search_retriever], weights=[1.0]
+            )
+        elif hybrid_bm25_weight >= 1:
+            ensemble_retriever = EnsembleRetriever(
+                retrievers=[bm25_retriever], weights=[1.0]
+            )
+        else:
+            ensemble_retriever = EnsembleRetriever(
+                retrievers=[bm25_retriever, vector_search_retriever],
+                weights=[hybrid_bm25_weight, 1.0 - hybrid_bm25_weight],
+            )
+
         compressor = RerankCompressor(
             embedding_function=embedding_function,
             top_n=k_reranker,
@@ -313,6 +330,7 @@ def query_collection_with_hybrid_search(
     reranking_function,
     k_reranker: int,
     r: float,
+    hybrid_bm25_weight: float,
 ) -> dict:
     results = []
     error = False
@@ -346,6 +364,7 @@ def query_collection_with_hybrid_search(
                 reranking_function=reranking_function,
                 k_reranker=k_reranker,
                 r=r,
+                hybrid_bm25_weight=hybrid_bm25_weight,
             )
             return result, None
         except Exception as e:
@@ -386,12 +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,
@@ -400,6 +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):
@@ -424,169 +445,233 @@ 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,
     reranking_function,
     k_reranker,
     r,
+    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 user.role == "admin" 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"])
+
+        # 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,
-                                )
-                            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:
@@ -659,7 +744,7 @@ 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,
@@ -681,6 +766,60 @@ 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],
@@ -704,7 +843,7 @@ 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,
@@ -745,38 +884,33 @@ 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
 
 
@@ -806,7 +940,7 @@ 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:
@@ -818,7 +952,9 @@ class RerankCompressor(BaseDocumentCompressor):
             )
             scores = util.cos_sim(query_embedding, document_embedding)[0]
 
-        docs_with_scores = list(zip(documents, scores.tolist()))
+        docs_with_scores = list(
+            zip(documents, scores.tolist() if not isinstance(scores, list) else scores)
+        )
         if self.r_score:
             docs_with_scores = [
                 (d, s) for d, s in docs_with_scores if s >= self.r_score

backend/open_webui/retrieval/vector/connector.py

@@ -1,30 +0,0 @@
-from open_webui.config import VECTOR_DB
-
-if VECTOR_DB == "milvus":
-    from open_webui.retrieval.vector.dbs.milvus import MilvusClient
-
-    VECTOR_DB_CLIENT = MilvusClient()
-elif VECTOR_DB == "qdrant":
-    from open_webui.retrieval.vector.dbs.qdrant import QdrantClient
-
-    VECTOR_DB_CLIENT = QdrantClient()
-elif VECTOR_DB == "opensearch":
-    from open_webui.retrieval.vector.dbs.opensearch import OpenSearchClient
-
-    VECTOR_DB_CLIENT = OpenSearchClient()
-elif VECTOR_DB == "pgvector":
-    from open_webui.retrieval.vector.dbs.pgvector import PgvectorClient
-
-    VECTOR_DB_CLIENT = PgvectorClient()
-elif VECTOR_DB == "elasticsearch":
-    from open_webui.retrieval.vector.dbs.elasticsearch import ElasticsearchClient
-
-    VECTOR_DB_CLIENT = ElasticsearchClient()
-elif VECTOR_DB == "pinecone":
-    from open_webui.retrieval.vector.dbs.pinecone import PineconeClient
-
-    VECTOR_DB_CLIENT = PineconeClient()
-else:
-    from open_webui.retrieval.vector.dbs.chroma import ChromaClient
-
-    VECTOR_DB_CLIENT = ChromaClient()

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

@@ -3,7 +3,6 @@ from pymilvus import FieldSchema, DataType
 import json
 import logging
 from typing import Optional
-
 from open_webui.retrieval.vector.main import (
     VectorDBBase,
     VectorItem,
@@ -14,6 +13,11 @@ from open_webui.config import (
     MILVUS_URI,
     MILVUS_DB,
     MILVUS_TOKEN,
+    MILVUS_INDEX_TYPE,
+    MILVUS_METRIC_TYPE,
+    MILVUS_HNSW_M,
+    MILVUS_HNSW_EFCONSTRUCTION,
+    MILVUS_IVF_FLAT_NLIST,
 )
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -33,7 +37,6 @@ class MilvusClient(VectorDBBase):
         ids = []
         documents = []
         metadatas = []
-
         for match in result:
             _ids = []
             _documents = []
@@ -42,11 +45,9 @@ class MilvusClient(VectorDBBase):
                 _ids.append(item.get("id"))
                 _documents.append(item.get("data", {}).get("text"))
                 _metadatas.append(item.get("metadata"))
-
             ids.append(_ids)
             documents.append(_documents)
             metadatas.append(_metadatas)
-
         return GetResult(
             **{
                 "ids": ids,
@@ -60,13 +61,11 @@ class MilvusClient(VectorDBBase):
         distances = []
         documents = []
         metadatas = []
-
         for match in result:
             _ids = []
             _distances = []
             _documents = []
             _metadatas = []
-
             for item in match:
                 _ids.append(item.get("id"))
                 # normalize milvus score from [-1, 1] to [0, 1] range
@@ -75,12 +74,10 @@ class MilvusClient(VectorDBBase):
                 _distances.append(_dist)
                 _documents.append(item.get("entity", {}).get("data", {}).get("text"))
                 _metadatas.append(item.get("entity", {}).get("metadata"))
-
             ids.append(_ids)
             distances.append(_distances)
             documents.append(_documents)
             metadatas.append(_metadatas)
-
         return SearchResult(
             **{
                 "ids": ids,
@@ -113,11 +110,39 @@ class MilvusClient(VectorDBBase):
         )
 
         index_params = self.client.prepare_index_params()
+
+        # Use configurations from config.py
+        index_type = MILVUS_INDEX_TYPE.upper()
+        metric_type = MILVUS_METRIC_TYPE.upper()
+
+        log.info(f"Using Milvus index type: {index_type}, metric type: {metric_type}")
+
+        index_creation_params = {}
+        if index_type == "HNSW":
+            index_creation_params = {
+                "M": MILVUS_HNSW_M,
+                "efConstruction": MILVUS_HNSW_EFCONSTRUCTION,
+            }
+            log.info(f"HNSW params: {index_creation_params}")
+        elif index_type == "IVF_FLAT":
+            index_creation_params = {"nlist": MILVUS_IVF_FLAT_NLIST}
+            log.info(f"IVF_FLAT params: {index_creation_params}")
+        elif index_type in ["FLAT", "AUTOINDEX"]:
+            log.info(f"Using {index_type} index with no specific build-time params.")
+        else:
+            log.warning(
+                f"Unsupported MILVUS_INDEX_TYPE: '{index_type}'. "
+                f"Supported types: HNSW, IVF_FLAT, FLAT, AUTOINDEX. "
+                f"Milvus will use its default for the collection if this type is not directly supported for index creation."
+            )
+            # For unsupported types, pass the type directly to Milvus; it might handle it or use a default.
+            # If Milvus errors out, the user needs to correct the MILVUS_INDEX_TYPE env var.
+
         index_params.add_index(
             field_name="vector",
-            index_type="HNSW",
-            metric_type="COSINE",
-            params={"M": 16, "efConstruction": 100},
+            index_type=index_type,
+            metric_type=metric_type,
+            params=index_creation_params,
         )
 
         self.client.create_collection(
@@ -125,6 +150,9 @@ class MilvusClient(VectorDBBase):
             schema=schema,
             index_params=index_params,
         )
+        log.info(
+            f"Successfully created collection '{self.collection_prefix}_{collection_name}' with index type '{index_type}' and metric '{metric_type}'."
+        )
 
     def has_collection(self, collection_name: str) -> bool:
         # Check if the collection exists based on the collection name.
@@ -145,84 +173,113 @@ class MilvusClient(VectorDBBase):
     ) -> Optional[SearchResult]:
         # Search for the nearest neighbor items based on the vectors and return 'limit' number of results.
         collection_name = collection_name.replace("-", "_")
+        # For some index types like IVF_FLAT, search params like nprobe can be set.
+        # Example: search_params = {"nprobe": 10} if using IVF_FLAT
+        # For simplicity, not adding configurable search_params here, but could be extended.
         result = self.client.search(
             collection_name=f"{self.collection_prefix}_{collection_name}",
             data=vectors,
             limit=limit,
             output_fields=["data", "metadata"],
+            # search_params=search_params # Potentially add later if needed
         )
-
         return self._result_to_search_result(result)
 
     def query(self, collection_name: str, filter: dict, limit: Optional[int] = None):
         # Construct the filter string for querying
         collection_name = collection_name.replace("-", "_")
         if not self.has_collection(collection_name):
+            log.warning(
+                f"Query attempted on non-existent collection: {self.collection_prefix}_{collection_name}"
+            )
             return None
-
         filter_string = " && ".join(
             [
                 f'metadata["{key}"] == {json.dumps(value)}'
                 for key, value in filter.items()
             ]
         )
-
         max_limit = 16383  # The maximum number of records per request
         all_results = []
-
         if limit is None:
-            limit = float("inf")  # Use infinity as a placeholder for no limit
+            # Milvus default limit for query if not specified is 16384, but docs mention iteration.
+            # Let's set a practical high number if "all" is intended, or handle true pagination.
+            # For now, if limit is None, we'll fetch in batches up to a very large number.
+            # This part could be refined based on expected use cases for "get all".
+            # For this function signature, None implies "as many as possible" up to Milvus limits.
+            limit = (
+                16384 * 10
+            )  # A large number to signify fetching many, will be capped by actual data or max_limit per call.
+            log.info(
+                f"Limit not specified for query, fetching up to {limit} results in batches."
+            )
 
         # Initialize offset and remaining to handle pagination
         offset = 0
         remaining = limit
 
         try:
+            log.info(
+                f"Querying collection {self.collection_prefix}_{collection_name} with filter: '{filter_string}', limit: {limit}"
+            )
             # Loop until there are no more items to fetch or the desired limit is reached
             while remaining > 0:
-                log.info(f"remaining: {remaining}")
                 current_fetch = min(
-                    max_limit, remaining
-                )  # Determine how many items to fetch in this iteration
+                    max_limit, remaining if isinstance(remaining, int) else max_limit
+                )
+                log.debug(
+                    f"Querying with offset: {offset}, current_fetch: {current_fetch}"
+                )
 
                 results = self.client.query(
                     collection_name=f"{self.collection_prefix}_{collection_name}",
                     filter=filter_string,
-                    output_fields=["*"],
+                    output_fields=[
+                        "id",
+                        "data",
+                        "metadata",
+                    ],  # Explicitly list needed fields. Vector not usually needed in query.
                     limit=current_fetch,
                     offset=offset,
                 )
 
                 if not results:
+                    log.debug("No more results from query.")
                     break
 
                 all_results.extend(results)
                 results_count = len(results)
-                remaining -= (
-                    results_count  # Decrease remaining by the number of items fetched
-                )
+                log.debug(f"Fetched {results_count} results in this batch.")
+
+                if isinstance(remaining, int):
+                    remaining -= results_count
+
                 offset += results_count
 
-                # Break the loop if the results returned are less than the requested fetch count
+                # Break the loop if the results returned are less than the requested fetch count (means end of data)
                 if results_count < current_fetch:
+                    log.debug(
+                        "Fetched less than requested, assuming end of results for this query."
+                    )
                     break
 
-            log.debug(all_results)
+            log.info(f"Total results from query: {len(all_results)}")
             return self._result_to_get_result([all_results])
         except Exception as e:
             log.exception(
-                f"Error querying collection {collection_name} with limit {limit}: {e}"
+                f"Error querying collection {self.collection_prefix}_{collection_name} with filter '{filter_string}' and limit {limit}: {e}"
             )
             return None
 
     def get(self, collection_name: str) -> Optional[GetResult]:
-        # Get all the items in the collection.
+        # Get all the items in the collection. This can be very resource-intensive for large collections.
         collection_name = collection_name.replace("-", "_")
-        result = self.client.query(
-            collection_name=f"{self.collection_prefix}_{collection_name}",
-            filter='id != ""',
+        log.warning(
+            f"Fetching ALL items from collection '{self.collection_prefix}_{collection_name}'. This might be slow for large collections."
         )
-        return self._result_to_get_result([result])
+        # Using query with a trivial filter to get all items.
+        # This will use the paginated query logic.
+        return self.query(collection_name=collection_name, filter={}, limit=None)
 
     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.
@@ -230,10 +287,23 @@ class MilvusClient(VectorDBBase):
         if not self.client.has_collection(
             collection_name=f"{self.collection_prefix}_{collection_name}"
         ):
+            log.info(
+                f"Collection {self.collection_prefix}_{collection_name} does not exist. Creating now."
+            )
+            if not items:
+                log.error(
+                    f"Cannot create collection {self.collection_prefix}_{collection_name} without items to determine dimension."
+                )
+                raise ValueError(
+                    "Cannot create Milvus collection without items to determine vector dimension."
+                )
             self._create_collection(
                 collection_name=collection_name, dimension=len(items[0]["vector"])
             )
 
+        log.info(
+            f"Inserting {len(items)} items into collection {self.collection_prefix}_{collection_name}."
+        )
         return self.client.insert(
             collection_name=f"{self.collection_prefix}_{collection_name}",
             data=[
@@ -253,10 +323,23 @@ class MilvusClient(VectorDBBase):
         if not self.client.has_collection(
             collection_name=f"{self.collection_prefix}_{collection_name}"
         ):
+            log.info(
+                f"Collection {self.collection_prefix}_{collection_name} does not exist for upsert. Creating now."
+            )
+            if not items:
+                log.error(
+                    f"Cannot create collection {self.collection_prefix}_{collection_name} for upsert without items to determine dimension."
+                )
+                raise ValueError(
+                    "Cannot create Milvus collection for upsert without items to determine vector dimension."
+                )
             self._create_collection(
                 collection_name=collection_name, dimension=len(items[0]["vector"])
             )
 
+        log.info(
+            f"Upserting {len(items)} items into collection {self.collection_prefix}_{collection_name}."
+        )
         return self.client.upsert(
             collection_name=f"{self.collection_prefix}_{collection_name}",
             data=[
@@ -276,30 +359,55 @@ class MilvusClient(VectorDBBase):
         ids: Optional[list[str]] = None,
         filter: Optional[dict] = None,
     ):
-        # Delete the items from the collection based on the ids.
+        # Delete the items from the collection based on the ids or filter.
         collection_name = collection_name.replace("-", "_")
+        if not self.has_collection(collection_name):
+            log.warning(
+                f"Delete attempted on non-existent collection: {self.collection_prefix}_{collection_name}"
+            )
+            return None
+
         if ids:
+            log.info(
+                f"Deleting items by IDs from {self.collection_prefix}_{collection_name}. IDs: {ids}"
+            )
             return self.client.delete(
                 collection_name=f"{self.collection_prefix}_{collection_name}",
                 ids=ids,
             )
         elif filter:
-            # Convert the filter dictionary to a string using JSON_CONTAINS.
             filter_string = " && ".join(
                 [
                     f'metadata["{key}"] == {json.dumps(value)}'
                     for key, value in filter.items()
                 ]
             )
-
+            log.info(
+                f"Deleting items by filter from {self.collection_prefix}_{collection_name}. Filter: {filter_string}"
+            )
             return self.client.delete(
                 collection_name=f"{self.collection_prefix}_{collection_name}",
                 filter=filter_string,
             )
+        else:
+            log.warning(
+                f"Delete operation on {self.collection_prefix}_{collection_name} called without IDs or filter. No action taken."
+            )
+            return None
 
     def reset(self):
-        # Resets the database. This will delete all collections and item entries.
+        # Resets the database. This will delete all collections and item entries that match the prefix.
+        log.warning(
+            f"Resetting Milvus: Deleting all collections with prefix '{self.collection_prefix}'."
+        )
         collection_names = self.client.list_collections()
-        for collection_name in collection_names:
-            if collection_name.startswith(self.collection_prefix):
-                self.client.drop_collection(collection_name=collection_name)
+        deleted_collections = []
+        for collection_name_full in collection_names:
+            if collection_name_full.startswith(self.collection_prefix):
+                try:
+                    self.client.drop_collection(collection_name=collection_name_full)
+                    deleted_collections.append(collection_name_full)
+                    log.info(f"Deleted collection: {collection_name_full}")
+                except Exception as e:
+                    log.error(f"Error deleting collection {collection_name_full}: {e}")
+        log.info(f"Milvus reset complete. Deleted collections: {deleted_collections}")

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

@@ -157,10 +157,10 @@ 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(
@@ -206,6 +206,7 @@ class OpenSearchClient(VectorDBBase):
                 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(
@@ -228,6 +229,7 @@ class OpenSearchClient(VectorDBBase):
                 for item in batch
             ]
             bulk(self.client, actions)
+        self.client.indices.refresh(self._get_index_name(collection_name))
 
     def delete(
         self,
@@ -251,11 +253,12 @@ 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/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
@@ -28,7 +32,16 @@ from open_webui.retrieval.vector.main import (
     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 +52,27 @@ 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 +84,24 @@ 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 +111,15 @@ 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()
 
@@ -147,44 +197,39 @@ 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"],
-                )
-                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}")
-            raise
-
-    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 existing:
-                    existing.vector = vector
-                    existing.text = item["text"]
-                    existing.vmetadata = item["metadata"]
-                    existing.collection_name = (
-                        collection_name  # Update collection_name if necessary
+            if PGVECTOR_PGCRYPTO:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    # Use raw SQL for BYTEA/pgcrypto
+                    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": json.dumps(item["metadata"]),
+                            "key": PGVECTOR_PGCRYPTO_KEY,
+                        },
                     )
-                else:
+                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,
@@ -192,11 +237,78 @@ class PgvectorClient(VectorDBBase):
                         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}'."
-            )
+                    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}")
+            raise
+
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        try:
+            if PGVECTOR_PGCRYPTO:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    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": json.dumps(item["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 = 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 +342,32 @@ 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))
@@ -299,17 +427,43 @@ 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
@@ -331,20 +485,38 @@ 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]]
 
             return GetResult(ids=ids, documents=documents, metadatas=metadatas)
         except Exception as e:
@@ -358,17 +530,33 @@ 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:

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

@@ -1,7 +1,22 @@
 from typing import Optional, List, Dict, Any, Union
 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,
     VectorItem,
@@ -40,8 +55,28 @@ class PineconeClient(VectorDBBase):
         self.metric = PINECONE_METRIC
         self.cloud = PINECONE_CLOUD
 
-        # Initialize Pinecone client
-        self.client = Pinecone(api_key=self.api_key)
+        # Initialize Pinecone client for improved performance
+        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)
 
         # Create index if it doesn't exist
         self._initialize_index()
@@ -82,12 +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]]:
@@ -135,8 +211,8 @@ class PineconeClient(VectorDBBase):
         metadatas = []
 
         for match in matches:
-            metadata = match.get("metadata", {})
-            ids.append(match["id"])
+            metadata = getattr(match, "metadata", {}) or {}
+            ids.append(match.id if hasattr(match, "id") else match["id"])
             documents.append(metadata.get("text", ""))
             metadatas.append(metadata)
 
@@ -162,7 +238,8 @@ class PineconeClient(VectorDBBase):
                 filter={"collection_name": collection_name_with_prefix},
                 include_metadata=False,
             )
-            return len(response.matches) > 0
+            matches = getattr(response, "matches", []) or []
+            return len(matches) > 0
         except Exception as e:
             log.exception(
                 f"Error checking collection '{collection_name_with_prefix}': {e}"
@@ -191,27 +268,30 @@ class PineconeClient(VectorDBBase):
             log.warning("No items to insert")
             return
 
+        start_time = time.time()
+
         collection_name_with_prefix = self._get_collection_name_with_prefix(
             collection_name
         )
         points = self._create_points(items, collection_name_with_prefix)
 
-        # Insert in batches for better performance and reliability
+        # Parallelize batch inserts for performance
+        executor = self._executor
+        futures = []
         for i in range(0, len(points), BATCH_SIZE):
             batch = points[i : i + BATCH_SIZE]
+            futures.append(executor.submit(self.index.upsert, vectors=batch))
+        for future in concurrent.futures.as_completed(futures):
             try:
-                self.index.upsert(vectors=batch)
-                log.debug(
-                    f"Inserted batch of {len(batch)} vectors into '{collection_name_with_prefix}'"
-                )
+                future.result()
             except Exception as e:
-                log.error(
-                    f"Error inserting batch into '{collection_name_with_prefix}': {e}"
-                )
+                log.error(f"Error inserting batch: {e}")
                 raise
-
+        elapsed = time.time() - start_time
+        log.debug(f"Insert of {len(points)} vectors took {elapsed:.2f} seconds")
         log.info(
-            f"Successfully inserted {len(items)} vectors 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:
@@ -220,27 +300,94 @@ class PineconeClient(VectorDBBase):
             log.warning("No items to upsert")
             return
 
+        start_time = time.time()
+
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+        points = self._create_points(items, collection_name_with_prefix)
+
+        # Parallelize batch upserts for performance
+        executor = self._executor
+        futures = []
+        for i in range(0, len(points), BATCH_SIZE):
+            batch = points[i : i + BATCH_SIZE]
+            futures.append(executor.submit(self.index.upsert, vectors=batch))
+        for future in concurrent.futures.as_completed(futures):
+            try:
+                future.result()
+            except Exception as e:
+                log.error(f"Error upserting batch: {e}")
+                raise
+        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 "
+            f"into '{collection_name_with_prefix}'"
+        )
+
+    async def insert_async(self, collection_name: str, items: List[VectorItem]) -> None:
+        """Async version of insert using asyncio and run_in_executor for improved performance."""
+        if not items:
+            log.warning("No items to insert")
+            return
+
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+        points = self._create_points(items, collection_name_with_prefix)
+
+        # Create batches
+        batches = [
+            points[i : i + BATCH_SIZE] for i in range(0, len(points), BATCH_SIZE)
+        ]
+        loop = asyncio.get_event_loop()
+        tasks = [
+            loop.run_in_executor(
+                None, functools.partial(self.index.upsert, vectors=batch)
+            )
+            for batch in batches
+        ]
+        results = await asyncio.gather(*tasks, return_exceptions=True)
+        for result in results:
+            if isinstance(result, Exception):
+                log.error(f"Error in async insert batch: {result}")
+                raise result
+        log.info(
+            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:
+        """Async version of upsert using asyncio and run_in_executor for improved performance."""
+        if not items:
+            log.warning("No items to upsert")
+            return
+
         collection_name_with_prefix = self._get_collection_name_with_prefix(
             collection_name
         )
         points = self._create_points(items, collection_name_with_prefix)
 
-        # Upsert in batches
-        for i in range(0, len(points), BATCH_SIZE):
-            batch = points[i : i + BATCH_SIZE]
-            try:
-                self.index.upsert(vectors=batch)
-                log.debug(
-                    f"Upserted batch of {len(batch)} vectors into '{collection_name_with_prefix}'"
-                )
-            except Exception as e:
-                log.error(
-                    f"Error upserting batch into '{collection_name_with_prefix}': {e}"
-                )
-                raise
-
+        # Create batches
+        batches = [
+            points[i : i + BATCH_SIZE] for i in range(0, len(points), BATCH_SIZE)
+        ]
+        loop = asyncio.get_event_loop()
+        tasks = [
+            loop.run_in_executor(
+                None, functools.partial(self.index.upsert, vectors=batch)
+            )
+            for batch in batches
+        ]
+        results = await asyncio.gather(*tasks, return_exceptions=True)
+        for result in results:
+            if isinstance(result, Exception):
+                log.error(f"Error in async upsert batch: {result}")
+                raise result
         log.info(
-            f"Successfully upserted {len(items)} vectors into '{collection_name_with_prefix}'"
+            f"Successfully async upserted {len(points)} vectors in batches "
+            f"into '{collection_name_with_prefix}'"
         )
 
     def search(
@@ -270,7 +417,8 @@ class PineconeClient(VectorDBBase):
                 filter={"collection_name": collection_name_with_prefix},
             )
 
-            if not query_response.matches:
+            matches = getattr(query_response, "matches", []) or []
+            if not matches:
                 # Return empty result if no matches
                 return SearchResult(
                     ids=[[]],
@@ -280,13 +428,13 @@ class PineconeClient(VectorDBBase):
                 )
 
             # Convert to GetResult format
-            get_result = self._result_to_get_result(query_response.matches)
+            get_result = self._result_to_get_result(matches)
 
             # Calculate normalized distances based on metric
             distances = [
                 [
-                    self._normalize_distance(match.score)
-                    for match in query_response.matches
+                    self._normalize_distance(getattr(match, "score", 0.0))
+                    for match in matches
                 ]
             ]
 
@@ -328,7 +476,8 @@ class PineconeClient(VectorDBBase):
                 include_metadata=True,
             )
 
-            return self._result_to_get_result(query_response.matches)
+            matches = getattr(query_response, "matches", []) or []
+            return self._result_to_get_result(matches)
 
         except Exception as e:
             log.error(f"Error querying collection '{collection_name}': {e}")
@@ -352,7 +501,8 @@ class PineconeClient(VectorDBBase):
                 filter={"collection_name": collection_name_with_prefix},
             )
 
-            return self._result_to_get_result(query_response.matches)
+            matches = getattr(query_response, "matches", []) or []
+            return self._result_to_get_result(matches)
 
         except Exception as e:
             log.error(f"Error getting collection '{collection_name}': {e}")
@@ -378,10 +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:
@@ -410,3 +562,20 @@ class PineconeClient(VectorDBBase):
         except Exception as e:
             log.error(f"Failed to reset Pinecone index: {e}")
             raise
+
+    def close(self):
+        """Shut down resources."""
+        try:
+            # The new Pinecone client doesn't need explicit closing
+            pass
+        except Exception as e:
+            log.warning(f"Failed to clean up Pinecone resources: {e}")
+        self._executor.shutdown(wait=True)
+
+    def __enter__(self):
+        """Enter context manager."""
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """Exit context manager, ensuring resources are cleaned up."""
+        self.close()

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

@@ -18,6 +18,7 @@ from open_webui.config import (
     QDRANT_ON_DISK,
     QDRANT_GRPC_PORT,
     QDRANT_PREFER_GRPC,
+    QDRANT_COLLECTION_PREFIX,
 )
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -29,7 +30,7 @@ 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
@@ -86,6 +87,25 @@ class QdrantClient(VectorDBBase):
             ),
         )
 
+        # 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):

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

@@ -0,0 +1,350 @@
+import logging
+from typing import Optional, Tuple, List, Dict, Any
+from urllib.parse import urlparse
+
+import grpc
+from open_webui.config import (
+    QDRANT_API_KEY,
+    QDRANT_GRPC_PORT,
+    QDRANT_ON_DISK,
+    QDRANT_PREFER_GRPC,
+    QDRANT_URI,
+    QDRANT_COLLECTION_PREFIX,
+)
+from open_webui.env import SRC_LOG_LEVELS
+from open_webui.retrieval.vector.main import (
+    GetResult,
+    SearchResult,
+    VectorDBBase,
+    VectorItem,
+)
+from qdrant_client import QdrantClient as Qclient
+from qdrant_client.http.exceptions import UnexpectedResponse
+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 = 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
+
+        if not self.QDRANT_URI:
+            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
+
+        self.client = (
+            Qclient(
+                host=host,
+                port=http_port,
+                grpc_port=self.GRPC_PORT,
+                prefer_grpc=self.PREFER_GRPC,
+                api_key=self.QDRANT_API_KEY,
+            )
+            if self.PREFER_GRPC
+            else Qclient(url=self.QDRANT_URI, api_key=self.QDRANT_API_KEY)
+        )
+
+        # Main collection types for multi-tenancy
+        self.MEMORY_COLLECTION = f"{self.collection_prefix}_memories"
+        self.KNOWLEDGE_COLLECTION = f"{self.collection_prefix}_knowledge"
+        self.FILE_COLLECTION = f"{self.collection_prefix}_files"
+        self.WEB_SEARCH_COLLECTION = f"{self.collection_prefix}_web-search"
+        self.HASH_BASED_COLLECTION = f"{self.collection_prefix}_hash-based"
+
+    def _result_to_get_result(self, points) -> GetResult:
+        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])
+
+    def _get_collection_and_tenant_id(self, collection_name: str) -> Tuple[str, str]:
+        """
+        Maps the traditional collection name to multi-tenant collection and tenant ID.
+
+        Returns:
+            tuple: (collection_name, tenant_id)
+        """
+        # Check for user memory collections
+        tenant_id = collection_name
+
+        if collection_name.startswith("user-memory-"):
+            return self.MEMORY_COLLECTION, tenant_id
+
+        # Check for file collections
+        elif collection_name.startswith("file-"):
+            return self.FILE_COLLECTION, tenant_id
+
+        # Check for web search collections
+        elif collection_name.startswith("web-search-"):
+            return self.WEB_SEARCH_COLLECTION, tenant_id
+
+        # Handle hash-based collections (YouTube and web URLs)
+        elif len(collection_name) == 63 and all(
+            c in "0123456789abcdef" for c in collection_name
+        ):
+            return self.HASH_BASED_COLLECTION, tenant_id
+
+        else:
+            return self.KNOWLEDGE_COLLECTION, tenant_id
+
+    def _create_multi_tenant_collection(
+        self, mt_collection_name: str, dimension: int = DEFAULT_DIMENSION
+    ):
+        """
+        Creates a collection with multi-tenancy configuration and payload indexes for tenant_id and metadata fields.
+        """
+        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,
+            ),
+        )
+        log.info(
+            f"Multi-tenant collection {mt_collection_name} created with dimension {dimension}!"
+        )
+
+        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=field,
+                field_schema=models.KeywordIndexParams(
+                    type=models.KeywordIndexType.KEYWORD,
+                    on_disk=self.QDRANT_ON_DISK,
+                ),
+            )
+
+    def _create_points(
+        self, items: List[VectorItem], tenant_id: str
+    ) -> List[PointStruct]:
+        """
+        Create point structs from vector items with tenant ID.
+        """
+        return [
+            PointStruct(
+                id=item["id"],
+                vector=item["vector"],
+                payload={
+                    "text": item["text"],
+                    "metadata": item["metadata"],
+                    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
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+        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[str, Any]] = None,
+    ):
+        """
+        Delete vectors by ID or filter from a collection with tenant isolation.
+        """
+        if not self.client:
+            return None
+
+        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
+
+        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)
+            ),
+        )
+
+    def search(
+        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 or not vectors:
+            return None
+        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, search returns None")
+            return 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
+        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, query returns None")
+            return None
+        if limit is None:
+            limit = NO_LIMIT
+        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])
+        points = self.client.query_points(
+            collection_name=mt_collection,
+            query_filter=combined_filter,
+            limit=limit,
+        )
+        return self._result_to_get_result(points.points)
+
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        """
+        Get all items in a collection with tenant isolation.
+        """
+        if not self.client:
+            return None
+        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, get returns None")
+            return None
+        tenant_filter = _tenant_filter(tenant_id)
+        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)
+
+    def upsert(self, collection_name: str, items: List[VectorItem]):
+        """
+        Upsert items with tenant ID.
+        """
+        if not self.client or not items:
+            return None
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+        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
+
+    def insert(self, collection_name: str, items: List[VectorItem]):
+        """
+        Insert items with tenant ID.
+        """
+        return self.upsert(collection_name, items)
+
+    def reset(self):
+        """
+        Reset the database by deleting all collections.
+        """
+        if not self.client:
+            return None
+        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):
+        """
+        Delete a collection.
+        """
+        if not self.client:
+            return None
+        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
+        self.client.delete(
+            collection_name=mt_collection,
+            points_selector=models.FilterSelector(
+                filter=models.Filter(must=[_tenant_filter(tenant_id)])
+            ),
+        )

backend/open_webui/retrieval/vector/factory.py

@@ -0,0 +1,55 @@
+from open_webui.retrieval.vector.main import VectorDBBase
+from open_webui.retrieval.vector.type import VectorType
+from open_webui.config import VECTOR_DB, ENABLE_QDRANT_MULTITENANCY_MODE
+
+
+class Vector:
+
+    @staticmethod
+    def get_vector(vector_type: str) -> VectorDBBase:
+        """
+        get vector db instance by vector type
+        """
+        match vector_type:
+            case VectorType.MILVUS:
+                from open_webui.retrieval.vector.dbs.milvus import MilvusClient
+
+                return MilvusClient()
+            case VectorType.QDRANT:
+                if ENABLE_QDRANT_MULTITENANCY_MODE:
+                    from open_webui.retrieval.vector.dbs.qdrant_multitenancy import (
+                        QdrantClient,
+                    )
+
+                    return QdrantClient()
+                else:
+                    from open_webui.retrieval.vector.dbs.qdrant import QdrantClient
+
+                    return QdrantClient()
+            case VectorType.PINECONE:
+                from open_webui.retrieval.vector.dbs.pinecone import PineconeClient
+
+                return PineconeClient()
+            case VectorType.OPENSEARCH:
+                from open_webui.retrieval.vector.dbs.opensearch import OpenSearchClient
+
+                return OpenSearchClient()
+            case VectorType.PGVECTOR:
+                from open_webui.retrieval.vector.dbs.pgvector import PgvectorClient
+
+                return PgvectorClient()
+            case VectorType.ELASTICSEARCH:
+                from open_webui.retrieval.vector.dbs.elasticsearch import (
+                    ElasticsearchClient,
+                )
+
+                return ElasticsearchClient()
+            case VectorType.CHROMA:
+                from open_webui.retrieval.vector.dbs.chroma import ChromaClient
+
+                return ChromaClient()
+            case _:
+                raise ValueError(f"Unsupported vector type: {vector_type}")
+
+
+VECTOR_DB_CLIENT = Vector.get_vector(VECTOR_DB)

backend/open_webui/retrieval/vector/type.py

@@ -0,0 +1,11 @@
+from enum import StrEnum
+
+
+class VectorType(StrEnum):
+    MILVUS = "milvus"
+    QDRANT = "qdrant"
+    CHROMA = "chroma"
+    PINECONE = "pinecone"
+    ELASTICSEARCH = "elasticsearch"
+    OPENSEARCH = "opensearch"
+    PGVECTOR = "pgvector"

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/retrieval/web/searchapi.py

@@ -42,7 +42,9 @@ def search_searchapi(
         results = get_filtered_results(results, filter_list)
     return [
         SearchResult(
-            link=result["link"], title=result["title"], snippet=result["snippet"]
+            link=result["link"],
+            title=result.get("title"),
+            snippet=result.get("snippet"),
         )
         for result in results[:count]
     ]

backend/open_webui/retrieval/web/serpapi.py

@@ -42,7 +42,9 @@ def search_serpapi(
         results = get_filtered_results(results, filter_list)
     return [
         SearchResult(
-            link=result["link"], title=result["title"], snippet=result["snippet"]
+            link=result["link"],
+            title=result.get("title"),
+            snippet=result.get("snippet"),
         )
         for result in results[:count]
     ]

backend/open_webui/retrieval/web/utils.py

@@ -25,7 +25,7 @@ from langchain_community.document_loaders.firecrawl import FireCrawlLoader
 from langchain_community.document_loaders.base import BaseLoader
 from langchain_core.documents import Document
 from open_webui.retrieval.loaders.tavily import TavilyLoader
-from open_webui.retrieval.loaders.external import ExternalLoader
+from open_webui.retrieval.loaders.external_web import ExternalWebLoader
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.config import (
     ENABLE_RAG_LOCAL_WEB_FETCH,
@@ -39,7 +39,7 @@ from open_webui.config import (
     EXTERNAL_WEB_LOADER_URL,
     EXTERNAL_WEB_LOADER_API_KEY,
 )
-from open_webui.env import SRC_LOG_LEVELS
+from open_webui.env import SRC_LOG_LEVELS, AIOHTTP_CLIENT_SESSION_SSL
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
@@ -515,7 +515,8 @@ class SafeWebBaseLoader(WebBaseLoader):
                         kwargs["ssl"] = False
 
                     async with session.get(
-                        url, **(self.requests_kwargs | kwargs)
+                        url,
+                        **(self.requests_kwargs | kwargs),
                     ) as response:
                         if self.raise_for_status:
                             response.raise_for_status()
@@ -628,7 +629,7 @@ def get_web_loader(
         web_loader_args["extract_depth"] = TAVILY_EXTRACT_DEPTH.value
 
     if WEB_LOADER_ENGINE.value == "external":
-        WebLoaderClass = ExternalLoader
+        WebLoaderClass = ExternalWebLoader
         web_loader_args["external_url"] = EXTERNAL_WEB_LOADER_URL.value
         web_loader_args["external_api_key"] = EXTERNAL_WEB_LOADER_API_KEY.value
 

backend/open_webui/routers/audio.py

@@ -7,16 +7,21 @@ from functools import lru_cache
 from pathlib import Path
 from pydub import AudioSegment
 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,
     FastAPI,
     File,
+    Form,
     HTTPException,
     Request,
     UploadFile,
@@ -38,6 +43,7 @@ from open_webui.config import (
 
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import (
+    AIOHTTP_CLIENT_SESSION_SSL,
     AIOHTTP_CLIENT_TIMEOUT,
     ENV,
     SRC_LOG_LEVELS,
@@ -49,7 +55,7 @@ from open_webui.env import (
 router = APIRouter()
 
 # Constants
-MAX_FILE_SIZE_MB = 25
+MAX_FILE_SIZE_MB = 20
 MAX_FILE_SIZE = MAX_FILE_SIZE_MB * 1024 * 1024  # Convert MB to bytes
 AZURE_MAX_FILE_SIZE_MB = 200
 AZURE_MAX_FILE_SIZE = AZURE_MAX_FILE_SIZE_MB * 1024 * 1024  # Convert MB to bytes
@@ -71,31 +77,47 @@ from pydub import AudioSegment
 from pydub.utils import mediainfo
 
 
-def get_audio_format(file_path):
-    """Check if the given file needs to be converted to a different format."""
+def is_audio_conversion_required(file_path):
+    """
+    Check if the given audio file needs conversion to mp3.
+    """
+    SUPPORTED_FORMATS = {"flac", "m4a", "mp3", "mp4", "mpeg", "wav", "webm"}
+
     if not os.path.isfile(file_path):
         log.error(f"File not found: {file_path}")
         return False
 
-    info = mediainfo(file_path)
-    if (
-        info.get("codec_name") == "aac"
-        and info.get("codec_type") == "audio"
-        and info.get("codec_tag_string") == "mp4a"
-    ):
-        return "mp4"
-    elif info.get("format_name") == "ogg":
-        return "ogg"
-    elif info.get("format_name") == "matroska,webm":
-        return "webm"
-    return None
+    try:
+        info = mediainfo(file_path)
+        codec_name = info.get("codec_name", "").lower()
+        codec_type = info.get("codec_type", "").lower()
+        codec_tag_string = info.get("codec_tag_string", "").lower()
+
+        if codec_name == "aac" and codec_type == "audio" and codec_tag_string == "mp4a":
+            # File is AAC/mp4a audio, recommend mp3 conversion
+            return True
+
+        # If the codec name is in the supported formats
+        if codec_name in SUPPORTED_FORMATS:
+            return False
+
+        return True
+    except Exception as e:
+        log.error(f"Error getting audio format: {e}")
+        return False
 
 
-def convert_audio_to_wav(file_path, output_path, conversion_type):
-    """Convert MP4/OGG audio file to WAV format."""
-    audio = AudioSegment.from_file(file_path, format=conversion_type)
-    audio.export(output_path, format="wav")
-    log.info(f"Converted {file_path} to {output_path}")
+def convert_audio_to_mp3(file_path):
+    """Convert audio file to mp3 format."""
+    try:
+        output_path = os.path.splitext(file_path)[0] + ".mp3"
+        audio = AudioSegment.from_file(file_path)
+        audio.export(output_path, format="mp3")
+        log.info(f"Converted {file_path} to {output_path}")
+        return output_path
+    except Exception as e:
+        log.error(f"Error converting audio file: {e}")
+        return None
 
 
 def set_faster_whisper_model(model: str, auto_update: bool = False):
@@ -147,6 +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
@@ -181,6 +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,
@@ -215,6 +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
@@ -229,6 +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": {
@@ -248,6 +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,
@@ -297,6 +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
 
@@ -305,7 +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={
@@ -313,7 +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,
@@ -322,14 +354,16 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                             else {}
                         ),
                     },
-                ) as r:
-                    r.raise_for_status()
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
+                )
 
-                    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)
 
@@ -337,18 +371,18 @@ 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 "error" in res:
-                        detail = f"External: {res['error'].get('message', '')}"
-            except Exception:
-                detail = f"External: {e}"
+            if r is not None:
+                status_code = r.status
+                res = await r.json()
+                if "error" in res:
+                    detail = f"External: {res['error'].get('message', '')}"
 
             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":
@@ -377,6 +411,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                         "Content-Type": "application/json",
                         "xi-api-key": request.app.state.config.TTS_API_KEY,
                     },
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as r:
                     r.raise_for_status()
 
@@ -435,6 +470,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                         "X-Microsoft-OutputFormat": output_format,
                     },
                     data=data,
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as r:
                     r.raise_for_status()
 
@@ -503,12 +539,13 @@ async def speech(request: Request, user=Depends(get_verified_user)):
         return FileResponse(file_path)
 
 
-def transcribe(request: Request, file_path):
-    log.info(f"transcribe: {file_path}")
+def transcription_handler(request, file_path, metadata):
     filename = os.path.basename(file_path)
     file_dir = os.path.dirname(file_path)
     id = filename.split(".")[0]
 
+    metadata = metadata or {}
+
     if request.app.state.config.STT_ENGINE == "":
         if request.app.state.faster_whisper_model is None:
             request.app.state.faster_whisper_model = set_faster_whisper_model(
@@ -520,7 +557,7 @@ def transcribe(request: Request, file_path):
             file_path,
             beam_size=5,
             vad_filter=request.app.state.config.WHISPER_VAD_FILTER,
-            language=WHISPER_LANGUAGE,
+            language=metadata.get("language") or WHISPER_LANGUAGE,
         )
         log.info(
             "Detected language '%s' with probability %f"
@@ -538,16 +575,6 @@ def transcribe(request: Request, file_path):
         log.debug(data)
         return data
     elif request.app.state.config.STT_ENGINE == "openai":
-        audio_format = get_audio_format(file_path)
-        if audio_format:
-            os.rename(file_path, file_path.replace(".wav", f".{audio_format}"))
-            # Convert unsupported audio file to WAV format
-            convert_audio_to_wav(
-                file_path.replace(".wav", f".{audio_format}"),
-                file_path,
-                audio_format,
-            )
-
         r = None
         try:
             r = requests.post(
@@ -556,7 +583,14 @@ def transcribe(request: Request, file_path):
                     "Authorization": f"Bearer {request.app.state.config.STT_OPENAI_API_KEY}"
                 },
                 files={"file": (filename, open(file_path, "rb"))},
-                data={"model": request.app.state.config.STT_MODEL},
+                data={
+                    "model": request.app.state.config.STT_MODEL,
+                    **(
+                        {"language": metadata.get("language")}
+                        if metadata.get("language")
+                        else {}
+                    ),
+                },
             )
 
             r.raise_for_status()
@@ -606,7 +640,7 @@ def transcribe(request: Request, file_path):
 
             # 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,
@@ -764,41 +798,143 @@ def transcribe(request: Request, file_path):
             )
 
 
+def transcribe(request: Request, file_path: str, metadata: Optional[dict] = None):
+    log.info(f"transcribe: {file_path} {metadata}")
+
+    if is_audio_conversion_required(file_path):
+        file_path = convert_audio_to_mp3(file_path)
+
+    try:
+        file_path = compress_audio(file_path)
+    except Exception as e:
+        log.exception(e)
+
+    # Always produce a list of chunk paths (could be one entry if small)
+    try:
+        chunk_paths = split_audio(file_path, MAX_FILE_SIZE)
+        print(f"Chunk paths: {chunk_paths}")
+    except Exception as e:
+        log.exception(e)
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.DEFAULT(e),
+        )
+
+    results = []
+    try:
+        with ThreadPoolExecutor() as executor:
+            # Submit tasks for each chunk_path
+            futures = [
+                executor.submit(transcription_handler, request, chunk_path, metadata)
+                for chunk_path in chunk_paths
+            ]
+            # Gather results as they complete
+            for future in futures:
+                try:
+                    results.append(future.result())
+                except Exception as transcribe_exc:
+                    raise HTTPException(
+                        status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                        detail=f"Error transcribing chunk: {transcribe_exc}",
+                    )
+    finally:
+        # Clean up only the temporary chunks, never the original file
+        for chunk_path in chunk_paths:
+            if chunk_path != file_path and os.path.isfile(chunk_path):
+                try:
+                    os.remove(chunk_path)
+                except Exception:
+                    pass
+
+    return {
+        "text": " ".join([result["text"] for result in results]),
+    }
+
+
 def compress_audio(file_path):
     if os.path.getsize(file_path) > MAX_FILE_SIZE:
+        id = os.path.splitext(os.path.basename(file_path))[
+            0
+        ]  # Handles names with multiple dots
         file_dir = os.path.dirname(file_path)
+
         audio = AudioSegment.from_file(file_path)
         audio = audio.set_frame_rate(16000).set_channels(1)  # Compress audio
-        compressed_path = f"{file_dir}/{id}_compressed.opus"
-        audio.export(compressed_path, format="opus", bitrate="32k")
-        log.debug(f"Compressed audio to {compressed_path}")
 
-        if (
-            os.path.getsize(compressed_path) > MAX_FILE_SIZE
-        ):  # Still larger than MAX_FILE_SIZE after compression
-            raise Exception(ERROR_MESSAGES.FILE_TOO_LARGE(size=f"{MAX_FILE_SIZE_MB}MB"))
+        compressed_path = os.path.join(file_dir, f"{id}_compressed.mp3")
+        audio.export(compressed_path, format="mp3", bitrate="32k")
+        # log.debug(f"Compressed audio to {compressed_path}")  # Uncomment if log is defined
+
         return compressed_path
     else:
         return file_path
 
 
+def split_audio(file_path, max_bytes, format="mp3", bitrate="32k"):
+    """
+    Splits audio into chunks not exceeding max_bytes.
+    Returns a list of chunk file paths. If audio fits, returns list with original path.
+    """
+    file_size = os.path.getsize(file_path)
+    if file_size <= max_bytes:
+        return [file_path]  # Nothing to split
+
+    audio = AudioSegment.from_file(file_path)
+    duration_ms = len(audio)
+    orig_size = file_size
+
+    approx_chunk_ms = max(int(duration_ms * (max_bytes / orig_size)) - 1000, 1000)
+    chunks = []
+    start = 0
+    i = 0
+
+    base, _ = os.path.splitext(file_path)
+
+    while start < duration_ms:
+        end = min(start + approx_chunk_ms, duration_ms)
+        chunk = audio[start:end]
+        chunk_path = f"{base}_chunk_{i}.{format}"
+        chunk.export(chunk_path, format=format, bitrate=bitrate)
+
+        # Reduce chunk duration if still too large
+        while os.path.getsize(chunk_path) > max_bytes and (end - start) > 5000:
+            end = start + ((end - start) // 2)
+            chunk = audio[start:end]
+            chunk.export(chunk_path, format=format, bitrate=bitrate)
+
+        if os.path.getsize(chunk_path) > max_bytes:
+            os.remove(chunk_path)
+            raise Exception("Audio chunk cannot be reduced below max file size.")
+
+        chunks.append(chunk_path)
+        start = end
+        i += 1
+
+    return chunks
+
+
 @router.post("/transcriptions")
 def transcription(
     request: Request,
     file: UploadFile = File(...),
+    language: Optional[str] = Form(None),
     user=Depends(get_verified_user),
 ):
     log.info(f"file.content_type: {file.content_type}")
 
-    supported_filetypes = (
-        "audio/mpeg",
-        "audio/wav",
-        "audio/ogg",
-        "audio/x-m4a",
-        "audio/webm",
+    stt_supported_content_types = getattr(
+        request.app.state.config, "STT_SUPPORTED_CONTENT_TYPES", []
     )
 
-    if not file.content_type.startswith(supported_filetypes):
+    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,
             detail=ERROR_MESSAGES.FILE_NOT_SUPPORTED,
@@ -819,19 +955,18 @@ def transcription(
             f.write(contents)
 
         try:
-            try:
-                file_path = compress_audio(file_path)
-            except Exception as e:
-                log.exception(e)
+            metadata = None
 
-                raise HTTPException(
-                    status_code=status.HTTP_400_BAD_REQUEST,
-                    detail=ERROR_MESSAGES.DEFAULT(e),
-                )
+            if language:
+                metadata = {"language": language}
+
+            result = transcribe(request, file_path, metadata)
+
+            return {
+                **result,
+                "filename": os.path.basename(file_path),
+            }
 
-            data = transcribe(request, file_path)
-            file_path = file_path.split("/")[-1]
-            return {**data, "filename": file_path}
         except Exception as e:
             log.exception(e)
 

backend/open_webui/routers/auths.py

@@ -19,19 +19,21 @@ from open_webui.models.auths import (
     UserResponse,
 )
 from open_webui.models.users import Users
+from open_webui.models.groups import Groups
 
 from open_webui.constants import ERROR_MESSAGES, WEBHOOK_MESSAGES
 from open_webui.env import (
     WEBUI_AUTH,
     WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
     WEBUI_AUTH_TRUSTED_NAME_HEADER,
+    WEBUI_AUTH_TRUSTED_GROUPS_HEADER,
     WEBUI_AUTH_COOKIE_SAME_SITE,
     WEBUI_AUTH_COOKIE_SECURE,
     WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
     SRC_LOG_LEVELS,
 )
 from fastapi import APIRouter, Depends, HTTPException, Request, status
-from fastapi.responses import RedirectResponse, Response
+from fastapi.responses import RedirectResponse, Response, JSONResponse
 from open_webui.config import OPENID_PROVIDER_URL, ENABLE_OAUTH_SIGNUP, ENABLE_LDAP
 from pydantic import BaseModel
 
@@ -51,11 +53,10 @@ from open_webui.utils.access_control import get_permissions
 
 from typing import Optional, List
 
-from ssl import CERT_REQUIRED, PROTOCOL_TLS
+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()
 
@@ -186,6 +187,9 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
     LDAP_APP_PASSWORD = request.app.state.config.LDAP_APP_PASSWORD
     LDAP_USE_TLS = request.app.state.config.LDAP_USE_TLS
     LDAP_CA_CERT_FILE = request.app.state.config.LDAP_CA_CERT_FILE
+    LDAP_VALIDATE_CERT = (
+        CERT_REQUIRED if request.app.state.config.LDAP_VALIDATE_CERT else CERT_NONE
+    )
     LDAP_CIPHERS = (
         request.app.state.config.LDAP_CIPHERS
         if request.app.state.config.LDAP_CIPHERS
@@ -197,7 +201,7 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
 
     try:
         tls = Tls(
-            validate=CERT_REQUIRED,
+            validate=LDAP_VALIDATE_CERT,
             version=PROTOCOL_TLS,
             ca_certs_file=LDAP_CA_CERT_FILE,
             ciphers=LDAP_CIPHERS,
@@ -224,17 +228,33 @@ 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:
+        if not search_success or not connection_app.entries:
             raise HTTPException(400, detail="User not found in the LDAP server")
 
         entry = connection_app.entries[0]
@@ -254,6 +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,
@@ -296,7 +379,7 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                         500, detail="Internal error occurred during LDAP user creation."
                     )
 
-            user = Auths.authenticate_user_by_trusted_header(email)
+            user = Auths.authenticate_user_by_email(email)
 
             if user:
                 expires_delta = parse_duration(request.app.state.config.JWT_EXPIRES_IN)
@@ -329,6 +412,22 @@ 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",
@@ -360,21 +459,29 @@ async def signin(request: Request, response: Response, form_data: SigninForm):
         if WEBUI_AUTH_TRUSTED_EMAIL_HEADER not in request.headers:
             raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_TRUSTED_HEADER)
 
-        trusted_email = request.headers[WEBUI_AUTH_TRUSTED_EMAIL_HEADER].lower()
-        trusted_name = trusted_email
+        email = request.headers[WEBUI_AUTH_TRUSTED_EMAIL_HEADER].lower()
+        name = email
+
         if WEBUI_AUTH_TRUSTED_NAME_HEADER:
-            trusted_name = request.headers.get(
-                WEBUI_AUTH_TRUSTED_NAME_HEADER, trusted_email
-            )
-        if not Users.get_user_by_email(trusted_email.lower()):
+            name = request.headers.get(WEBUI_AUTH_TRUSTED_NAME_HEADER, email)
+
+        if not Users.get_user_by_email(email.lower()):
             await signup(
                 request,
                 response,
-                SignupForm(
-                    email=trusted_email, password=str(uuid.uuid4()), name=trusted_name
-                ),
+                SignupForm(email=email, password=str(uuid.uuid4()), name=name),
             )
-        user = Auths.authenticate_user_by_trusted_header(trusted_email)
+
+        user = Auths.authenticate_user_by_email(email)
+        if WEBUI_AUTH_TRUSTED_GROUPS_HEADER and user and user.role != "admin":
+            group_names = request.headers.get(
+                WEBUI_AUTH_TRUSTED_GROUPS_HEADER, ""
+            ).split(",")
+            group_names = [name.strip() for name in group_names if name.strip()]
+
+            if group_names:
+                Groups.sync_groups_by_group_names(user.id, group_names)
+
     elif WEBUI_AUTH == False:
         admin_email = "admin@localhost"
         admin_password = "admin"
@@ -478,10 +585,6 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
             "admin" if user_count == 0 else request.app.state.config.DEFAULT_USER_ROLE
         )
 
-        if user_count == 0:
-            # Disable signup after the first user is created
-            request.app.state.config.ENABLE_SIGNUP = False
-
         # 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:
             raise HTTPException(
@@ -541,6 +644,10 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
                 user.id, request.app.state.config.USER_PERMISSIONS
             )
 
+            if user_count == 0:
+                # Disable signup after the first user is created
+                request.app.state.config.ENABLE_SIGNUP = False
+
             return {
                 "token": token,
                 "token_type": "Bearer",
@@ -562,21 +669,32 @@ 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:
             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()
                             logout_url = openid_data.get("end_session_endpoint")
                             if logout_url:
                                 response.delete_cookie("oauth_id_token")
-                                return RedirectResponse(
+
+                                return JSONResponse(
+                                    status_code=200,
+                                    content={
+                                        "status": True,
+                                        "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,
-                                    url=f"{logout_url}?id_token_hint={oauth_id_token}",
                                 )
                         else:
                             raise HTTPException(
@@ -591,12 +709,18 @@ async def signout(request: Request, response: Response):
                 )
 
     if WEBUI_AUTH_SIGNOUT_REDIRECT_URL:
-        return RedirectResponse(
+        return JSONResponse(
+            status_code=200,
+            content={
+                "status": True,
+                "redirect_url": WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
+            },
             headers=response.headers,
-            url=WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
         )
 
-    return {"status": True}
+    return JSONResponse(
+        status_code=200, content={"status": True}, headers=response.headers
+    )
 
 
 ############################
@@ -696,6 +820,9 @@ async def get_admin_config(request: Request, user=Depends(get_admin_user)):
         "ENABLE_CHANNELS": request.app.state.config.ENABLE_CHANNELS,
         "ENABLE_NOTES": request.app.state.config.ENABLE_NOTES,
         "ENABLE_USER_WEBHOOKS": request.app.state.config.ENABLE_USER_WEBHOOKS,
+        "PENDING_USER_OVERLAY_TITLE": request.app.state.config.PENDING_USER_OVERLAY_TITLE,
+        "PENDING_USER_OVERLAY_CONTENT": request.app.state.config.PENDING_USER_OVERLAY_CONTENT,
+        "RESPONSE_WATERMARK": request.app.state.config.RESPONSE_WATERMARK,
     }
 
 
@@ -713,6 +840,9 @@ class AdminConfig(BaseModel):
     ENABLE_CHANNELS: bool
     ENABLE_NOTES: bool
     ENABLE_USER_WEBHOOKS: bool
+    PENDING_USER_OVERLAY_TITLE: Optional[str] = None
+    PENDING_USER_OVERLAY_CONTENT: Optional[str] = None
+    RESPONSE_WATERMARK: Optional[str] = None
 
 
 @router.post("/admin/config")
@@ -750,6 +880,15 @@ async def update_admin_config(
 
     request.app.state.config.ENABLE_USER_WEBHOOKS = form_data.ENABLE_USER_WEBHOOKS
 
+    request.app.state.config.PENDING_USER_OVERLAY_TITLE = (
+        form_data.PENDING_USER_OVERLAY_TITLE
+    )
+    request.app.state.config.PENDING_USER_OVERLAY_CONTENT = (
+        form_data.PENDING_USER_OVERLAY_CONTENT
+    )
+
+    request.app.state.config.RESPONSE_WATERMARK = form_data.RESPONSE_WATERMARK
+
     return {
         "SHOW_ADMIN_DETAILS": request.app.state.config.SHOW_ADMIN_DETAILS,
         "WEBUI_URL": request.app.state.config.WEBUI_URL,
@@ -764,6 +903,9 @@ async def update_admin_config(
         "ENABLE_CHANNELS": request.app.state.config.ENABLE_CHANNELS,
         "ENABLE_NOTES": request.app.state.config.ENABLE_NOTES,
         "ENABLE_USER_WEBHOOKS": request.app.state.config.ENABLE_USER_WEBHOOKS,
+        "PENDING_USER_OVERLAY_TITLE": request.app.state.config.PENDING_USER_OVERLAY_TITLE,
+        "PENDING_USER_OVERLAY_CONTENT": request.app.state.config.PENDING_USER_OVERLAY_CONTENT,
+        "RESPONSE_WATERMARK": request.app.state.config.RESPONSE_WATERMARK,
     }
 
 
@@ -779,6 +921,7 @@ class LdapServerConfig(BaseModel):
     search_filters: str = ""
     use_tls: bool = True
     certificate_path: Optional[str] = None
+    validate_cert: bool = True
     ciphers: Optional[str] = "ALL"
 
 
@@ -796,6 +939,7 @@ async def get_ldap_server(request: Request, user=Depends(get_admin_user)):
         "search_filters": request.app.state.config.LDAP_SEARCH_FILTERS,
         "use_tls": request.app.state.config.LDAP_USE_TLS,
         "certificate_path": request.app.state.config.LDAP_CA_CERT_FILE,
+        "validate_cert": request.app.state.config.LDAP_VALIDATE_CERT,
         "ciphers": request.app.state.config.LDAP_CIPHERS,
     }
 
@@ -831,6 +975,7 @@ async def update_ldap_server(
     request.app.state.config.LDAP_SEARCH_FILTERS = form_data.search_filters
     request.app.state.config.LDAP_USE_TLS = form_data.use_tls
     request.app.state.config.LDAP_CA_CERT_FILE = form_data.certificate_path
+    request.app.state.config.LDAP_VALIDATE_CERT = form_data.validate_cert
     request.app.state.config.LDAP_CIPHERS = form_data.ciphers
 
     return {
@@ -845,6 +990,7 @@ async def update_ldap_server(
         "search_filters": request.app.state.config.LDAP_SEARCH_FILTERS,
         "use_tls": request.app.state.config.LDAP_USE_TLS,
         "certificate_path": request.app.state.config.LDAP_CA_CERT_FILE,
+        "validate_cert": request.app.state.config.LDAP_VALIDATE_CERT,
         "ciphers": request.app.state.config.LDAP_CIPHERS,
     }
 

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)
 
 
 ############################

backend/open_webui/routers/chats.py

@@ -76,17 +76,34 @@ async def delete_all_user_chats(request: Request, user=Depends(get_verified_user
 @router.get("/list/user/{user_id}", response_model=list[ChatTitleIdResponse])
 async def get_user_chat_list_by_user_id(
     user_id: str,
+    page: Optional[int] = None,
+    query: Optional[str] = None,
+    order_by: Optional[str] = None,
+    direction: Optional[str] = None,
     user=Depends(get_admin_user),
-    skip: int = 0,
-    limit: int = 50,
 ):
     if not ENABLE_ADMIN_CHAT_ACCESS:
         raise HTTPException(
             status_code=status.HTTP_401_UNAUTHORIZED,
             detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
         )
+
+    if page is None:
+        page = 1
+
+    limit = 60
+    skip = (page - 1) * limit
+
+    filter = {}
+    if query:
+        filter["query"] = query
+    if order_by:
+        filter["order_by"] = order_by
+    if direction:
+        filter["direction"] = direction
+
     return Chats.get_chat_list_by_user_id(
-        user_id, include_archived=True, skip=skip, limit=limit
+        user_id, include_archived=True, filter=filter, skip=skip, limit=limit
     )
 
 
@@ -194,10 +211,10 @@ async def get_chats_by_folder_id(folder_id: str, user=Depends(get_verified_user)
 ############################
 
 
-@router.get("/pinned", response_model=list[ChatResponse])
+@router.get("/pinned", response_model=list[ChatTitleIdResponse])
 async def get_user_pinned_chats(user=Depends(get_verified_user)):
     return [
-        ChatResponse(**chat.model_dump())
+        ChatTitleIdResponse(**chat.model_dump())
         for chat in Chats.get_pinned_chats_by_user_id(user.id)
     ]
 
@@ -267,9 +284,37 @@ async def get_all_user_chats_in_db(user=Depends(get_admin_user)):
 
 @router.get("/archived", response_model=list[ChatTitleIdResponse])
 async def get_archived_session_user_chat_list(
-    user=Depends(get_verified_user), skip: int = 0, limit: int = 50
+    page: Optional[int] = None,
+    query: Optional[str] = None,
+    order_by: Optional[str] = None,
+    direction: Optional[str] = None,
+    user=Depends(get_verified_user),
 ):
-    return Chats.get_archived_chat_list_by_user_id(user.id, skip, limit)
+    if page is None:
+        page = 1
+
+    limit = 60
+    skip = (page - 1) * limit
+
+    filter = {}
+    if query:
+        filter["query"] = query
+    if order_by:
+        filter["order_by"] = order_by
+    if direction:
+        filter["direction"] = direction
+
+    chat_list = [
+        ChatTitleIdResponse(**chat.model_dump())
+        for chat in Chats.get_archived_chat_list_by_user_id(
+            user.id,
+            filter=filter,
+            skip=skip,
+            limit=limit,
+        )
+    ]
+
+    return chat_list
 
 
 ############################
@@ -639,8 +684,10 @@ 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

@@ -39,32 +39,39 @@ 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,
     }
 
 

backend/open_webui/routers/evaluations.py

@@ -74,13 +74,17 @@ class FeedbackUserResponse(FeedbackResponse):
 @router.get("/feedbacks/all", response_model=list[FeedbackUserResponse])
 async def get_all_feedbacks(user=Depends(get_admin_user)):
     feedbacks = Feedbacks.get_all_feedbacks()
-    return [
-        FeedbackUserResponse(
-            **feedback.model_dump(),
-            user=UserResponse(**Users.get_user_by_id(feedback.user_id).model_dump()),
+
+    feedback_list = []
+    for feedback in feedbacks:
+        user = Users.get_user_by_id(feedback.user_id)
+        feedback_list.append(
+            FeedbackUserResponse(
+                **feedback.model_dump(),
+                user=UserResponse(**user.model_dump()) if user else None,
+            )
         )
-        for feedback in feedbacks
-    ]
+    return feedback_list
 
 
 @router.delete("/feedbacks/all")
@@ -92,12 +96,7 @@ async def delete_all_feedbacks(user=Depends(get_admin_user)):
 @router.get("/feedbacks/all/export", response_model=list[FeedbackModel])
 async def get_all_feedbacks(user=Depends(get_admin_user)):
     feedbacks = Feedbacks.get_all_feedbacks()
-    return [
-        FeedbackModel(
-            **feedback.model_dump(), user=Users.get_user_by_id(feedback.user_id)
-        )
-        for feedback in feedbacks
-    ]
+    return feedbacks
 
 
 @router.get("/feedbacks/user", response_model=list[FeedbackUserResponse])

backend/open_webui/routers/files.py

@@ -1,6 +1,7 @@
 import logging
 import os
 import uuid
+import json
 from fnmatch import fnmatch
 from pathlib import Path
 from typing import Optional
@@ -10,6 +11,7 @@ from fastapi import (
     APIRouter,
     Depends,
     File,
+    Form,
     HTTPException,
     Request,
     UploadFile,
@@ -84,17 +86,44 @@ def has_access_to_file(
 def upload_file(
     request: Request,
     file: UploadFile = File(...),
-    user=Depends(get_verified_user),
-    file_metadata: dict = None,
+    metadata: Optional[dict | str] = Form(None),
     process: bool = Query(True),
+    internal: bool = False,
+    user=Depends(get_verified_user),
 ):
     log.info(f"file.content_type: {file.content_type}")
 
-    file_metadata = file_metadata if file_metadata else {}
+    if isinstance(metadata, str):
+        try:
+            metadata = json.loads(metadata)
+        except json.JSONDecodeError:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail=ERROR_MESSAGES.DEFAULT("Invalid metadata format"),
+            )
+    file_metadata = metadata if metadata else {}
+
     try:
         unsanitized_filename = file.filename
         filename = os.path.basename(unsanitized_filename)
 
+        file_extension = os.path.splitext(filename)[1]
+        # Remove the leading dot from the file extension
+        file_extension = file_extension[1:] if file_extension else ""
+
+        if (not internal) and request.app.state.config.ALLOWED_FILE_EXTENSIONS:
+            request.app.state.config.ALLOWED_FILE_EXTENSIONS = [
+                ext for ext in request.app.state.config.ALLOWED_FILE_EXTENSIONS if ext
+            ]
+
+            if file_extension not in request.app.state.config.ALLOWED_FILE_EXTENSIONS:
+                raise HTTPException(
+                    status_code=status.HTTP_400_BAD_REQUEST,
+                    detail=ERROR_MESSAGES.DEFAULT(
+                        f"File type {file_extension} is not allowed"
+                    ),
+                )
+
         # replace filename with uuid
         id = str(uuid.uuid4())
         name = filename
@@ -125,33 +154,36 @@ def upload_file(
         )
         if process:
             try:
-
-                if file.content_type.startswith(
-                    (
-                        "audio/mpeg",
-                        "audio/wav",
-                        "audio/ogg",
-                        "audio/x-m4a",
-                        "audio/webm",
+                if file.content_type:
+                    stt_supported_content_types = getattr(
+                        request.app.state.config, "STT_SUPPORTED_CONTENT_TYPES", []
                     )
-                ):
-                    file_path = Storage.get_file(file_path)
-                    result = transcribe(request, file_path)
 
-                    process_file(
-                        request,
-                        ProcessFileForm(file_id=id, content=result.get("text", "")),
-                        user=user,
+                    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)
+
+                        process_file(
+                            request,
+                            ProcessFileForm(file_id=id, content=result.get("text", "")),
+                            user=user,
+                        )
+                    elif (not file.content_type.startswith(("image/", "video/"))) or (
+                        request.app.state.config.CONTENT_EXTRACTION_ENGINE == "external"
+                    ):
+                        process_file(request, ProcessFileForm(file_id=id), user=user)
+                else:
+                    log.info(
+                        f"File type {file.content_type} is not provided, but trying to process anyway"
                     )
-                elif file.content_type not in [
-                    "image/png",
-                    "image/jpeg",
-                    "image/gif",
-                    "video/mp4",
-                    "video/ogg",
-                    "video/quicktime",
-                    "video/webm",
-                ]:
                     process_file(request, ProcessFileForm(file_id=id), user=user)
 
                 file_item = Files.get_file_by_id(id=id)
@@ -177,7 +209,7 @@ def upload_file(
         log.exception(e)
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,
-            detail=ERROR_MESSAGES.DEFAULT(e),
+            detail=ERROR_MESSAGES.DEFAULT("Error uploading file"),
         )
 
 

backend/open_webui/routers/folders.py

@@ -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:

backend/open_webui/routers/functions.py

@@ -1,5 +1,8 @@
 import os
+import re
+
 import logging
+import aiohttp
 from pathlib import Path
 from typing import Optional
 
@@ -9,12 +12,18 @@ from open_webui.models.functions import (
     FunctionResponse,
     Functions,
 )
-from open_webui.utils.plugin import load_function_module_by_id, replace_imports
+from open_webui.utils.plugin import (
+    load_function_module_by_id,
+    replace_imports,
+    get_function_module_from_cache,
+)
 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.auth import get_admin_user, get_verified_user
 from open_webui.env import SRC_LOG_LEVELS
+from pydantic import BaseModel, HttpUrl
+
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
@@ -42,6 +51,97 @@ async def get_functions(user=Depends(get_admin_user)):
     return Functions.get_functions()
 
 
+############################
+# LoadFunctionFromLink
+############################
+
+
+class LoadUrlForm(BaseModel):
+    url: HttpUrl
+
+
+def github_url_to_raw_url(url: str) -> str:
+    # Handle 'tree' (folder) URLs (add main.py at the end)
+    m1 = re.match(r"https://github\.com/([^/]+)/([^/]+)/tree/([^/]+)/(.*)", url)
+    if m1:
+        org, repo, branch, path = m1.groups()
+        return f"https://raw.githubusercontent.com/{org}/{repo}/refs/heads/{branch}/{path.rstrip('/')}/main.py"
+
+    # Handle 'blob' (file) URLs
+    m2 = re.match(r"https://github\.com/([^/]+)/([^/]+)/blob/([^/]+)/(.*)", url)
+    if m2:
+        org, repo, branch, path = m2.groups()
+        return (
+            f"https://raw.githubusercontent.com/{org}/{repo}/refs/heads/{branch}/{path}"
+        )
+
+    # No match; return as-is
+    return url
+
+
+@router.post("/load/url", response_model=Optional[dict])
+async def load_function_from_url(
+    request: Request, form_data: LoadUrlForm, user=Depends(get_admin_user)
+):
+    # NOTE: This is NOT a SSRF vulnerability:
+    # This endpoint is admin-only (see get_admin_user), meant for *trusted* internal use,
+    # and does NOT accept untrusted user input. Access is enforced by authentication.
+
+    url = str(form_data.url)
+    if not url:
+        raise HTTPException(status_code=400, detail="Please enter a valid URL")
+
+    url = github_url_to_raw_url(url)
+    url_parts = url.rstrip("/").split("/")
+
+    file_name = url_parts[-1]
+    function_name = (
+        file_name[:-3]
+        if (
+            file_name.endswith(".py")
+            and (not file_name.startswith(("main.py", "index.py", "__init__.py")))
+        )
+        else url_parts[-2] if len(url_parts) > 1 else "function"
+    )
+
+    try:
+        async with aiohttp.ClientSession(trust_env=True) as session:
+            async with session.get(
+                url, headers={"Content-Type": "application/json"}
+            ) as resp:
+                if resp.status != 200:
+                    raise HTTPException(
+                        status_code=resp.status, detail="Failed to fetch the function"
+                    )
+                data = await resp.text()
+                if not data:
+                    raise HTTPException(
+                        status_code=400, detail="No data received from the URL"
+                    )
+        return {
+            "name": function_name,
+            "content": data,
+        }
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Error importing function: {e}")
+
+
+############################
+# SyncFunctions
+############################
+
+
+class SyncFunctionsForm(FunctionForm):
+    functions: list[FunctionModel] = []
+
+
+@router.post("/sync", response_model=Optional[FunctionModel])
+async def sync_functions(
+    request: Request, form_data: SyncFunctionsForm, user=Depends(get_admin_user)
+):
+    return Functions.sync_functions(user.id, form_data.functions)
+
+
 ############################
 # CreateNewFunction
 ############################
@@ -262,11 +362,9 @@ async def get_function_valves_spec_by_id(
 ):
     function = Functions.get_function_by_id(id)
     if function:
-        if id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[id]
-        else:
-            function_module, function_type, frontmatter = load_function_module_by_id(id)
-            request.app.state.FUNCTIONS[id] = function_module
+        function_module, function_type, frontmatter = get_function_module_from_cache(
+            request, id
+        )
 
         if hasattr(function_module, "Valves"):
             Valves = function_module.Valves
@@ -290,11 +388,9 @@ async def update_function_valves_by_id(
 ):
     function = Functions.get_function_by_id(id)
     if function:
-        if id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[id]
-        else:
-            function_module, function_type, frontmatter = load_function_module_by_id(id)
-            request.app.state.FUNCTIONS[id] = function_module
+        function_module, function_type, frontmatter = get_function_module_from_cache(
+            request, id
+        )
 
         if hasattr(function_module, "Valves"):
             Valves = function_module.Valves
@@ -353,11 +449,9 @@ async def get_function_user_valves_spec_by_id(
 ):
     function = Functions.get_function_by_id(id)
     if function:
-        if id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[id]
-        else:
-            function_module, function_type, frontmatter = load_function_module_by_id(id)
-            request.app.state.FUNCTIONS[id] = function_module
+        function_module, function_type, frontmatter = get_function_module_from_cache(
+            request, id
+        )
 
         if hasattr(function_module, "UserValves"):
             UserValves = function_module.UserValves
@@ -377,11 +471,9 @@ async def update_function_user_valves_by_id(
     function = Functions.get_function_by_id(id)
 
     if function:
-        if id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[id]
-        else:
-            function_module, function_type, frontmatter = load_function_module_by_id(id)
-            request.app.state.FUNCTIONS[id] = function_module
+        function_module, function_type, frontmatter = get_function_module_from_cache(
+            request, id
+        )
 
         if hasattr(function_module, "UserValves"):
             UserValves = function_module.UserValves

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(
@@ -333,10 +342,11 @@ def get_models(request: Request, user=Depends(get_verified_user)):
             return [
                 {"id": "dall-e-2", "name": "DALL·E 2"},
                 {"id": "dall-e-3", "name": "DALL·E 3"},
+                {"id": "gpt-image-1", "name": "GPT-IMAGE 1"},
             ]
         elif request.app.state.config.IMAGE_GENERATION_ENGINE == "gemini":
             return [
-                {"id": "imagen-3-0-generate-002", "name": "imagen-3.0 generate-002"},
+                {"id": "imagen-3.0-generate-002", "name": "imagen-3.0 generate-002"},
             ]
         elif request.app.state.config.IMAGE_GENERATION_ENGINE == "comfyui":
             # TODO - get models from comfyui
@@ -419,7 +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"
@@ -427,7 +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):
@@ -450,7 +460,7 @@ def load_url_image_data(url, headers=None):
         return None
 
 
-def upload_image(request, image_metadata, image_data, content_type, user):
+def upload_image(request, image_data, content_type, metadata, user):
     image_format = mimetypes.guess_extension(content_type)
     file = UploadFile(
         file=io.BytesIO(image_data),
@@ -459,7 +469,7 @@ def upload_image(request, image_metadata, image_data, content_type, user):
             "content-type": content_type,
         },
     )
-    file_item = upload_file(request, file, user, file_metadata=image_metadata)
+    file_item = upload_file(request, file, metadata=metadata, internal=True, user=user)
     url = request.app.url_path_for("get_file_content_by_id", id=file_item.id)
     return url
 
@@ -470,7 +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:
@@ -482,7 +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
@@ -526,7 +543,7 @@ async def image_generations(
                 else:
                     image_data, content_type = load_b64_image_data(image["b64_json"])
 
-                url = upload_image(request, data, image_data, content_type, user)
+                url = upload_image(request, image_data, content_type, data, user)
                 images.append({"url": url})
             return images
 
@@ -560,7 +577,7 @@ async def image_generations(
                 image_data, content_type = load_b64_image_data(
                     image["bytesBase64Encoded"]
                 )
-                url = upload_image(request, data, image_data, content_type, user)
+                url = upload_image(request, image_data, content_type, data, user)
                 images.append({"url": url})
 
             return images
@@ -611,9 +628,9 @@ async def image_generations(
                 image_data, content_type = load_url_image_data(image["url"], headers)
                 url = upload_image(
                     request,
-                    form_data.model_dump(exclude_none=True),
                     image_data,
                     content_type,
+                    form_data.model_dump(exclude_none=True),
                     user,
                 )
                 images.append({"url": url})
@@ -664,9 +681,9 @@ async def image_generations(
                 image_data, content_type = load_b64_image_data(image)
                 url = upload_image(
                     request,
-                    {**data, "info": res["info"]},
                     image_data,
                     content_type,
+                    {**data, "info": res["info"]},
                     user,
                 )
                 images.append({"url": url})

backend/open_webui/routers/knowledge.py

@@ -10,7 +10,7 @@ from open_webui.models.knowledge import (
     KnowledgeUserResponse,
 )
 from open_webui.models.files import Files, FileModel, FileMetadataResponse
-from open_webui.retrieval.vector.connector import VECTOR_DB_CLIENT
+from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT
 from open_webui.routers.retrieval import (
     process_file,
     ProcessFileForm,

backend/open_webui/routers/memories.py

@@ -4,7 +4,7 @@ import logging
 from typing import Optional
 
 from open_webui.models.memories import Memories, MemoryModel
-from open_webui.retrieval.vector.connector import VECTOR_DB_CLIENT
+from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT
 from open_webui.utils.auth import get_verified_user
 from open_webui.env import SRC_LOG_LEVELS
 

backend/open_webui/routers/notes.py

@@ -51,7 +51,14 @@ 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 +70,8 @@ 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 +126,9 @@ 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 +160,8 @@ 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(
@@ -199,9 +199,8 @@ 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

@@ -9,11 +9,16 @@ import os
 import random
 import re
 import time
+from datetime import datetime
+
 from typing import Optional, Union
 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 (
@@ -54,6 +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,
@@ -83,7 +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,
@@ -134,7 +140,7 @@ 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,
@@ -145,8 +151,23 @@ 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)
 
@@ -166,20 +187,14 @@ async def send_post_request(
             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",
         )
 
 
@@ -229,7 +244,7 @@ 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,
@@ -300,7 +315,23 @@ async def update_config(
     }
 
 
-@cached(ttl=1)
+def merge_ollama_models_lists(model_lists):
+    merged_models = {}
+
+    for idx, model_list in enumerate(model_lists):
+        if model_list is not None:
+            for model in model_list:
+                id = model["model"]
+                if id not in merged_models:
+                    model["urls"] = [idx]
+                    merged_models[id] = model
+                else:
+                    merged_models[id]["urls"].append(idx)
+
+    return list(merged_models.values())
+
+
+@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:
@@ -340,6 +371,8 @@ async def get_all_models(request: Request, user: UserModel = None):
                     ),  # Legacy support
                 )
 
+                connection_type = api_config.get("connection_type", "local")
+
                 prefix_id = api_config.get("prefix_id", None)
                 tags = api_config.get("tags", [])
                 model_ids = api_config.get("model_ids", [])
@@ -352,31 +385,18 @@ async def get_all_models(request: Request, user: UserModel = None):
                         )
                     )
 
-                if prefix_id:
-                    for model in response.get("models", []):
+                for model in response.get("models", []):
+                    if prefix_id:
                         model["model"] = f"{prefix_id}.{model['model']}"
 
-                if tags:
-                    for model in response.get("models", []):
+                    if tags:
                         model["tags"] = tags
 
-        def merge_models_lists(model_lists):
-            merged_models = {}
-
-            for idx, model_list in enumerate(model_lists):
-                if model_list is not None:
-                    for model in model_list:
-                        id = model["model"]
-                        if id not in merged_models:
-                            model["urls"] = [idx]
-                            merged_models[id] = model
-                        else:
-                            merged_models[id]["urls"].append(idx)
-
-            return list(merged_models.values())
+                    if connection_type:
+                        model["connection_type"] = connection_type
 
         models = {
-            "models": merge_models_lists(
+            "models": merge_ollama_models_lists(
                 map(
                     lambda response: response.get("models", []) if response else None,
                     responses,
@@ -384,6 +404,22 @@ async def get_all_models(request: Request, user: UserModel = None):
             )
         }
 
+        try:
+            loaded_models = await get_ollama_loaded_models(request, user=user)
+            expires_map = {
+                m["name"]: m["expires_at"]
+                for m in loaded_models["models"]
+                if "expires_at" in m
+            }
+
+            for m in models["models"]:
+                if m["name"] in expires_map:
+                    # Parse ISO8601 datetime with offset, get unix timestamp as int
+                    dt = datetime.fromisoformat(expires_map[m["name"]])
+                    m["expires_at"] = int(dt.timestamp())
+        except Exception as e:
+            log.debug(f"Failed to get loaded models: {e}")
+
     else:
         models = {"models": []}
 
@@ -428,7 +464,7 @@ 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,
@@ -464,6 +500,68 @@ async def get_ollama_tags(
     return models
 
 
+@router.get("/api/ps")
+async def get_ollama_loaded_models(request: Request, user=Depends(get_admin_user)):
+    """
+    List models that are currently loaded into Ollama memory, and which node they are loaded on.
+    """
+    if request.app.state.config.ENABLE_OLLAMA_API:
+        request_tasks = []
+        for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS):
+            if (str(idx) not in request.app.state.config.OLLAMA_API_CONFIGS) and (
+                url not in request.app.state.config.OLLAMA_API_CONFIGS  # Legacy support
+            ):
+                request_tasks.append(send_get_request(f"{url}/api/ps", user=user))
+            else:
+                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
+                    str(idx),
+                    request.app.state.config.OLLAMA_API_CONFIGS.get(
+                        url, {}
+                    ),  # Legacy support
+                )
+
+                enable = api_config.get("enable", True)
+                key = api_config.get("key", None)
+
+                if enable:
+                    request_tasks.append(
+                        send_get_request(f"{url}/api/ps", key, user=user)
+                    )
+                else:
+                    request_tasks.append(asyncio.ensure_future(asyncio.sleep(0, None)))
+
+        responses = await asyncio.gather(*request_tasks)
+
+        for idx, response in enumerate(responses):
+            if response:
+                url = request.app.state.config.OLLAMA_BASE_URLS[idx]
+                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
+                    str(idx),
+                    request.app.state.config.OLLAMA_API_CONFIGS.get(
+                        url, {}
+                    ),  # Legacy support
+                )
+
+                prefix_id = api_config.get("prefix_id", None)
+
+                for model in response.get("models", []):
+                    if prefix_id:
+                        model["model"] = f"{prefix_id}.{model['model']}"
+
+        models = {
+            "models": merge_ollama_models_lists(
+                map(
+                    lambda response: response.get("models", []) if response else None,
+                    responses,
+                )
+            )
+        }
+    else:
+        models = {"models": []}
+
+    return models
+
+
 @router.get("/api/version")
 @router.get("/api/version/{url_idx}")
 async def get_ollama_versions(request: Request, url_idx: Optional[int] = None):
@@ -537,34 +635,77 @@ async def get_ollama_versions(request: Request, url_idx: Optional[int] = None):
         return {"version": False}
 
 
-@router.get("/api/ps")
-async def get_ollama_loaded_models(request: Request, user=Depends(get_verified_user)):
-    """
-    List models that are currently loaded into Ollama memory, and which node they are loaded on.
-    """
-    if request.app.state.config.ENABLE_OLLAMA_API:
-        request_tasks = [
-            send_get_request(
-                f"{url}/api/ps",
-                request.app.state.config.OLLAMA_API_CONFIGS.get(
-                    str(idx),
-                    request.app.state.config.OLLAMA_API_CONFIGS.get(
-                        url, {}
-                    ),  # Legacy support
-                ).get("key", None),
+class ModelNameForm(BaseModel):
+    model: Optional[str] = None
+    model_config = ConfigDict(
+        extra="allow",
+    )
+
+
+@router.post("/api/unload")
+async def unload_model(
+    request: Request,
+    form_data: ModelNameForm,
+    user=Depends(get_admin_user),
+):
+    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 the model to unload."
+        )
+
+    # Refresh/load models if needed, get mapping from name to URLs
+    await get_all_models(request, user=user)
+    models = request.app.state.OLLAMA_MODELS
+
+    # Canonicalize model name (if not supplied with version)
+    if ":" not in model_name:
+        model_name = f"{model_name}:latest"
+
+    if model_name not in models:
+        raise HTTPException(
+            status_code=400, detail=ERROR_MESSAGES.MODEL_NOT_FOUND(model_name)
+        )
+    url_indices = models[model_name]["urls"]
+
+    # Send unload to ALL url_indices
+    results = []
+    errors = []
+    for idx in url_indices:
+        url = request.app.state.config.OLLAMA_BASE_URLS[idx]
+        api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
+            str(idx), request.app.state.config.OLLAMA_API_CONFIGS.get(url, {})
+        )
+        key = get_api_key(idx, url, request.app.state.config.OLLAMA_API_CONFIGS)
+
+        prefix_id = api_config.get("prefix_id", None)
+        if prefix_id and model_name.startswith(f"{prefix_id}."):
+            model_name = model_name[len(f"{prefix_id}.") :]
+
+        payload = {"model": model_name, "keep_alive": 0, "prompt": ""}
+
+        try:
+            res = await send_post_request(
+                url=f"{url}/api/generate",
+                payload=json.dumps(payload),
+                stream=False,
+                key=key,
                 user=user,
             )
-            for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS)
-        ]
-        responses = await asyncio.gather(*request_tasks)
+            results.append({"url_idx": idx, "success": True, "response": res})
+        except Exception as e:
+            log.exception(f"Failed to unload model on node {idx}: {e}")
+            errors.append({"url_idx": idx, "success": False, "error": str(e)})
 
-        return dict(zip(request.app.state.config.OLLAMA_BASE_URLS, responses))
-    else:
-        return {}
+    if len(errors) > 0:
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to unload model on {len(errors)} nodes: {errors}",
+        )
 
-
-class ModelNameForm(BaseModel):
-    name: str
+    return {"status": True}
 
 
 @router.post("/api/pull")
@@ -575,11 +716,14 @@ 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",
@@ -590,7 +734,7 @@ async def pull_model(
 
 
 class PushModelForm(BaseModel):
-    name: str
+    model: str
     insecure: Optional[bool] = None
     stream: Optional[bool] = None
 
@@ -607,12 +751,12 @@ 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]
@@ -690,7 +834,7 @@ 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,
@@ -731,16 +875,21 @@ 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]
@@ -750,13 +899,13 @@ 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,
@@ -792,16 +941,21 @@ 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)
@@ -815,7 +969,7 @@ 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,
@@ -824,7 +978,7 @@ 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()
 
@@ -902,7 +1056,7 @@ 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,
@@ -989,7 +1143,7 @@ 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,
@@ -1109,6 +1263,9 @@ 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):
@@ -1146,7 +1303,9 @@ 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"]
 
@@ -1160,13 +1319,10 @@ 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":
@@ -1199,7 +1355,7 @@ 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),
@@ -1348,8 +1504,10 @@ 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":
@@ -1585,7 +1743,9 @@ async def upload_model(
     if url_idx is None:
         url_idx = 0
     ollama_url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
-    file_path = os.path.join(UPLOAD_DIR, file.filename)
+
+    filename = os.path.basename(file.filename)
+    file_path = os.path.join(UPLOAD_DIR, filename)
     os.makedirs(UPLOAD_DIR, exist_ok=True)
 
     # --- P1: save file locally ---
@@ -1630,13 +1790,13 @@ async def upload_model(
                 os.remove(file_path)
 
                 # Create model in ollama
-                model_name, ext = os.path.splitext(file.filename)
+                model_name, ext = os.path.splitext(filename)
                 log.info(f"Created Model: {model_name}")  # DEBUG
 
                 create_payload = {
                     "model": model_name,
                     # Reference the file by its original name => the uploaded blob's digest
-                    "files": {file.filename: f"sha256:{file_hash}"},
+                    "files": {filename: f"sha256:{file_hash}"},
                 }
                 log.info(f"Model Payload: {create_payload}")  # DEBUG
 
@@ -1653,7 +1813,7 @@ async def upload_model(
                     done_msg = {
                         "done": True,
                         "blob": f"sha256:{file_hash}",
-                        "name": file.filename,
+                        "name": filename,
                         "model_created": model_name,
                     }
                     yield f"data: {json.dumps(done_msg)}\n\n"

backend/open_webui/routers/openai.py

@@ -8,7 +8,7 @@ from typing import Literal, Optional, overload
 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
@@ -21,6 +21,7 @@ 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,
@@ -66,7 +67,7 @@ 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,
@@ -225,7 +226,7 @@ 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,
@@ -353,21 +354,22 @@ async def get_all_models_responses(request: Request, user: UserModel) -> list:
                 ),  # Legacy support
             )
 
+            connection_type = api_config.get("connection_type", "external")
             prefix_id = api_config.get("prefix_id", None)
             tags = api_config.get("tags", [])
 
-            if prefix_id:
-                for model in (
-                    response if isinstance(response, list) else response.get("data", [])
-                ):
+            for model in (
+                response if isinstance(response, list) else response.get("data", [])
+            ):
+                if prefix_id:
                     model["id"] = f"{prefix_id}.{model['id']}"
 
-            if tags:
-                for model in (
-                    response if isinstance(response, list) else response.get("data", [])
-                ):
+                if tags:
                     model["tags"] = tags
 
+                if connection_type:
+                    model["connection_type"] = connection_type
+
     log.debug(f"get_all_models:responses() {responses}")
     return responses
 
@@ -385,7 +387,7 @@ 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()")
 
@@ -415,6 +417,7 @@ async def get_all_models(request: Request, user: UserModel) -> dict[str, list]:
                             "name": model.get("name", model["id"]),
                             "owned_by": "openai",
                             "openai": model,
+                            "connection_type": model.get("connection_type", "external"),
                             "urlIdx": idx,
                         }
                         for model in models
@@ -461,60 +464,74 @@ async def get_models(
         url = request.app.state.config.OPENAI_API_BASE_URLS[url_idx]
         key = request.app.state.config.OPENAI_API_KEYS[url_idx]
 
+        api_config = request.app.state.config.OPENAI_API_CONFIGS.get(
+            str(url_idx),
+            request.app.state.config.OPENAI_API_CONFIGS.get(url, {}),  # Legacy support
+        )
+
         r = None
         async with aiohttp.ClientSession(
             trust_env=True,
             timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST),
         ) as session:
             try:
-                async with session.get(
-                    f"{url}/models",
-                    headers={
-                        "Authorization": f"Bearer {key}",
-                        "Content-Type": "application/json",
-                        **(
-                            {
-                                "X-OpenWebUI-User-Name": user.name,
-                                "X-OpenWebUI-User-Id": user.id,
-                                "X-OpenWebUI-User-Email": user.email,
-                                "X-OpenWebUI-User-Role": user.role,
-                            }
-                            if ENABLE_FORWARD_USER_INFO_HEADERS
-                            else {}
-                        ),
-                    },
-                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
-                ) as r:
-                    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)
+                headers = {
+                    "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
+                        else {}
+                    ),
+                }
 
-                    response_data = await r.json()
+                if api_config.get("azure", False):
+                    models = {
+                        "data": api_config.get("model_ids", []) or [],
+                        "object": "list",
+                    }
+                else:
+                    headers["Authorization"] = f"Bearer {key}"
 
-                    # Check if we're calling OpenAI API based on the URL
-                    if "api.openai.com" in url:
-                        # Filter models according to the specified conditions
-                        response_data["data"] = [
-                            model
-                            for model in response_data.get("data", [])
-                            if not any(
-                                name in model["id"]
-                                for name in [
-                                    "babbage",
-                                    "dall-e",
-                                    "davinci",
-                                    "embedding",
-                                    "tts",
-                                    "whisper",
-                                ]
-                            )
-                        ]
+                    async with session.get(
+                        f"{url}/models",
+                        headers=headers,
+                        ssl=AIOHTTP_CLIENT_SESSION_SSL,
+                    ) as r:
+                        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)
 
-                    models = response_data
+                        response_data = await r.json()
+
+                        # Check if we're calling OpenAI API based on the URL
+                        if "api.openai.com" in url:
+                            # Filter models according to the specified conditions
+                            response_data["data"] = [
+                                model
+                                for model in response_data.get("data", [])
+                                if not any(
+                                    name in model["id"]
+                                    for name in [
+                                        "babbage",
+                                        "dall-e",
+                                        "davinci",
+                                        "embedding",
+                                        "tts",
+                                        "whisper",
+                                    ]
+                                )
+                            ]
+
+                        models = response_data
             except aiohttp.ClientError as e:
                 # ClientError covers all aiohttp requests issues
                 log.exception(f"Client error: {str(e)}")
@@ -536,6 +553,8 @@ class ConnectionVerificationForm(BaseModel):
     url: str
     key: str
 
+    config: Optional[dict] = None
+
 
 @router.post("/verify")
 async def verify_connection(
@@ -544,39 +563,64 @@ async def verify_connection(
     url = form_data.url
     key = form_data.key
 
+    api_config = form_data.config or {}
+
     async with aiohttp.ClientSession(
         trust_env=True,
         timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST),
     ) as session:
         try:
-            async with session.get(
-                f"{url}/models",
-                headers={
-                    "Authorization": f"Bearer {key}",
-                    "Content-Type": "application/json",
-                    **(
-                        {
-                            "X-OpenWebUI-User-Name": user.name,
-                            "X-OpenWebUI-User-Id": user.id,
-                            "X-OpenWebUI-User-Email": user.email,
-                            "X-OpenWebUI-User-Role": user.role,
-                        }
-                        if ENABLE_FORWARD_USER_INFO_HEADERS
-                        else {}
-                    ),
-                },
-                ssl=AIOHTTP_CLIENT_SESSION_SSL,
-            ) as r:
-                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)
+            headers = {
+                "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
+                    else {}
+                ),
+            }
 
-                response_data = await r.json()
-                return response_data
+            if api_config.get("azure", False):
+                headers["api-key"] = key
+                api_version = api_config.get("api_version", "") or "2023-03-15-preview"
+
+                async with session.get(
+                    url=f"{url}/openai/models?api-version={api_version}",
+                    headers=headers,
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
+                ) as r:
+                    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)
+
+                    response_data = await r.json()
+                    return response_data
+            else:
+                headers["Authorization"] = f"Bearer {key}"
+
+                async with session.get(
+                    f"{url}/models",
+                    headers=headers,
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
+                ) as r:
+                    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)
+
+                    response_data = await r.json()
+                    return response_data
 
         except aiohttp.ClientError as e:
             # ClientError covers all aiohttp requests issues
@@ -590,6 +634,74 @@ async def verify_connection(
             raise HTTPException(status_code=500, detail=error_detail)
 
 
+def get_azure_allowed_params(api_version: str) -> set[str]:
+    allowed_params = {
+        "messages",
+        "temperature",
+        "role",
+        "content",
+        "contentPart",
+        "contentPartImage",
+        "enhancements",
+        "dataSources",
+        "n",
+        "stream",
+        "stop",
+        "max_tokens",
+        "presence_penalty",
+        "frequency_penalty",
+        "logit_bias",
+        "user",
+        "function_call",
+        "functions",
+        "tools",
+        "tool_choice",
+        "top_p",
+        "log_probs",
+        "top_logprobs",
+        "response_format",
+        "seed",
+        "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
+        if "max_tokens" in payload:
+            payload["max_completion_tokens"] = payload["max_tokens"]
+            del payload["max_tokens"]
+
+        # Remove temperature if not 1 for o-series models
+        if "temperature" in payload and payload["temperature"] != 1:
+            log.debug(
+                f"Removing temperature parameter for o-series model {model} as only default value (1) is supported"
+            )
+            del payload["temperature"]
+
+    # Filter out unsupported parameters
+    payload = {k: v for k, v in payload.items() if k in allowed_params}
+
+    url = f"{url}/openai/deployments/{model}"
+    return url, payload
+
+
 @router.post("/chat/completions")
 async def generate_chat_completion(
     request: Request,
@@ -615,8 +727,12 @@ 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":
@@ -690,6 +806,38 @@ async def generate_chat_completion(
             convert_logit_bias_input_to_json(payload["logit_bias"])
         )
 
+    headers = {
+        "Content-Type": "application/json",
+        **(
+            {
+                "HTTP-Referer": "https://openwebui.com/",
+                "X-Title": "Open WebUI",
+            }
+            if "openrouter.ai" in url
+            else {}
+        ),
+        **(
+            {
+                "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
+            else {}
+        ),
+    }
+
+    if api_config.get("azure", False):
+        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}"
+    else:
+        request_url = f"{url}/chat/completions"
+        headers["Authorization"] = f"Bearer {key}"
+
     payload = json.dumps(payload)
 
     r = None
@@ -704,30 +852,9 @@ async def generate_chat_completion(
 
         r = await session.request(
             method="POST",
-            url=f"{url}/chat/completions",
+            url=request_url,
             data=payload,
-            headers={
-                "Authorization": f"Bearer {key}",
-                "Content-Type": "application/json",
-                **(
-                    {
-                        "HTTP-Referer": "https://openwebui.com/",
-                        "X-Title": "Open WebUI",
-                    }
-                    if "openrouter.ai" in url
-                    else {}
-                ),
-                **(
-                    {
-                        "X-OpenWebUI-User-Name": user.name,
-                        "X-OpenWebUI-User-Id": user.id,
-                        "X-OpenWebUI-User-Email": user.email,
-                        "X-OpenWebUI-User-Role": user.role,
-                    }
-                    if ENABLE_FORWARD_USER_INFO_HEADERS
-                    else {}
-                ),
-            },
+            headers=headers,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
         )
 
@@ -772,6 +899,88 @@ async def generate_chat_completion(
             await session.close()
 
 
+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 {}
+                ),
+            },
+        )
+        r.raise_for_status()
+        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:
+            response_data = await r.json()
+            return response_data
+    except Exception as e:
+        log.exception(e)
+        detail = None
+        if r is not None:
+            try:
+                res = await r.json()
+                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",
+        )
+    finally:
+        if not streaming and session:
+            if r:
+                r.close()
+            await session.close()
+
+
 @router.api_route("/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])
 async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
     """
@@ -783,31 +992,52 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
     idx = 0
     url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
     key = request.app.state.config.OPENAI_API_KEYS[idx]
+    api_config = request.app.state.config.OPENAI_API_CONFIGS.get(
+        str(idx),
+        request.app.state.config.OPENAI_API_CONFIGS.get(
+            request.app.state.config.OPENAI_API_BASE_URLS[idx], {}
+        ),  # Legacy support
+    )
 
     r = None
     session = None
     streaming = False
 
     try:
+        headers = {
+            "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
+                else {}
+            ),
+        }
+
+        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_version
+
+            payload = json.loads(body)
+            url, payload = convert_to_azure_payload(url, payload, api_version)
+            body = json.dumps(payload).encode()
+
+            request_url = f"{url}/{path}?api-version={api_version}"
+        else:
+            headers["Authorization"] = f"Bearer {key}"
+            request_url = f"{url}/{path}"
+
         session = aiohttp.ClientSession(trust_env=True)
         r = await session.request(
             method=request.method,
-            url=f"{url}/{path}",
+            url=request_url,
             data=body,
-            headers={
-                "Authorization": f"Bearer {key}",
-                "Content-Type": "application/json",
-                **(
-                    {
-                        "X-OpenWebUI-User-Name": user.name,
-                        "X-OpenWebUI-User-Id": user.id,
-                        "X-OpenWebUI-User-Email": user.email,
-                        "X-OpenWebUI-User-Role": user.role,
-                    }
-                    if ENABLE_FORWARD_USER_INFO_HEADERS
-                    else {}
-                ),
-            },
+            headers=headers,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
         )
         r.raise_for_status()

backend/open_webui/routers/pipelines.py

@@ -18,7 +18,7 @@ from pydantic import BaseModel
 from starlette.responses import FileResponse
 from typing import Optional
 
-from open_webui.env import SRC_LOG_LEVELS
+from open_webui.env import SRC_LOG_LEVELS, AIOHTTP_CLIENT_SESSION_SSL
 from open_webui.config import CACHE_DIR
 from open_webui.constants import ERROR_MESSAGES
 
@@ -69,7 +69,10 @@ async def process_pipeline_inlet_filter(request, payload, user, models):
     async with aiohttp.ClientSession(trust_env=True) as session:
         for filter in sorted_filters:
             urlIdx = filter.get("urlIdx")
-            if urlIdx is None:
+
+            try:
+                urlIdx = int(urlIdx)
+            except:
                 continue
 
             url = request.app.state.config.OPENAI_API_BASE_URLS[urlIdx]
@@ -89,6 +92,7 @@ async def process_pipeline_inlet_filter(request, payload, user, models):
                     f"{url}/{filter['id']}/filter/inlet",
                     headers=headers,
                     json=request_data,
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as response:
                     payload = await response.json()
                     response.raise_for_status()
@@ -118,7 +122,10 @@ async def process_pipeline_outlet_filter(request, payload, user, models):
     async with aiohttp.ClientSession(trust_env=True) as session:
         for filter in sorted_filters:
             urlIdx = filter.get("urlIdx")
-            if urlIdx is None:
+
+            try:
+                urlIdx = int(urlIdx)
+            except:
                 continue
 
             url = request.app.state.config.OPENAI_API_BASE_URLS[urlIdx]
@@ -138,6 +145,7 @@ async def process_pipeline_outlet_filter(request, payload, user, models):
                     f"{url}/{filter['id']}/filter/outlet",
                     headers=headers,
                     json=request_data,
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as response:
                     payload = await response.json()
                     response.raise_for_status()
@@ -197,8 +205,10 @@ async def upload_pipeline(
     user=Depends(get_admin_user),
 ):
     log.info(f"upload_pipeline: urlIdx={urlIdx}, filename={file.filename}")
+    filename = os.path.basename(file.filename)
+
     # Check if the uploaded file is a python file
-    if not (file.filename and file.filename.endswith(".py")):
+    if not (filename and filename.endswith(".py")):
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,
             detail="Only Python (.py) files are allowed.",
@@ -206,7 +216,7 @@ async def upload_pipeline(
 
     upload_folder = f"{CACHE_DIR}/pipelines"
     os.makedirs(upload_folder, exist_ok=True)
-    file_path = os.path.join(upload_folder, file.filename)
+    file_path = os.path.join(upload_folder, filename)
 
     r = None
     try:

backend/open_webui/routers/retrieval.py

@@ -3,6 +3,8 @@ import logging
 import mimetypes
 import os
 import shutil
+import asyncio
+
 
 import uuid
 from datetime import datetime
@@ -27,6 +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
@@ -34,7 +37,7 @@ from open_webui.models.knowledge import Knowledges
 from open_webui.storage.provider import Storage
 
 
-from open_webui.retrieval.vector.connector import VECTOR_DB_CLIENT
+from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT
 
 # Document loaders
 from open_webui.retrieval.loaders.main import Loader
@@ -67,6 +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,
@@ -135,7 +139,10 @@ def get_ef(
 
 
 def get_rf(
+    engine: str = "",
     reranking_model: Optional[str] = None,
+    external_reranker_url: str = "",
+    external_reranker_api_key: str = "",
     auto_update: bool = False,
 ):
     rf = None
@@ -153,19 +160,33 @@ def get_rf(
                 log.error(f"ColBERT: {e}")
                 raise Exception(ERROR_MESSAGES.DEFAULT(e))
         else:
-            import sentence_transformers
+            if engine == "external":
+                try:
+                    from open_webui.retrieval.models.external import ExternalReranker
+
+                    rf = ExternalReranker(
+                        url=external_reranker_url,
+                        api_key=external_reranker_api_key,
+                        model=reranking_model,
+                    )
+                except Exception as e:
+                    log.error(f"ExternalReranking: {e}")
+                    raise Exception(ERROR_MESSAGES.DEFAULT(e))
+            else:
+                import sentence_transformers
+
+                try:
+                    rf = sentence_transformers.CrossEncoder(
+                        get_model_path(reranking_model, auto_update),
+                        device=DEVICE_TYPE,
+                        trust_remote_code=RAG_RERANKING_MODEL_TRUST_REMOTE_CODE,
+                        backend=SENTENCE_TRANSFORMERS_CROSS_ENCODER_BACKEND,
+                        model_kwargs=SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS,
+                    )
+                except Exception as e:
+                    log.error(f"CrossEncoder: {e}")
+                    raise Exception(ERROR_MESSAGES.DEFAULT("CrossEncoder error"))
 
-            try:
-                rf = sentence_transformers.CrossEncoder(
-                    get_model_path(reranking_model, auto_update),
-                    device=DEVICE_TYPE,
-                    trust_remote_code=RAG_RERANKING_MODEL_TRUST_REMOTE_CODE,
-                    backend=SENTENCE_TRANSFORMERS_CROSS_ENCODER_BACKEND,
-                    model_kwargs=SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS,
-                )
-            except Exception as e:
-                log.error(f"CrossEncoder: {e}")
-                raise Exception(ERROR_MESSAGES.DEFAULT("CrossEncoder error"))
     return rf
 
 
@@ -188,7 +209,7 @@ class ProcessUrlForm(CollectionNameForm):
 
 
 class SearchForm(BaseModel):
-    query: str
+    queries: List[str]
 
 
 @router.get("/")
@@ -220,14 +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,
         },
-    }
-
-
-@router.get("/reranking")
-async def get_reraanking_config(request: Request, user=Depends(get_admin_user)):
-    return {
-        "status": True,
-        "reranking_model": request.app.state.config.RAG_RERANKING_MODEL,
+        "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,
+        },
     }
 
 
@@ -241,9 +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
@@ -260,7 +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
@@ -277,6 +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
             )
@@ -293,14 +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 {
@@ -316,6 +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}")
@@ -325,41 +383,6 @@ async def update_embedding_config(
         )
 
 
-class RerankingModelUpdateForm(BaseModel):
-    reranking_model: str
-
-
-@router.post("/reranking/update")
-async def update_reranking_config(
-    request: Request, form_data: RerankingModelUpdateForm, user=Depends(get_admin_user)
-):
-    log.info(
-        f"Updating reranking model: {request.app.state.config.RAG_RERANKING_MODEL} to {form_data.reranking_model}"
-    )
-    try:
-        request.app.state.config.RAG_RERANKING_MODEL = form_data.reranking_model
-
-        try:
-            request.app.state.rf = get_rf(
-                request.app.state.config.RAG_RERANKING_MODEL,
-                True,
-            )
-        except Exception as e:
-            log.error(f"Error loading reranking model: {e}")
-            request.app.state.config.ENABLE_RAG_HYBRID_SEARCH = False
-
-        return {
-            "status": True,
-            "reranking_model": request.app.state.config.RAG_RERANKING_MODEL,
-        }
-    except Exception as e:
-        log.exception(f"Problem updating reranking model: {e}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail=ERROR_MESSAGES.DEFAULT(e),
-        )
-
-
 @router.get("/config")
 async def get_rag_config(request: Request, user=Depends(get_admin_user)):
     return {
@@ -373,16 +396,37 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         "ENABLE_RAG_HYBRID_SEARCH": request.app.state.config.ENABLE_RAG_HYBRID_SEARCH,
         "TOP_K_RERANKER": request.app.state.config.TOP_K_RERANKER,
         "RELEVANCE_THRESHOLD": request.app.state.config.RELEVANCE_THRESHOLD,
+        "HYBRID_BM25_WEIGHT": request.app.state.config.HYBRID_BM25_WEIGHT,
         # Content extraction settings
         "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_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_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_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,
+        # Reranking settings
+        "RAG_RERANKING_MODEL": request.app.state.config.RAG_RERANKING_MODEL,
+        "RAG_RERANKING_ENGINE": request.app.state.config.RAG_RERANKING_ENGINE,
+        "RAG_EXTERNAL_RERANKER_URL": request.app.state.config.RAG_EXTERNAL_RERANKER_URL,
+        "RAG_EXTERNAL_RERANKER_API_KEY": request.app.state.config.RAG_EXTERNAL_RERANKER_API_KEY,
         # Chunking settings
         "TEXT_SPLITTER": request.app.state.config.TEXT_SPLITTER,
         "CHUNK_SIZE": request.app.state.config.CHUNK_SIZE,
@@ -390,6 +434,9 @@ 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,
         "ENABLE_ONEDRIVE_INTEGRATION": request.app.state.config.ENABLE_ONEDRIVE_INTEGRATION,
@@ -402,6 +449,7 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
             "WEB_SEARCH_CONCURRENT_REQUESTS": request.app.state.config.WEB_SEARCH_CONCURRENT_REQUESTS,
             "WEB_SEARCH_DOMAIN_FILTER_LIST": request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
             "BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL": request.app.state.config.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL,
+            "BYPASS_WEB_SEARCH_WEB_LOADER": request.app.state.config.BYPASS_WEB_SEARCH_WEB_LOADER,
             "SEARXNG_QUERY_URL": request.app.state.config.SEARXNG_QUERY_URL,
             "YACY_QUERY_URL": request.app.state.config.YACY_QUERY_URL,
             "YACY_USERNAME": request.app.state.config.YACY_USERNAME,
@@ -426,6 +474,8 @@ 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,
@@ -454,6 +504,7 @@ class WebConfig(BaseModel):
     WEB_SEARCH_CONCURRENT_REQUESTS: Optional[int] = None
     WEB_SEARCH_DOMAIN_FILTER_LIST: Optional[List[str]] = []
     BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL: Optional[bool] = None
+    BYPASS_WEB_SEARCH_WEB_LOADER: Optional[bool] = None
     SEARXNG_QUERY_URL: Optional[str] = None
     YACY_QUERY_URL: Optional[str] = None
     YACY_USERNAME: Optional[str] = None
@@ -478,6 +529,8 @@ 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
@@ -507,18 +560,41 @@ class ConfigForm(BaseModel):
     ENABLE_RAG_HYBRID_SEARCH: Optional[bool] = None
     TOP_K_RERANKER: Optional[int] = None
     RELEVANCE_THRESHOLD: Optional[float] = None
+    HYBRID_BM25_WEIGHT: Optional[float] = None
 
     # Content extraction settings
     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_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_USE_LLM: Optional[bool] = None
+    DATALAB_MARKER_OUTPUT_FORMAT: Optional[str] = None
+    EXTERNAL_DOCUMENT_LOADER_URL: Optional[str] = None
+    EXTERNAL_DOCUMENT_LOADER_API_KEY: Optional[str] = None
+
     TIKA_SERVER_URL: Optional[str] = None
     DOCLING_SERVER_URL: Optional[str] = None
     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
 
+    # Reranking settings
+    RAG_RERANKING_MODEL: Optional[str] = None
+    RAG_RERANKING_ENGINE: Optional[str] = None
+    RAG_EXTERNAL_RERANKER_URL: Optional[str] = None
+    RAG_EXTERNAL_RERANKER_API_KEY: Optional[str] = None
+
     # Chunking settings
     TEXT_SPLITTER: Optional[str] = None
     CHUNK_SIZE: Optional[int] = None
@@ -527,6 +603,9 @@ 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
     ENABLE_GOOGLE_DRIVE_INTEGRATION: Optional[bool] = None
@@ -582,6 +661,11 @@ async def update_rag_config(
         if form_data.RELEVANCE_THRESHOLD is not None
         else request.app.state.config.RELEVANCE_THRESHOLD
     )
+    request.app.state.config.HYBRID_BM25_WEIGHT = (
+        form_data.HYBRID_BM25_WEIGHT
+        if form_data.HYBRID_BM25_WEIGHT is not None
+        else request.app.state.config.HYBRID_BM25_WEIGHT
+    )
 
     # Content extraction settings
     request.app.state.config.CONTENT_EXTRACTION_ENGINE = (
@@ -594,6 +678,61 @@ async def update_rag_config(
         if form_data.PDF_EXTRACT_IMAGES is not None
         else request.app.state.config.PDF_EXTRACT_IMAGES
     )
+    request.app.state.config.DATALAB_MARKER_API_KEY = (
+        form_data.DATALAB_MARKER_API_KEY
+        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_SKIP_CACHE = (
+        form_data.DATALAB_MARKER_SKIP_CACHE
+        if form_data.DATALAB_MARKER_SKIP_CACHE is not None
+        else request.app.state.config.DATALAB_MARKER_SKIP_CACHE
+    )
+    request.app.state.config.DATALAB_MARKER_FORCE_OCR = (
+        form_data.DATALAB_MARKER_FORCE_OCR
+        if form_data.DATALAB_MARKER_FORCE_OCR is not None
+        else request.app.state.config.DATALAB_MARKER_FORCE_OCR
+    )
+    request.app.state.config.DATALAB_MARKER_PAGINATE = (
+        form_data.DATALAB_MARKER_PAGINATE
+        if form_data.DATALAB_MARKER_PAGINATE is not None
+        else request.app.state.config.DATALAB_MARKER_PAGINATE
+    )
+    request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR = (
+        form_data.DATALAB_MARKER_STRIP_EXISTING_OCR
+        if form_data.DATALAB_MARKER_STRIP_EXISTING_OCR is not None
+        else request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR
+    )
+    request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = (
+        form_data.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
+        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_OUTPUT_FORMAT = (
+        form_data.DATALAB_MARKER_OUTPUT_FORMAT
+        if form_data.DATALAB_MARKER_OUTPUT_FORMAT is not None
+        else request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT
+    )
+    request.app.state.config.DATALAB_MARKER_USE_LLM = (
+        form_data.DATALAB_MARKER_USE_LLM
+        if form_data.DATALAB_MARKER_USE_LLM is not None
+        else request.app.state.config.DATALAB_MARKER_USE_LLM
+    )
+    request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL = (
+        form_data.EXTERNAL_DOCUMENT_LOADER_URL
+        if form_data.EXTERNAL_DOCUMENT_LOADER_URL is not None
+        else request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL
+    )
+    request.app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY = (
+        form_data.EXTERNAL_DOCUMENT_LOADER_API_KEY
+        if form_data.EXTERNAL_DOCUMENT_LOADER_API_KEY is not None
+        else request.app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY
+    )
     request.app.state.config.TIKA_SERVER_URL = (
         form_data.TIKA_SERVER_URL
         if form_data.TIKA_SERVER_URL is not None
@@ -614,6 +753,29 @@ async def update_rag_config(
         if form_data.DOCLING_OCR_LANG is not None
         else request.app.state.config.DOCLING_OCR_LANG
     )
+
+    request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION = (
+        form_data.DOCLING_DO_PICTURE_DESCRIPTION
+        if form_data.DOCLING_DO_PICTURE_DESCRIPTION is not None
+        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
@@ -630,6 +792,55 @@ async def update_rag_config(
         else request.app.state.config.MISTRAL_OCR_API_KEY
     )
 
+    # Reranking settings
+    request.app.state.config.RAG_RERANKING_ENGINE = (
+        form_data.RAG_RERANKING_ENGINE
+        if form_data.RAG_RERANKING_ENGINE is not None
+        else request.app.state.config.RAG_RERANKING_ENGINE
+    )
+
+    request.app.state.config.RAG_EXTERNAL_RERANKER_URL = (
+        form_data.RAG_EXTERNAL_RERANKER_URL
+        if form_data.RAG_EXTERNAL_RERANKER_URL is not None
+        else request.app.state.config.RAG_EXTERNAL_RERANKER_URL
+    )
+
+    request.app.state.config.RAG_EXTERNAL_RERANKER_API_KEY = (
+        form_data.RAG_EXTERNAL_RERANKER_API_KEY
+        if form_data.RAG_EXTERNAL_RERANKER_API_KEY is not None
+        else request.app.state.config.RAG_EXTERNAL_RERANKER_API_KEY
+    )
+
+    log.info(
+        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
+
+        try:
+            request.app.state.rf = get_rf(
+                request.app.state.config.RAG_RERANKING_ENGINE,
+                request.app.state.config.RAG_RERANKING_MODEL,
+                request.app.state.config.RAG_EXTERNAL_RERANKER_URL,
+                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
+    except Exception as e:
+        log.exception(f"Problem updating reranking model: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=ERROR_MESSAGES.DEFAULT(e),
+        )
+
     # Chunking settings
     request.app.state.config.TEXT_SPLITTER = (
         form_data.TEXT_SPLITTER
@@ -648,15 +859,18 @@ 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
+        if form_data.ALLOWED_FILE_EXTENSIONS is not None
+        else request.app.state.config.ALLOWED_FILE_EXTENSIONS
     )
 
     # Integration settings
@@ -690,6 +904,9 @@ async def update_rag_config(
         request.app.state.config.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL = (
             form_data.web.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL
         )
+        request.app.state.config.BYPASS_WEB_SEARCH_WEB_LOADER = (
+            form_data.web.BYPASS_WEB_SEARCH_WEB_LOADER
+        )
         request.app.state.config.SEARXNG_QUERY_URL = form_data.web.SEARXNG_QUERY_URL
         request.app.state.config.YACY_QUERY_URL = form_data.web.YACY_QUERY_URL
         request.app.state.config.YACY_USERNAME = form_data.web.YACY_USERNAME
@@ -726,6 +943,10 @@ 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
 
@@ -776,16 +997,37 @@ async def update_rag_config(
         "ENABLE_RAG_HYBRID_SEARCH": request.app.state.config.ENABLE_RAG_HYBRID_SEARCH,
         "TOP_K_RERANKER": request.app.state.config.TOP_K_RERANKER,
         "RELEVANCE_THRESHOLD": request.app.state.config.RELEVANCE_THRESHOLD,
+        "HYBRID_BM25_WEIGHT": request.app.state.config.HYBRID_BM25_WEIGHT,
         # Content extraction settings
         "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_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_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_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,
+        # Reranking settings
+        "RAG_RERANKING_MODEL": request.app.state.config.RAG_RERANKING_MODEL,
+        "RAG_RERANKING_ENGINE": request.app.state.config.RAG_RERANKING_ENGINE,
+        "RAG_EXTERNAL_RERANKER_URL": request.app.state.config.RAG_EXTERNAL_RERANKER_URL,
+        "RAG_EXTERNAL_RERANKER_API_KEY": request.app.state.config.RAG_EXTERNAL_RERANKER_API_KEY,
         # Chunking settings
         "TEXT_SPLITTER": request.app.state.config.TEXT_SPLITTER,
         "CHUNK_SIZE": request.app.state.config.CHUNK_SIZE,
@@ -793,6 +1035,9 @@ 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,
         "ENABLE_ONEDRIVE_INTEGRATION": request.app.state.config.ENABLE_ONEDRIVE_INTEGRATION,
@@ -805,6 +1050,7 @@ async def update_rag_config(
             "WEB_SEARCH_CONCURRENT_REQUESTS": request.app.state.config.WEB_SEARCH_CONCURRENT_REQUESTS,
             "WEB_SEARCH_DOMAIN_FILTER_LIST": request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
             "BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL": request.app.state.config.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL,
+            "BYPASS_WEB_SEARCH_WEB_LOADER": request.app.state.config.BYPASS_WEB_SEARCH_WEB_LOADER,
             "SEARXNG_QUERY_URL": request.app.state.config.SEARXNG_QUERY_URL,
             "YACY_QUERY_URL": request.app.state.config.YACY_QUERY_URL,
             "YACY_USERNAME": request.app.state.config.YACY_USERNAME,
@@ -829,6 +1075,8 @@ 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,
@@ -906,6 +1154,7 @@ 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}"
@@ -918,11 +1167,56 @@ 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)
 
@@ -973,14 +1267,27 @@ 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(
@@ -1094,10 +1401,27 @@ def process_file(
                 file_path = Storage.get_file(file_path)
                 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_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_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_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,
@@ -1477,6 +1801,16 @@ 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(
@@ -1527,6 +1861,8 @@ 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 (
@@ -1568,16 +1904,34 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
 async def process_web_search(
     request: Request, form_data: SearchForm, user=Depends(get_verified_user)
 ):
+
+    urls = []
     try:
         logging.info(
-            f"trying to web search with {request.app.state.config.WEB_SEARCH_ENGINE, form_data.query}"
-        )
-        web_results = await run_in_threadpool(
-            search_web,
-            request,
-            request.app.state.config.WEB_SEARCH_ENGINE,
-            form_data.query,
+            f"trying to web search with {request.app.state.config.WEB_SEARCH_ENGINE, form_data.queries}"
         )
+
+        search_tasks = [
+            run_in_threadpool(
+                search_web,
+                request,
+                request.app.state.config.WEB_SEARCH_ENGINE,
+                query,
+            )
+            for query in form_data.queries
+        ]
+
+        search_results = await asyncio.gather(*search_tasks)
+
+        for result in search_results:
+            if result:
+                for item in result:
+                    if item and item.link:
+                        urls.append(item.link)
+
+        urls = list(dict.fromkeys(urls))
+        log.debug(f"urls: {urls}")
+
     except Exception as e:
         log.exception(e)
 
@@ -1586,20 +1940,37 @@ async def process_web_search(
             detail=ERROR_MESSAGES.WEB_SEARCH_ERROR(e),
         )
 
-    log.debug(f"web_results: {web_results}")
-
     try:
-        urls = [result.link for result in web_results]
-        loader = get_web_loader(
-            urls,
-            verify_ssl=request.app.state.config.ENABLE_WEB_LOADER_SSL_VERIFICATION,
-            requests_per_second=request.app.state.config.WEB_SEARCH_CONCURRENT_REQUESTS,
-            trust_env=request.app.state.config.WEB_SEARCH_TRUST_ENV,
-        )
-        docs = await loader.aload()
+        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,
+                    metadata={
+                        "source": result.link,
+                        "title": result.title,
+                        "snippet": result.snippet,
+                        "link": result.link,
+                    },
+                )
+                for result in search_results
+                if hasattr(result, "snippet")
+            ]
+        else:
+            loader = get_web_loader(
+                urls,
+                verify_ssl=request.app.state.config.ENABLE_WEB_LOADER_SSL_VERIFICATION,
+                requests_per_second=request.app.state.config.WEB_SEARCH_CONCURRENT_REQUESTS,
+                trust_env=request.app.state.config.WEB_SEARCH_TRUST_ENV,
+            )
+            docs = await loader.aload()
+
         urls = [
             doc.metadata.get("source") for doc in docs if doc.metadata.get("source")
-        ]  # only keep URLs
+        ]  # only keep the urls returned by the loader
 
         if request.app.state.config.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL:
             return {
@@ -1616,29 +1987,28 @@ async def process_web_search(
                 "loaded_count": len(docs),
             }
         else:
-            collection_names = []
-            for doc_idx, doc in enumerate(docs):
-                if doc and doc.page_content:
-                    try:
-                        collection_name = f"web-search-{calculate_sha256_string(form_data.query + '-' + urls[doc_idx])}"[
-                            :63
-                        ]
+            # Create a single collection for all documents
+            collection_name = (
+                f"web-search-{calculate_sha256_string('-'.join(form_data.queries))}"[
+                    :63
+                ]
+            )
 
-                        collection_names.append(collection_name)
-                        await run_in_threadpool(
-                            save_docs_to_vector_db,
-                            request,
-                            [doc],
-                            collection_name,
-                            overwrite=True,
-                            user=user,
-                        )
-                    except Exception as e:
-                        log.debug(f"error saving doc {doc_idx}: {e}")
+            try:
+                await run_in_threadpool(
+                    save_docs_to_vector_db,
+                    request,
+                    docs,
+                    collection_name,
+                    overwrite=True,
+                    user=user,
+                )
+            except Exception as e:
+                log.debug(f"error saving docs: {e}")
 
             return {
                 "status": True,
-                "collection_names": collection_names,
+                "collection_names": [collection_name],
                 "filenames": urls,
                 "loaded_count": len(docs),
             }
@@ -1679,7 +2049,13 @@ 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=(
@@ -1687,6 +2063,11 @@ def query_doc_handler(
                     if form_data.r
                     else request.app.state.config.RELEVANCE_THRESHOLD
                 ),
+                hybrid_bm25_weight=(
+                    form_data.hybrid_bm25_weight
+                    if form_data.hybrid_bm25_weight
+                    else request.app.state.config.HYBRID_BM25_WEIGHT
+                ),
                 user=user,
             )
         else:
@@ -1713,6 +2094,7 @@ class QueryCollectionsForm(BaseModel):
     k_reranker: Optional[int] = None
     r: Optional[float] = None
     hybrid: Optional[bool] = None
+    hybrid_bm25_weight: Optional[float] = None
 
 
 @router.post("/query/collection")
@@ -1730,7 +2112,9 @@ 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
+                ),
                 k_reranker=form_data.k_reranker
                 or request.app.state.config.TOP_K_RERANKER,
                 r=(
@@ -1738,6 +2122,11 @@ def query_collection_handler(
                     if form_data.r
                     else request.app.state.config.RELEVANCE_THRESHOLD
                 ),
+                hybrid_bm25_weight=(
+                    form_data.hybrid_bm25_weight
+                    if form_data.hybrid_bm25_weight
+                    else request.app.state.config.HYBRID_BM25_WEIGHT
+                ),
             )
         else:
             return query_collection(

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,
@@ -20,14 +21,12 @@ from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.constants import TASKS
 
 from open_webui.routers.pipelines import process_pipeline_inlet_filter
-from open_webui.utils.filter import (
-    get_sorted_filter_ids,
-    process_filter_functions,
-)
+
 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,
@@ -61,6 +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,
@@ -79,6 +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
@@ -97,6 +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
     )
@@ -136,6 +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,
@@ -186,20 +198,100 @@ async def generate_title(
     else:
         template = DEFAULT_TITLE_GENERATION_PROMPT_TEMPLATE
 
-    messages = form_data["messages"]
-
-    # Remove reasoning details from the messages
-    for message in messages:
-        message["content"] = re.sub(
-            r"<details\s+type=\"reasoning\"[^>]*>.*?<\/details>",
-            "",
-            message["content"],
-            flags=re.S,
-        ).strip()
-
     content = title_generation_template(
         template,
-        messages,
+        form_data["messages"],
+        {
+            "name": user.name,
+            "location": user.info.get("location") if user.info else None,
+        },
+    )
+
+    max_tokens = (
+        models[task_model_id].get("info", {}).get("params", {}).get("max_tokens", 1000)
+    )
+
+    payload = {
+        "model": task_model_id,
+        "messages": [{"role": "user", "content": content}],
+        "stream": False,
+        **(
+            {"max_tokens": max_tokens}
+            if models[task_model_id].get("owned_by") == "ollama"
+            else {
+                "max_completion_tokens": max_tokens,
+            }
+        ),
+        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
+            "task": str(TASKS.TITLE_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("/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,
@@ -210,16 +302,9 @@ async def generate_title(
         "model": task_model_id,
         "messages": [{"role": "user", "content": content}],
         "stream": False,
-        **(
-            {"max_tokens": 1000}
-            if models[task_model_id].get("owned_by") == "ollama"
-            else {
-                "max_completion_tokens": 1000,
-            }
-        ),
         "metadata": {
             **(request.state.metadata if hasattr(request.state, "metadata") else {}),
-            "task": str(TASKS.TITLE_GENERATION),
+            "task": str(TASKS.FOLLOW_UP_GENERATION),
             "task_body": form_data,
             "chat_id": form_data.get("chat_id", None),
         },

backend/open_webui/routers/tools.py

@@ -2,6 +2,9 @@ import logging
 from pathlib import Path
 from typing import Optional
 import time
+import re
+import aiohttp
+from pydantic import BaseModel, HttpUrl
 
 from open_webui.models.tools import (
     ToolForm,
@@ -21,6 +24,7 @@ from open_webui.env import SRC_LOG_LEVELS
 
 from open_webui.utils.tools import get_tool_servers_data
 
+
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
@@ -51,11 +55,11 @@ async def get_tools(request: Request, user=Depends(get_verified_user)):
                 **{
                     "id": f"server:{server['idx']}",
                     "user_id": f"server:{server['idx']}",
-                    "name": server["openapi"]
+                    "name": server.get("openapi", {})
                     .get("info", {})
                     .get("title", "Tool Server"),
                     "meta": {
-                        "description": server["openapi"]
+                        "description": server.get("openapi", {})
                         .get("info", {})
                         .get("description", ""),
                     },
@@ -95,6 +99,81 @@ async def get_tool_list(user=Depends(get_verified_user)):
     return tools
 
 
+############################
+# LoadFunctionFromLink
+############################
+
+
+class LoadUrlForm(BaseModel):
+    url: HttpUrl
+
+
+def github_url_to_raw_url(url: str) -> str:
+    # Handle 'tree' (folder) URLs (add main.py at the end)
+    m1 = re.match(r"https://github\.com/([^/]+)/([^/]+)/tree/([^/]+)/(.*)", url)
+    if m1:
+        org, repo, branch, path = m1.groups()
+        return f"https://raw.githubusercontent.com/{org}/{repo}/refs/heads/{branch}/{path.rstrip('/')}/main.py"
+
+    # Handle 'blob' (file) URLs
+    m2 = re.match(r"https://github\.com/([^/]+)/([^/]+)/blob/([^/]+)/(.*)", url)
+    if m2:
+        org, repo, branch, path = m2.groups()
+        return (
+            f"https://raw.githubusercontent.com/{org}/{repo}/refs/heads/{branch}/{path}"
+        )
+
+    # No match; return as-is
+    return url
+
+
+@router.post("/load/url", response_model=Optional[dict])
+async def load_tool_from_url(
+    request: Request, form_data: LoadUrlForm, user=Depends(get_admin_user)
+):
+    # NOTE: This is NOT a SSRF vulnerability:
+    # This endpoint is admin-only (see get_admin_user), meant for *trusted* internal use,
+    # and does NOT accept untrusted user input. Access is enforced by authentication.
+
+    url = str(form_data.url)
+    if not url:
+        raise HTTPException(status_code=400, detail="Please enter a valid URL")
+
+    url = github_url_to_raw_url(url)
+    url_parts = url.rstrip("/").split("/")
+
+    file_name = url_parts[-1]
+    tool_name = (
+        file_name[:-3]
+        if (
+            file_name.endswith(".py")
+            and (not file_name.startswith(("main.py", "index.py", "__init__.py")))
+        )
+        else url_parts[-2] if len(url_parts) > 1 else "function"
+    )
+
+    try:
+        async with aiohttp.ClientSession(trust_env=True) as session:
+            async with session.get(
+                url, headers={"Content-Type": "application/json"}
+            ) as resp:
+                if resp.status != 200:
+                    raise HTTPException(
+                        status_code=resp.status, detail="Failed to fetch the tool"
+                    )
+                data = await resp.text()
+                if not data:
+                    raise HTTPException(
+                        status_code=400, detail="No data received from the URL"
+                    )
+        return {
+            "name": tool_name,
+            "content": data,
+        }
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Error importing tool: {e}")
+
+
 ############################
 # ExportTools
 ############################

backend/open_webui/routers/users.py

@@ -14,7 +14,11 @@ 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
@@ -29,12 +33,30 @@ 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
 ############################
 
 
-PAGE_ITEM_COUNT = 10
+PAGE_ITEM_COUNT = 30
 
 
 @router.get("/", response_model=UserListResponse)
@@ -111,6 +133,7 @@ class SharingPermissions(BaseModel):
 
 class ChatPermissions(BaseModel):
     controls: bool = True
+    system_prompt: bool = True
     file_upload: bool = True
     delete: bool = True
     edit: bool = True
@@ -165,22 +188,6 @@ 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 +326,18 @@ async def get_user_by_id(user_id: str, user=Depends(get_verified_user)):
         )
 
 
+############################
+# 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 +352,22 @@ 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 +395,7 @@ 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,
@@ -22,7 +27,11 @@ from open_webui.env import (
     WEBSOCKET_SENTINEL_HOSTS,
 )
 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 +44,8 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["SOCKET"])
 
 
+REDIS = None
+
 if WEBSOCKET_MANAGER == "redis":
     if WEBSOCKET_SENTINEL_HOSTS:
         mgr = socketio.AsyncRedisManager(
@@ -69,6 +80,14 @@ 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
+        ),
+        async_mode=True,
+    )
+
     redis_sentinels = get_sentinels_from_env(
         WEBSOCKET_SENTINEL_HOSTS, WEBSOCKET_SENTINEL_PORT
     )
@@ -101,14 +120,37 @@ else:
     SESSION_POOL = {}
     USER_POOL = {}
     USAGE_POOL = {}
+
     aquire_func = release_func = renew_func = lambda: True
 
 
+YDOC_MANAGER = YdocManager(
+    redis=REDIS,
+    redis_key_prefix="open-webui: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 +177,6 @@ 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()
@@ -157,20 +194,59 @@ 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):
-    model_id = data["model"]
-    # Record the timestamp for the last update
-    current_time = int(time.time())
+    if sid in SESSION_POOL:
+        model_id = data["model"]
+        # Record the timestamp for the last update
+        current_time = int(time.time())
 
-    # Store the new usage data and task
-    USAGE_POOL[model_id] = {
-        **(USAGE_POOL[model_id] if model_id in USAGE_POOL else {}),
-        sid: {"updated_at": current_time},
-    }
-
-    # Broadcast the usage data to all clients
-    await sio.emit("usage", {"models": get_models_in_use()})
+        # Store the new usage data and task
+        USAGE_POOL[model_id] = {
+            **(USAGE_POOL[model_id] if model_id in USAGE_POOL else {}),
+            sid: {"updated_at": current_time},
+        }
 
 
 @sio.event
@@ -189,13 +265,6 @@ 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()})
-            return True
-
-    return False
-
 
 @sio.on("user-join")
 async def user_join(sid, data):
@@ -223,10 +292,6 @@ 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}
 
 
@@ -279,9 +344,239 @@ async def channel_events(sid, data):
         )
 
 
-@sio.on("user-list")
-async def user_list(sid):
-    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)
+        print(active_session_ids)
+        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)
+            )
+
+        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 (
+            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
@@ -296,7 +591,7 @@ 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")
@@ -389,30 +684,3 @@ 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

@@ -1,6 +1,8 @@
 import json
 import uuid
 from open_webui.utils.redis import get_redis_connection
+from typing import Optional, List, Tuple
+import pycrdt as Y
 
 
 class RedisLock:
@@ -89,3 +91,109 @@ class RedisDict:
         if key not in self:
             self[key] = default
         return self[key]
+
+
+class YdocManager:
+    def __init__(
+        self,
+        redis=None,
+        redis_key_prefix: str = "open-webui: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/static/assets/pdf-style.css

@@ -269,11 +269,6 @@ tbody + tbody {
 	margin-bottom: 0;
 }
 
-/* Add a rule to reset margin-bottom for <p> not followed by <ul> */
-.markdown-section p + ul {
-	margin-top: 0;
-}
-
 /* List item styles */
 .markdown-section li {
 	padding: 2px;

backend/open_webui/storage/provider.py

@@ -2,6 +2,7 @@ import os
 import shutil
 import json
 import logging
+import re
 from abc import ABC, abstractmethod
 from typing import BinaryIO, Tuple, Dict
 
@@ -17,6 +18,7 @@ from open_webui.config import (
     S3_SECRET_ACCESS_KEY,
     S3_USE_ACCELERATE_ENDPOINT,
     S3_ADDRESSING_STYLE,
+    S3_ENABLE_TAGGING,
     GCS_BUCKET_NAME,
     GOOGLE_APPLICATION_CREDENTIALS_JSON,
     AZURE_STORAGE_ENDPOINT,
@@ -135,23 +137,37 @@ class S3StorageProvider(StorageProvider):
         self.bucket_name = S3_BUCKET_NAME
         self.key_prefix = S3_KEY_PREFIX if S3_KEY_PREFIX else ""
 
+    @staticmethod
+    def sanitize_tag_value(s: str) -> str:
+        """Only include S3 allowed characters."""
+        return re.sub(r"[^a-zA-Z0-9 äöüÄÖÜß\+\-=\._:/@]", "", s)
+
     def upload_file(
         self, file: BinaryIO, filename: str, tags: Dict[str, str]
     ) -> Tuple[bytes, str]:
         """Handles uploading of the file to S3 storage."""
         _, file_path = LocalStorageProvider.upload_file(file, filename, tags)
-        tagging = {"TagSet": [{"Key": k, "Value": v} for k, v in tags.items()]}
+        s3_key = os.path.join(self.key_prefix, filename)
         try:
-            s3_key = os.path.join(self.key_prefix, filename)
             self.s3_client.upload_file(file_path, self.bucket_name, s3_key)
-            self.s3_client.put_object_tagging(
-                Bucket=self.bucket_name,
-                Key=s3_key,
-                Tagging=tagging,
-            )
+            if S3_ENABLE_TAGGING and tags:
+                sanitized_tags = {
+                    self.sanitize_tag_value(k): self.sanitize_tag_value(v)
+                    for k, v in tags.items()
+                }
+                tagging = {
+                    "TagSet": [
+                        {"Key": k, "Value": v} for k, v in sanitized_tags.items()
+                    ]
+                }
+                self.s3_client.put_object_tagging(
+                    Bucket=self.bucket_name,
+                    Key=s3_key,
+                    Tagging=tagging,
+                )
             return (
                 open(file_path, "rb").read(),
-                "s3://" + self.bucket_name + "/" + s3_key,
+                f"s3://{self.bucket_name}/{s3_key}",
             )
         except ClientError as e:
             raise RuntimeError(f"Error uploading file to S3: {e}")

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
+
+
+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 = "open-webui:tasks"
+REDIS_ITEM_TASKS_KEY = "open-webui:tasks:item"
+REDIS_PUBSUB_CHANNEL = "open-webui: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/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

@@ -13,6 +13,8 @@ import pytz
 from pytz import UTC
 from typing import Optional, Union, List, Dict
 
+from opentelemetry import trace
+
 from open_webui.models.users import Users
 
 from open_webui.constants import ERROR_MESSAGES
@@ -21,6 +23,7 @@ from open_webui.env import (
     TRUSTED_SIGNATURE_KEY,
     STATIC_DIR,
     SRC_LOG_LEVELS,
+    WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
 )
 
 from fastapi import BackgroundTasks, Depends, HTTPException, Request, Response, status
@@ -71,31 +74,37 @@ 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 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: {})()
+            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 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}")
     return False
@@ -155,6 +164,7 @@ 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),
 ):
@@ -194,7 +204,17 @@ def get_current_user(
                     status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.API_KEY_NOT_ALLOWED
                 )
 
-        return get_current_user_by_api_key(token)
+        user = get_current_user_by_api_key(token)
+
+        # Add user info to current span
+        current_span = trace.get_current_span()
+        if current_span:
+            current_span.set_attribute("client.user.id", user.id)
+            current_span.set_attribute("client.user.email", user.email)
+            current_span.set_attribute("client.user.role", user.role)
+            current_span.set_attribute("client.auth.type", "api_key")
+
+        return user
 
     # auth by jwt token
     try:
@@ -213,6 +233,29 @@ 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:
+                current_span.set_attribute("client.user.id", user.id)
+                current_span.set_attribute("client.user.email", user.email)
+                current_span.set_attribute("client.user.role", user.role)
+                current_span.set_attribute("client.auth.type", "jwt")
+
             # Refresh the user's last active timestamp asynchronously
             # to prevent blocking the request
             if background_tasks:
@@ -234,6 +277,14 @@ def get_current_user_by_api_key(api_key: str):
             detail=ERROR_MESSAGES.INVALID_TOKEN,
         )
     else:
+        # Add user info to current span
+        current_span = trace.get_current_span()
+        if current_span:
+            current_span.set_attribute("client.user.id", user.id)
+            current_span.set_attribute("client.user.email", user.email)
+            current_span.set_attribute("client.user.role", user.role)
+            current_span.set_attribute("client.auth.type", "api_key")
+
         Users.update_user_last_active_by_id(user.id)
 
     return user

backend/open_webui/utils/chat.py

@@ -40,7 +40,10 @@ from open_webui.models.functions import Functions
 from open_webui.models.models import Models
 
 
-from open_webui.utils.plugin import load_function_module_by_id
+from open_webui.utils.plugin import (
+    load_function_module_by_id,
+    get_function_module_from_cache,
+)
 from open_webui.utils.models import get_all_models, check_model_access
 from open_webui.utils.payload import convert_payload_openai_to_ollama
 from open_webui.utils.response import (
@@ -309,6 +312,7 @@ async def chat_completed(request: Request, form_data: dict, user: Any):
     metadata = {
         "chat_id": data["chat_id"],
         "message_id": data["id"],
+        "filter_ids": data.get("filter_ids", []),
         "session_id": data["session_id"],
         "user_id": user.id,
     }
@@ -316,12 +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,
@@ -330,7 +329,9 @@ async def chat_completed(request: Request, form_data: dict, user: Any):
     try:
         filter_functions = [
             Functions.get_function_by_id(filter_id)
-            for filter_id in get_sorted_filter_ids(model)
+            for filter_id in get_sorted_filter_ids(
+                request, model, metadata.get("filter_ids", [])
+            )
         ]
 
         result, _ = await process_filter_functions(
@@ -389,11 +390,7 @@ async def chat_action(request: Request, action_id: str, form_data: dict, user: A
         }
     )
 
-    if action_id in request.app.state.FUNCTIONS:
-        function_module = request.app.state.FUNCTIONS[action_id]
-    else:
-        function_module, _, _ = load_function_module_by_id(action_id)
-        request.app.state.FUNCTIONS[action_id] = function_module
+    function_module, _, _ = get_function_module_from_cache(request, action_id)
 
     if hasattr(function_module, "valves") and hasattr(function_module, "Valves"):
         valves = Functions.get_function_valves_by_id(action_id)
@@ -422,12 +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/code_interpreter.py

@@ -44,12 +44,14 @@ class JupyterCodeExecuter:
         :param password: Jupyter password (optional)
         :param timeout: WebSocket timeout in seconds (default: 60s)
         """
-        self.base_url = base_url.rstrip("/")
+        self.base_url = base_url
         self.code = code
         self.token = token
         self.password = password
         self.timeout = timeout
         self.kernel_id = ""
+        if self.base_url[-1] != "/":
+            self.base_url += "/"
         self.session = aiohttp.ClientSession(trust_env=True, base_url=self.base_url)
         self.params = {}
         self.result = ResultModel()
@@ -61,7 +63,7 @@ class JupyterCodeExecuter:
         if self.kernel_id:
             try:
                 async with self.session.delete(
-                    f"/api/kernels/{self.kernel_id}", params=self.params
+                    f"api/kernels/{self.kernel_id}", params=self.params
                 ) as response:
                     response.raise_for_status()
             except Exception as err:
@@ -81,7 +83,7 @@ class JupyterCodeExecuter:
     async def sign_in(self) -> None:
         # password authentication
         if self.password and not self.token:
-            async with self.session.get("/login") as response:
+            async with self.session.get("login") as response:
                 response.raise_for_status()
                 xsrf_token = response.cookies["_xsrf"].value
                 if not xsrf_token:
@@ -89,7 +91,7 @@ class JupyterCodeExecuter:
                 self.session.cookie_jar.update_cookies(response.cookies)
                 self.session.headers.update({"X-XSRFToken": xsrf_token})
             async with self.session.post(
-                "/login",
+                "login",
                 data={"_xsrf": xsrf_token, "password": self.password},
                 allow_redirects=False,
             ) as response:
@@ -101,17 +103,15 @@ class JupyterCodeExecuter:
             self.params.update({"token": self.token})
 
     async def init_kernel(self) -> None:
-        async with self.session.post(
-            url="/api/kernels", params=self.params
-        ) as response:
+        async with self.session.post(url="api/kernels", params=self.params) as response:
             response.raise_for_status()
             kernel_data = await response.json()
             self.kernel_id = kernel_data["id"]
 
     def init_ws(self) -> (str, dict):
-        ws_base = self.base_url.replace("http", "ws")
+        ws_base = self.base_url.replace("http", "ws", 1)
         ws_params = "?" + "&".join([f"{key}={val}" for key, val in self.params.items()])
-        websocket_url = f"{ws_base}/api/kernels/{self.kernel_id}/channels{ws_params if len(ws_params) > 1 else ''}"
+        websocket_url = f"{ws_base}api/kernels/{self.kernel_id}/channels{ws_params if len(ws_params) > 1 else ''}"
         ws_headers = {}
         if self.password and not self.token:
             ws_headers = {

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/filter.py

@@ -1,7 +1,10 @@
 import inspect
 import logging
 
-from open_webui.utils.plugin import load_function_module_by_id
+from open_webui.utils.plugin import (
+    load_function_module_by_id,
+    get_function_module_from_cache,
+)
 from open_webui.models.functions import Functions
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -9,7 +12,17 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
 
-def get_sorted_filter_ids(model: dict):
+def get_function_module(request, function_id, load_from_db=True):
+    """
+    Get the function module by its ID.
+    """
+    function_module, _, _ = get_function_module_from_cache(
+        request, function_id, load_from_db
+    )
+    return function_module
+
+
+def get_sorted_filter_ids(request, model: dict, enabled_filter_ids: list = None):
     def get_priority(function_id):
         function = Functions.get_function_by_id(function_id)
         if function is not None:
@@ -21,14 +34,26 @@ def get_sorted_filter_ids(model: dict):
     if "info" in model and "meta" in model["info"]:
         filter_ids.extend(model["info"]["meta"].get("filterIds", []))
         filter_ids = list(set(filter_ids))
-
-    enabled_filter_ids = [
+    active_filter_ids = [
         function.id
         for function in Functions.get_functions_by_type("filter", active_only=True)
     ]
 
-    filter_ids = [fid for fid in filter_ids if fid in enabled_filter_ids]
+    def get_active_status(filter_id):
+        function_module = get_function_module(request, filter_id)
+
+        if getattr(function_module, "toggle", None):
+            return filter_id in (enabled_filter_ids or [])
+
+        return True
+
+    active_filter_ids = [
+        filter_id for filter_id in active_filter_ids if get_active_status(filter_id)
+    ]
+
+    filter_ids = [fid for fid in filter_ids if fid in active_filter_ids]
     filter_ids.sort(key=get_priority)
+
     return filter_ids
 
 
@@ -43,12 +68,9 @@ async def process_filter_functions(
         if not filter:
             continue
 
-        if filter_id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[filter_id]
-        else:
-            function_module, _, _ = load_function_module_by_id(filter_id)
-            request.app.state.FUNCTIONS[filter_id] = function_module
-
+        function_module = get_function_module(
+            request, filter_id, load_from_db=(filter_type != "stream")
+        )
         # Prepare handler function
         handler = getattr(function_module, filter_type, None)
         if not handler:

backend/open_webui/utils/logger.py

@@ -5,7 +5,9 @@ from typing import TYPE_CHECKING
 
 from loguru import logger
 
+
 from open_webui.env import (
+    AUDIT_UVICORN_LOGGER_NAMES,
     AUDIT_LOG_FILE_ROTATION_SIZE,
     AUDIT_LOG_LEVEL,
     AUDIT_LOGS_FILE_PATH,
@@ -128,11 +130,13 @@ 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

@@ -34,11 +34,15 @@ def get_message_list(messages, message_id):
     :return: List of ordered messages starting from the root to the given message
     """
 
+    # Handle case where messages is None
+    if not messages:
+        return []  # Return empty list instead of None to prevent iteration errors
+
     # Find the message by its id
     current_message = messages.get(message_id)
 
     if not current_message:
-        return None
+        return []  # Return empty list instead of None to prevent iteration errors
 
     # Reconstruct the chain by following the parentId links
     message_list = []
@@ -47,7 +51,7 @@ def get_message_list(messages, message_id):
         message_list.insert(
             0, current_message
         )  # Insert the message at the beginning of the list
-        parent_id = current_message["parentId"]
+        parent_id = current_message.get("parentId")  # Use .get() for safety
         current_message = messages.get(parent_id) if parent_id else None
 
     return message_list
@@ -70,12 +74,12 @@ def get_last_user_message_item(messages: list[dict]) -> Optional[dict]:
 
 
 def get_content_from_message(message: dict) -> Optional[str]:
-    if isinstance(message["content"], list):
+    if isinstance(message.get("content"), list):
         for item in message["content"]:
             if item["type"] == "text":
                 return item["text"]
     else:
-        return message["content"]
+        return message.get("content")
     return None
 
 
@@ -130,7 +134,9 @@ def prepend_to_first_user_message_content(
     return messages
 
 
-def add_or_update_system_message(content: str, messages: list[dict]):
+def add_or_update_system_message(
+    content: str, messages: list[dict], append: bool = False
+):
     """
     Adds a new system message at the beginning of the messages list
     or updates the existing system message at the beginning.
@@ -141,7 +147,10 @@ def add_or_update_system_message(content: str, messages: list[dict]):
     """
 
     if messages and messages[0].get("role") == "system":
-        messages[0]["content"] = f"{content}\n{messages[0]['content']}"
+        if append:
+            messages[0]["content"] = f"{messages[0]['content']}\n{content}"
+        else:
+            messages[0]["content"] = f"{content}\n{messages[0]['content']}"
     else:
         # Insert at the beginning
         messages.insert(0, {"role": "system", "content": content})
@@ -199,6 +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:
@@ -211,6 +221,9 @@ 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
 
@@ -225,6 +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:
@@ -232,8 +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

@@ -1,5 +1,6 @@
 import time
 import logging
+import asyncio
 import sys
 
 from aiocache import cached
@@ -13,7 +14,10 @@ from open_webui.models.functions import Functions
 from open_webui.models.models import Models
 
 
-from open_webui.utils.plugin import load_function_module_by_id
+from open_webui.utils.plugin import (
+    load_function_module_by_id,
+    get_function_module_from_cache,
+)
 from open_webui.utils.access_control import has_access
 
 
@@ -30,38 +34,61 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
 
+async def fetch_ollama_models(request: Request, user: UserModel = None):
+    raw_ollama_models = await ollama.get_all_models(request, user=user)
+    return [
+        {
+            "id": model["model"],
+            "name": model["name"],
+            "object": "model",
+            "created": int(time.time()),
+            "owned_by": "ollama",
+            "ollama": model,
+            "connection_type": model.get("connection_type", "local"),
+            "tags": model.get("tags", []),
+        }
+        for model in raw_ollama_models["models"]
+    ]
+
+
+async def fetch_openai_models(request: Request, user: UserModel = None):
+    openai_response = await openai.get_all_models(request, user=user)
+    return openai_response["data"]
+
+
 async def get_all_base_models(request: Request, user: UserModel = None):
-    function_models = []
-    openai_models = []
-    ollama_models = []
+    openai_task = (
+        fetch_openai_models(request, user)
+        if request.app.state.config.ENABLE_OPENAI_API
+        else asyncio.sleep(0, result=[])
+    )
+    ollama_task = (
+        fetch_ollama_models(request, user)
+        if request.app.state.config.ENABLE_OLLAMA_API
+        else asyncio.sleep(0, result=[])
+    )
+    function_task = get_function_models(request)
 
-    if request.app.state.config.ENABLE_OPENAI_API:
-        openai_models = await openai.get_all_models(request, user=user)
-        openai_models = openai_models["data"]
+    openai_models, ollama_models, function_models = await asyncio.gather(
+        openai_task, ollama_task, function_task
+    )
 
-    if request.app.state.config.ENABLE_OLLAMA_API:
-        ollama_models = await ollama.get_all_models(request, user=user)
-        ollama_models = [
-            {
-                "id": model["model"],
-                "name": model["name"],
-                "object": "model",
-                "created": int(time.time()),
-                "owned_by": "ollama",
-                "ollama": model,
-                "tags": model.get("tags", []),
-            }
-            for model in ollama_models["models"]
-        ]
-
-    function_models = await get_function_models(request)
-    models = function_models + openai_models + ollama_models
-
-    return models
+    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:
@@ -110,9 +137,18 @@ async def get_all_models(request, user: UserModel = None):
         for function in Functions.get_functions_by_type("action", active_only=True)
     ]
 
+    global_filter_ids = [
+        function.id for function in Functions.get_global_filter_functions()
+    ]
+    enabled_filter_ids = [
+        function.id
+        for function in Functions.get_functions_by_type("filter", active_only=True)
+    ]
+
     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"
@@ -125,13 +161,20 @@ async def get_all_models(request, user: UserModel = None):
                         model["name"] = custom_model.name
                         model["info"] = custom_model.model_dump()
 
+                        # Set action_ids and filter_ids
                         action_ids = []
+                        filter_ids = []
+
                         if "info" in model and "meta" in model["info"]:
                             action_ids.extend(
                                 model["info"]["meta"].get("actionIds", [])
                             )
+                            filter_ids.extend(
+                                model["info"]["meta"].get("filterIds", [])
+                            )
 
                         model["action_ids"] = action_ids
+                        model["filter_ids"] = filter_ids
                     else:
                         models.remove(model)
 
@@ -140,7 +183,9 @@ async def get_all_models(request, user: UserModel = None):
         ):
             owned_by = "openai"
             pipe = None
+
             action_ids = []
+            filter_ids = []
 
             for model in models:
                 if (
@@ -154,9 +199,13 @@ async def get_all_models(request, user: UserModel = None):
 
             if custom_model.meta:
                 meta = custom_model.meta.model_dump()
+
                 if "actionIds" in meta:
                     action_ids.extend(meta["actionIds"])
 
+                if "filterIds" in meta:
+                    filter_ids.extend(meta["filterIds"])
+
             models.append(
                 {
                     "id": f"{custom_model.id}",
@@ -168,6 +217,7 @@ async def get_all_models(request, user: UserModel = None):
                     "preset": True,
                     **({"pipe": pipe} if pipe is not None else {}),
                     "action_ids": action_ids,
+                    "filter_ids": filter_ids,
                 }
             )
 
@@ -181,8 +231,11 @@ async def get_all_models(request, user: UserModel = None):
                     "id": f"{function.id}.{action['id']}",
                     "name": action.get("name", f"{function.name} ({action['id']})"),
                     "description": function.meta.description,
-                    "icon_url": action.get(
-                        "icon_url", function.meta.manifest.get("icon_url", None)
+                    "icon": action.get(
+                        "icon_url",
+                        function.meta.manifest.get("icon_url", None)
+                        or getattr(module, "icon_url", None)
+                        or getattr(module, "icon", None),
                     ),
                 }
                 for action in actions
@@ -193,16 +246,27 @@ async def get_all_models(request, user: UserModel = None):
                     "id": function.id,
                     "name": function.name,
                     "description": function.meta.description,
-                    "icon_url": function.meta.manifest.get("icon_url", None),
+                    "icon": function.meta.manifest.get("icon_url", None)
+                    or getattr(module, "icon_url", None)
+                    or getattr(module, "icon", None),
                 }
             ]
 
+    # Process filter_ids to get the filters
+    def get_filter_items_from_module(function, module):
+        return [
+            {
+                "id": function.id,
+                "name": function.name,
+                "description": function.meta.description,
+                "icon": function.meta.manifest.get("icon_url", None)
+                or getattr(module, "icon_url", None)
+                or getattr(module, "icon", None),
+            }
+        ]
+
     def get_function_module_by_id(function_id):
-        if function_id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[function_id]
-        else:
-            function_module, _, _ = load_function_module_by_id(function_id)
-            request.app.state.FUNCTIONS[function_id] = function_module
+        function_module, _, _ = get_function_module_from_cache(request, function_id)
         return function_module
 
     for model in models:
@@ -211,6 +275,11 @@ async def get_all_models(request, user: UserModel = None):
             for action_id in list(set(model.pop("action_ids", []) + global_action_ids))
             if action_id in enabled_action_ids
         ]
+        filter_ids = [
+            filter_id
+            for filter_id in list(set(model.pop("filter_ids", []) + global_filter_ids))
+            if filter_id in enabled_filter_ids
+        ]
 
         model["actions"] = []
         for action_id in action_ids:
@@ -222,6 +291,20 @@ async def get_all_models(request, user: UserModel = None):
             model["actions"].extend(
                 get_action_items_from_module(action_function, function_module)
             )
+
+        model["filters"] = []
+        for filter_id in filter_ids:
+            filter_function = Functions.get_function_by_id(filter_id)
+            if filter_function is None:
+                raise Exception(f"Filter not found: {filter_id}")
+
+            function_module = get_function_module_by_id(filter_id)
+
+            if getattr(function_module, "toggle", None):
+                model["filters"].extend(
+                    get_filter_items_from_module(filter_function, function_module)
+                )
+
     log.debug(f"get_all_models() returned {len(models)} models")
 
     request.app.state.MODELS = {model["id"]: model for model in models}

backend/open_webui/utils/oauth.py

@@ -34,12 +34,14 @@ from open_webui.config import (
     OAUTH_ALLOWED_ROLES,
     OAUTH_ADMIN_ROLES,
     OAUTH_ALLOWED_DOMAINS,
+    OAUTH_UPDATE_PICTURE_ON_LOGIN,
     WEBHOOK_URL,
     JWT_EXPIRES_IN,
     AppConfig,
 )
 from open_webui.constants import ERROR_MESSAGES, WEBHOOK_MESSAGES
 from open_webui.env import (
+    AIOHTTP_CLIENT_SESSION_SSL,
     WEBUI_NAME,
     WEBUI_AUTH_COOKIE_SAME_SITE,
     WEBUI_AUTH_COOKIE_SECURE,
@@ -72,6 +74,7 @@ auth_manager_config.OAUTH_ADMIN_ROLES = OAUTH_ADMIN_ROLES
 auth_manager_config.OAUTH_ALLOWED_DOMAINS = OAUTH_ALLOWED_DOMAINS
 auth_manager_config.WEBHOOK_URL = WEBHOOK_URL
 auth_manager_config.JWT_EXPIRES_IN = JWT_EXPIRES_IN
+auth_manager_config.OAUTH_UPDATE_PICTURE_ON_LOGIN = OAUTH_UPDATE_PICTURE_ON_LOGIN
 
 
 class OAuthManager:
@@ -282,6 +285,51 @@ class OAuthManager:
                     id=group_model.id, form_data=update_form, overwrite=False
                 )
 
+    async def _process_picture_url(
+        self, picture_url: str, access_token: str = None
+    ) -> str:
+        """Process a picture URL and return a base64 encoded data URL.
+
+        Args:
+            picture_url: The URL of the picture to process
+            access_token: Optional OAuth access token for authenticated requests
+
+        Returns:
+            A data URL containing the base64 encoded picture, or "/user.png" if processing fails
+        """
+        if not picture_url:
+            return "/user.png"
+
+        try:
+            get_kwargs = {}
+            if access_token:
+                get_kwargs["headers"] = {
+                    "Authorization": f"Bearer {access_token}",
+                }
+            async with aiohttp.ClientSession(trust_env=True) as session:
+                async with session.get(
+                    picture_url, **get_kwargs, ssl=AIOHTTP_CLIENT_SESSION_SSL
+                ) as resp:
+                    if resp.ok:
+                        picture = await resp.read()
+                        base64_encoded_picture = base64.b64encode(picture).decode(
+                            "utf-8"
+                        )
+                        guessed_mime_type = mimetypes.guess_type(picture_url)[0]
+                        if guessed_mime_type is None:
+                            guessed_mime_type = "image/jpeg"
+                        return (
+                            f"data:{guessed_mime_type};base64,{base64_encoded_picture}"
+                        )
+                    else:
+                        log.warning(
+                            f"Failed to fetch profile picture from {picture_url}"
+                        )
+                        return "/user.png"
+        except Exception as e:
+            log.error(f"Error processing profile picture '{picture_url}': {e}")
+            return "/user.png"
+
     async def handle_login(self, request, provider):
         if provider not in OAUTH_PROVIDERS:
             raise HTTPException(404)
@@ -326,7 +374,9 @@ class OAuthManager:
                     headers = {"Authorization": f"Bearer {access_token}"}
                     async with aiohttp.ClientSession(trust_env=True) as session:
                         async with session.get(
-                            "https://api.github.com/user/emails", headers=headers
+                            "https://api.github.com/user/emails",
+                            headers=headers,
+                            ssl=AIOHTTP_CLIENT_SESSION_SSL,
                         ) as resp:
                             if resp.ok:
                                 emails = await resp.json()
@@ -382,6 +432,22 @@ class OAuthManager:
             if user.role != determined_role:
                 Users.update_user_role_by_id(user.id, determined_role)
 
+            # Update profile picture if enabled and different from current
+            if auth_manager_config.OAUTH_UPDATE_PICTURE_ON_LOGIN:
+                picture_claim = auth_manager_config.OAUTH_PICTURE_CLAIM
+                if picture_claim:
+                    new_picture_url = user_data.get(
+                        picture_claim, OAUTH_PROVIDERS[provider].get("picture_url", "")
+                    )
+                    processed_picture_url = await self._process_picture_url(
+                        new_picture_url, token.get("access_token")
+                    )
+                    if processed_picture_url != user.profile_image_url:
+                        Users.update_user_profile_image_url_by_id(
+                            user.id, processed_picture_url
+                        )
+                        log.debug(f"Updated profile picture for user {user.email}")
+
         if not user:
             user_count = Users.get_num_users()
 
@@ -397,40 +463,9 @@ class OAuthManager:
                     picture_url = user_data.get(
                         picture_claim, OAUTH_PROVIDERS[provider].get("picture_url", "")
                     )
-                    if picture_url:
-                        # Download the profile image into a base64 string
-                        try:
-                            access_token = token.get("access_token")
-                            get_kwargs = {}
-                            if access_token:
-                                get_kwargs["headers"] = {
-                                    "Authorization": f"Bearer {access_token}",
-                                }
-                            async with aiohttp.ClientSession(trust_env=True) as session:
-                                async with session.get(
-                                    picture_url, **get_kwargs
-                                ) as resp:
-                                    if resp.ok:
-                                        picture = await resp.read()
-                                        base64_encoded_picture = base64.b64encode(
-                                            picture
-                                        ).decode("utf-8")
-                                        guessed_mime_type = mimetypes.guess_type(
-                                            picture_url
-                                        )[0]
-                                        if guessed_mime_type is None:
-                                            # assume JPG, browsers are tolerant enough of image formats
-                                            guessed_mime_type = "image/jpeg"
-                                        picture_url = f"data:{guessed_mime_type};base64,{base64_encoded_picture}"
-                                    else:
-                                        picture_url = "/user.png"
-                        except Exception as e:
-                            log.error(
-                                f"Error downloading profile image '{picture_url}': {e}"
-                            )
-                            picture_url = "/user.png"
-                    if not picture_url:
-                        picture_url = "/user.png"
+                    picture_url = await self._process_picture_url(
+                        picture_url, token.get("access_token")
+                    )
                 else:
                     picture_url = "/user.png"
 
@@ -501,5 +536,10 @@ class OAuthManager:
                 secure=WEBUI_AUTH_COOKIE_SECURE,
             )
         # Redirect back to the frontend with the JWT token
-        redirect_url = f"{request.base_url}auth#token={jwt_token}"
+
+        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}"
+
         return RedirectResponse(url=redirect_url, headers=response.headers)

backend/open_webui/utils/payload.py

@@ -1,5 +1,6 @@
 from open_webui.utils.task import prompt_template, prompt_variables_template
 from open_webui.utils.misc import (
+    deep_update,
     add_or_update_system_message,
 )
 
@@ -9,9 +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
 
@@ -45,20 +45,67 @@ def apply_model_params_to_body(
     if not params:
         return form_data
 
-    for key, cast_func in mappings.items():
-        if (value := params.get(key)) is not None:
-            form_data[key] = cast_func(value)
+    for key, value in params.items():
+        if value is not None:
+            if key in mappings:
+                cast_func = mappings[key]
+                if isinstance(cast_func, Callable):
+                    form_data[key] = cast_func(value)
+            else:
+                form_data[key] = value
 
     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,
+        "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)
+
     mappings = {
         "temperature": float,
         "top_p": float,
+        "min_p": float,
         "max_tokens": int,
         "frequency_penalty": float,
+        "presence_penalty": float,
         "reasoning_effort": str,
         "seed": lambda x: x,
         "stop": lambda x: [bytes(s, "utf-8").decode("unicode_escape") for s in x],
@@ -69,6 +116,23 @@ 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)
+
     # Convert OpenAI parameter names to Ollama parameter names if needed.
     name_differences = {
         "max_tokens": "num_predict",
@@ -111,16 +175,32 @@ 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]:
@@ -215,36 +295,48 @@ 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:
@@ -265,3 +357,32 @@ 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/plugin.py

@@ -115,7 +115,7 @@ def load_tool_module_by_id(tool_id, content=None):
         os.unlink(temp_file.name)
 
 
-def load_function_module_by_id(function_id, content=None):
+def load_function_module_by_id(function_id: str, content: str | None = None):
     if content is None:
         function = Functions.get_function_by_id(function_id)
         if not function:
@@ -166,6 +166,62 @@ def load_function_module_by_id(function_id, content=None):
         os.unlink(temp_file.name)
 
 
+def get_function_module_from_cache(request, function_id, load_from_db=True):
+    if load_from_db:
+        # Always load from the database by default
+        # This is useful for hooks like "inlet" or "outlet" where the content might change
+        # and we want to ensure the latest content is used.
+
+        function = Functions.get_function_by_id(function_id)
+        if not function:
+            raise Exception(f"Function not found: {function_id}")
+        content = function.content
+
+        new_content = replace_imports(content)
+        if new_content != content:
+            content = new_content
+            # Update the function content in the database
+            Functions.update_function_by_id(function_id, {"content": content})
+
+        if (
+            hasattr(request.app.state, "FUNCTION_CONTENTS")
+            and function_id in request.app.state.FUNCTION_CONTENTS
+        ) and (
+            hasattr(request.app.state, "FUNCTIONS")
+            and function_id in request.app.state.FUNCTIONS
+        ):
+            if request.app.state.FUNCTION_CONTENTS[function_id] == content:
+                return request.app.state.FUNCTIONS[function_id], None, None
+
+        function_module, function_type, frontmatter = load_function_module_by_id(
+            function_id, content
+        )
+    else:
+        # Load from cache (e.g. "stream" hook)
+        # This is useful for performance reasons
+
+        if (
+            hasattr(request.app.state, "FUNCTIONS")
+            and function_id in request.app.state.FUNCTIONS
+        ):
+            return request.app.state.FUNCTIONS[function_id], None, None
+
+        function_module, function_type, frontmatter = load_function_module_by_id(
+            function_id
+        )
+
+    if not hasattr(request.app.state, "FUNCTIONS"):
+        request.app.state.FUNCTIONS = {}
+
+    if not hasattr(request.app.state, "FUNCTION_CONTENTS"):
+        request.app.state.FUNCTION_CONTENTS = {}
+
+    request.app.state.FUNCTIONS[function_id] = function_module
+    request.app.state.FUNCTION_CONTENTS[function_id] = content
+
+    return function_module, function_type, frontmatter
+
+
 def install_frontmatter_requirements(requirements: str):
     if requirements:
         try:

backend/open_webui/utils/redis.py

@@ -1,7 +1,6 @@
 import socketio
-import redis
-from redis import asyncio as aioredis
 from urllib.parse import urlparse
+from typing import Optional
 
 
 def parse_redis_service_url(redis_url):
@@ -18,23 +17,46 @@ 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, async_mode=False, decode_responses=True
+):
+    if async_mode:
+        import redis.asyncio as redis
 
-        # Get a master connection from Sentinel
-        return sentinel.master_for(redis_config["service"])
+        # 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,
+            )
+            return sentinel.master_for(redis_config["service"])
+        elif redis_url:
+            return redis.from_url(redis_url, decode_responses=decode_responses)
+        else:
+            return None
     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,
+            )
+            return sentinel.master_for(redis_config["service"])
+        elif redis_url:
+            return redis.Redis.from_url(redis_url, decode_responses=decode_responses)
+        else:
+            return None
 
 
 def get_sentinels_from_env(sentinel_hosts_env, sentinel_port_env):

backend/open_webui/utils/response.py

@@ -83,6 +83,7 @@ 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 +95,7 @@ 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 +106,7 @@ 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 +120,71 @@ 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

@@ -22,7 +22,7 @@ def get_task_model_id(
     # Set the task model
     task_model_id = default_model_id
     # Check if the user has a custom task model and use that model
-    if models[task_model_id].get("owned_by") == "ollama":
+    if models[task_model_id].get("connection_type") == "local":
         if task_model and task_model in models:
             task_model_id = task_model
     else:
@@ -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/metrics.py

@@ -0,0 +1,110 @@
+"""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 fastapi import FastAPI, Request
+from opentelemetry import metrics
+from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import (
+    OTLPMetricExporter,
+)
+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 SERVICE_NAME, Resource
+
+from open_webui.env import OTEL_SERVICE_NAME, OTEL_EXPORTER_OTLP_ENDPOINT
+
+
+_EXPORT_INTERVAL_MILLIS = 10_000  # 10 seconds
+
+
+def _build_meter_provider() -> MeterProvider:
+    """Return a configured MeterProvider."""
+
+    # Periodic reader pushes metrics over OTLP/gRPC to collector
+    readers: List[PeriodicExportingMetricReader] = [
+        PeriodicExportingMetricReader(
+            OTLPMetricExporter(endpoint=OTEL_EXPORTER_OTLP_ENDPOINT),
+            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"],
+        ),
+    ]
+
+    provider = MeterProvider(
+        resource=Resource.create({SERVICE_NAME: OTEL_SERVICE_NAME}),
+        metric_readers=list(readers),
+        views=views,
+    )
+    return provider
+
+
+def setup_metrics(app: FastAPI) -> None:
+    """Attach OTel metrics middleware to *app* and initialise provider."""
+
+    metrics.set_meter_provider(_build_meter_provider())
+    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",
+    )
+
+    # 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

@@ -1,13 +1,26 @@
 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 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):
@@ -17,7 +30,30 @@ def setup(app: FastAPI, db_engine: Engine):
             resource=Resource.create(attributes={SERVICE_NAME: OTEL_SERVICE_NAME})
         )
     )
+
+    # 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)
+    if OTEL_OTLP_SPAN_EXPORTER == "http":
+        exporter = HttpOTLPSpanExporter(
+            endpoint=OTEL_EXPORTER_OTLP_ENDPOINT,
+            insecure=OTEL_EXPORTER_OTLP_INSECURE,
+            headers=headers,
+        )
+    else:
+        exporter = OTLPSpanExporter(
+            endpoint=OTEL_EXPORTER_OTLP_ENDPOINT,
+            insecure=OTEL_EXPORTER_OTLP_INSECURE,
+            headers=headers,
+        )
     trace.get_tracer_provider().add_span_processor(LazyBatchSpanProcessor(exporter))
     Instrumentor(app=app, db_engine=db_engine).instrument()
+
+    # set up metrics only if enabled
+    if ENABLE_OTEL_METRICS:
+        setup_metrics(app)

backend/open_webui/utils/tools.py

@@ -37,6 +37,7 @@ from open_webui.models.tools import Tools
 from open_webui.models.users import UserModel
 from open_webui.utils.plugin import load_tool_module_by_id
 from open_webui.env import (
+    SRC_LOG_LEVELS,
     AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA,
     AIOHTTP_CLIENT_SESSION_TOOL_SERVER_SSL,
 )
@@ -44,6 +45,7 @@ from open_webui.env import (
 import copy
 
 log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["MODELS"])
 
 
 def get_async_tool_function_and_apply_extra_params(
@@ -99,9 +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"],
@@ -158,7 +157,7 @@ def get_tools(
                 # TODO: Fix hack for OpenAI API
                 # Some times breaks OpenAI but others don't. Leaving the comment
                 for val in spec.get("parameters", {}).get("properties", {}).values():
-                    if val["type"] == "str":
+                    if val.get("type") == "str":
                         val["type"] = "string"
 
                 # Remove internal reserved parameters (e.g. __id__, __user__)
@@ -477,7 +476,7 @@ async def get_tool_server_data(token: str, url: str) -> Dict[str, Any]:
         "specs": convert_openapi_to_tool_payload(res),
     }
 
-    print("Fetched data:", data)
+    log.info(f"Fetched data: {data}")
     return data
 
 
@@ -488,8 +487,19 @@ async def get_tool_servers_data(
     server_entries = []
     for idx, server in enumerate(servers):
         if server.get("config", {}).get("enable"):
-            url_path = server.get("path", "openapi.json")
-            full_url = f"{server.get('url')}/{url_path}"
+            # 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}"
+
+            info = server.get("info", {})
 
             auth_type = server.get("auth_type", "bearer")
             token = None
@@ -498,26 +508,37 @@ async def get_tool_servers_data(
                 token = server.get("key", "")
             elif auth_type == "session":
                 token = session_token
-            server_entries.append((idx, server, full_url, token))
+            server_entries.append((idx, server, full_url, info, token))
 
     # Create async tasks to fetch data
-    tasks = [get_tool_server_data(token, url) for (_, _, url, token) in server_entries]
+    tasks = [
+        get_tool_server_data(token, url) for (_, _, url, _, token) in server_entries
+    ]
 
     # Execute tasks concurrently
     responses = await asyncio.gather(*tasks, return_exceptions=True)
 
     # Build final results with index and server metadata
     results = []
-    for (idx, server, url, _), response in zip(server_entries, responses):
+    for (idx, server, url, info, _), response in zip(server_entries, responses):
         if isinstance(response, Exception):
-            print(f"Failed to connect to {url} OpenAPI tool server")
+            log.error(f"Failed to connect to {url} OpenAPI tool server")
             continue
 
+        openapi_data = response.get("openapi", {})
+
+        if info and isinstance(openapi_data, dict):
+            if "name" in info:
+                openapi_data["info"]["title"] = info.get("name", "Tool Server")
+
+            if "description" in info:
+                openapi_data["info"]["description"] = info.get("description", "")
+
         results.append(
             {
                 "idx": idx,
                 "url": server.get("url"),
-                "openapi": response.get("openapi"),
+                "openapi": openapi_data,
                 "info": response.get("info"),
                 "specs": response.get("specs"),
             }
@@ -620,5 +641,5 @@ async def execute_tool_server(
 
     except Exception as err:
         error = str(err)
-        print("API Request Error:", error)
+        log.exception(f"API Request Error: {error}")
         return {"error": error}

backend/requirements.txt

@@ -1,21 +1,24 @@
 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
+requests==2.32.4
 aiohttp==3.11.11
 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
-peewee==3.17.9
+peewee==3.18.1
 peewee-migrate==1.12.2
 psycopg2-binary==2.9.9
 pgvector==0.4.0
@@ -29,6 +32,8 @@ boto3==1.35.53
 argon2-cffi==23.1.0
 APScheduler==3.10.4
 
+pycrdt==0.12.25
+
 RestrictedPython==8.0
 
 loguru==0.7.3
@@ -37,16 +42,17 @@ asgiref==3.8.1
 # AI libraries
 openai
 anthropic
-google-generativeai==0.8.4
+google-genai==1.15.0
+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
 chromadb==0.6.3
 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
@@ -64,7 +70,7 @@ 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
@@ -73,13 +79,13 @@ 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==11.2.1
 opencv-python-headless==4.11.0.86
 rapidocr-onnxruntime==1.4.4
 rank-bm25==0.2.2
@@ -93,12 +99,11 @@ authlib==1.4.1
 
 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~=8.0.0
+ddgs==9.0.0
 
 ## Google Drive
 google-api-python-client
@@ -113,7 +118,7 @@ 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
 
 

backend/start.sh

@@ -14,7 +14,11 @@ if [[ "${WEB_LOADER_ENGINE,,}" == "playwright" ]]; then
     python -c "import nltk; nltk.download('punkt_tab')"
 fi
 
-KEY_FILE=.webui_secret_key
+if [ -n "${WEBUI_SECRET_KEY_FILE}" ]; then
+    KEY_FILE="${WEBUI_SECRET_KEY_FILE}"
+else
+    KEY_FILE=".webui_secret_key"
+fi
 
 PORT="${PORT:-8080}"
 HOST="${HOST:-0.0.0.0}"
@@ -65,4 +69,6 @@ if [ -n "$SPACE_ID" ]; then
   export WEBUI_URL=${SPACE_HOST}
 fi
 
-WEBUI_SECRET_KEY="$WEBUI_SECRET_KEY" exec uvicorn open_webui.main:app --host "$HOST" --port "$PORT" --forwarded-allow-ips '*' --workers "${UVICORN_WORKERS:-1}"
+PYTHON_CMD=$(command -v python3 || command -v python)
+
+WEBUI_SECRET_KEY="$WEBUI_SECRET_KEY" exec "$PYTHON_CMD" -m uvicorn open_webui.main:app --host "$HOST" --port "$PORT" --forwarded-allow-ips '*' --workers "${UVICORN_WORKERS:-1}"

backend/start_windows.bat

@@ -18,13 +18,17 @@ IF /I "%WEB_LOADER_ENGINE%" == "playwright" (
 )
 
 SET "KEY_FILE=.webui_secret_key"
+IF NOT "%WEBUI_SECRET_KEY_FILE%" == "" (
+    SET "KEY_FILE=%WEBUI_SECRET_KEY_FILE%"
+)
+
 IF "%PORT%"=="" SET PORT=8080
 IF "%HOST%"=="" SET HOST=0.0.0.0
 SET "WEBUI_SECRET_KEY=%WEBUI_SECRET_KEY%"
 SET "WEBUI_JWT_SECRET_KEY=%WEBUI_JWT_SECRET_KEY%"
 
 :: Check if WEBUI_SECRET_KEY and WEBUI_JWT_SECRET_KEY are not set
-IF "%WEBUI_SECRET_KEY%%WEBUI_JWT_SECRET_KEY%" == " " (
+IF "%WEBUI_SECRET_KEY% %WEBUI_JWT_SECRET_KEY%" == " " (
     echo Loading WEBUI_SECRET_KEY from file, not provided as an environment variable.
 
     IF NOT EXIST "%KEY_FILE%" (