Merge pull request #14675 from open-webui/dev

0.6.14

.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/pull_request_template.md

@@ -70,3 +70,7 @@
 ### Screenshots or Videos
 
 - [Attach any relevant screenshots or videos demonstrating the changes]
+
+### Contributor License Agreement
+
+By submitting this pull request, I confirm that I have read and fully agree to the [Contributor License Agreement (CLA)](/CONTRIBUTOR_LICENSE_AGREEMENT), and I am providing my contributions under its terms.

.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 +5,336 @@ 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).
 
+## [0.6.14] - 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
+
+- 🌐 **Custom Azure TTS API URL Support Added**: You can now define a custom Azure Text-to-Speech endpoint—enabling flexibility for enterprise deployments and regional compliance.
+- ⚙️ **TOOL_SERVER_CONNECTIONS Environment Variable Suppor**: Easily configure and deploy tool servers via environment variables, streamlining setup and enabling faster enterprise provisioning.
+- 👥 **Enhanced OAuth Group Handling as String or List**: OAuth group data can now be passed as either a list or a comma-separated string, improving compatibility with varied identity provider formats and reducing onboarding friction.
+
+### Fixed
+
+- 🧠 **Embedding with Ollama Proxy Endpoints Restored**: Fixed an issue where missing API config broke embedding for proxied Ollama models—ensuring consistent performance and compatibility.
+- 🔐 **OIDC OAuth Login Issue Resolved**: Users can once again sign in seamlessly using OpenID Connect-based OAuth, eliminating login interruptions and improving reliability.
+- 📝 **Notes Feature Access Fixed for Non-Admins**: Fixed an issue preventing non-admin users from accessing the Notes feature, restoring full cross-role collaboration capabilities.
+- 🖼️ **Tika Loader Image Extraction Problem Resolved**: Ensured TikaLoader now processes 'extract_images' parameter correctly, restoring complete file extraction functionality in document workflows.
+- 🎨 **Automatic1111 Image Model Setting Applied Properly**: Fixed an issue where switching to a specific image model via the UI wasn’t reflected in generation, re-enabling full visual creativity control.
+- 🏷️ **Multiple XML Tags in Messages Now Parsed Correctly**: Fixed parsing issues when messages included multiple XML-style tags, ensuring clean and unbroken rendering of rich content in chats.
+- 🖌️ **OpenAI Image Generation Issues Resolved**: Resolved broken image output when using OpenAI’s image generation, ensuring fully functional visual creation workflows.
+- 🔎 **Tool Server Settings UI Privacy Restored**: Prevented restricted users from accessing tool server settings via search—restoring tight permissions control and safeguarding sensitive configurations.
+- 🎧 **WebM Audio Transcription Now Supported**: Fixed an issue where WebM files failed during audio transcription—these formats are now fully supported, ensuring smoother voice note workflows and broader file compatibility.
+
+## [0.6.6] - 2025-05-05
+
+### Added
+
+- 📝 **AI-Enhanced Notes (With Audio Transcription)**: Effortlessly create notes, attach meeting or voice audio, and let the AI instantly enhance, summarize, or refine your notes using audio transcriptions—making your documentation smarter, cleaner, and more insightful with minimal effort.
+- 🔊 **Meeting Audio Recording & Import**: Seamlessly record audio from your meetings or capture screen audio and attach it to your notes—making it easier to revisit, annotate, and extract insights from important discussions.
+- 📁 **Import Markdown Notes Effortlessly**: Bring your existing knowledge library into Open WebUI by importing your Markdown notes, so you can leverage all advanced note management and AI features right away.
+- 👥 **Notes Permissions by User Group**: Fine-tune access and editing rights for notes based on user roles or groups, so you can delegate writing or restrict sensitive information as needed.
+- ☁️ **OneDrive & SharePoint Integration**: Keep your content in sync by connecting notes and files directly with OneDrive or SharePoint—unlocking fast enterprise import/export and seamless collaboration with your existing workflows.
+- 🗂️ **Paginated User List in Admin Panel**: Effortlessly manage and search through large teams via the new paginated user list—saving time and streamlining user administration in big organizations.
+- 🕹️ **Granular Chat Share & Export Permissions**: Enjoy enhanced control over who can share or export chats, enabling tighter governance and privacy in team and enterprise settings.
+- 🛑 **User Role Change Confirmation Dialog**: Reduce accidental privilege changes with a required confirmation step before updating user roles—improving security and preventing costly mistakes in team management.
+- 🚨 **Audit Log for Failed Login Attempts**: Quickly detect unauthorized access attempts or troubleshoot user login problems with detailed logs of failed authentication right in the audit trail.
+- 💡 **Dedicated 'Generate Title' Button for Chats**: Swiftly organize every conversation—tap the new button to let AI create relevant, clear titles for all your chats, saving time and reducing clutter.
+- 💬 **Notification Sound Always-On Option**: Take control of your notifications by setting sound alerts to always play—helping you stay on top of important updates in busy environments.
+- 🆔 **S3 File Tagging Support**: Uploaded files to S3 now include tags for better organization, searching, and integration with your file management policies.
+- 🛡️ **OAuth Blocked Groups Support**: Gain more control over group-based access by explicitly blocking specified OAuth groups—ideal for complex identity or security requirements.
+- 🚀 **Optimized Faster Web Search & Multi-Threaded Queries**: Enjoy dramatically faster web search and RAG (retrieval augmented generation) with revamped multi-threaded search—get richer, more accurate results in less time.
+- 🔍 **All-Knowledge Parallel Search**: Searches across your entire knowledge base now happen in parallel even in non-hybrid mode, speeding up responses and improving knowledge accuracy for every question.
+- 🌐 **New Firecrawl & Yacy Web Search Integrations**: Expand your world of information with two new advanced search engines—Firecrawl for deeper web insight and Yacy for decentralized, privacy-friendly search capabilities.
+- 🧠 **Configurable Docling OCR Engine & Language**: Use environment variables to fine-tune Docling OCR engine and supported languages for smarter, more tailored document extraction and RAG workflows.
+- 🗝️ **Enhanced Sentence Transformers Configuration**: Added new environment variables for easier set up and advanced customization of Sentence Transformers—ensuring best fit for your embedding needs.
+- 🌲 **Pinecone Vector Database Integration**: Index, search, and manage knowledge at enterprise scale with full native support for Pinecone as your vector database—effortlessly handle even the biggest document sets.
+- 🔄 **Automatic Requirements Installation for Tools & Functions**: Never worry about lost dependencies on restart—external function and tool requirements are now auto-installed at boot, ensuring tools always “just work.”
+- 🔒 **Automatic Sign-Out on Token Expiry**: Security is smarter—users are now automatically logged out if their authentication token expires, protecting sensitive content and ensuring compliance without disruption.
+- 🎬 **Automatic YouTube Embed Detection**: Paste YouTube links and see instant in-chat video embeds—no more manual embedding, making knowledge sharing and media consumption even easier for every team.
+- 🔄 **Expanded Language & Locale Support**: Translations for Danish, French, Russian, Traditional Chinese, Simplified Chinese, Thai, Catalan, German, and Korean have been upgraded, offering smoother, more natural user experiences across the platform.
+
+### Fixed
+
+- 🔒 **Tighter HTML Token Security**: HTML rendering is now restricted to admin-uploaded tokens only, reducing any risk of XSS and keeping your data safe.
+- 🔐 **Refined HTML Security and Token Handling**: Further hardened how HTML tokens and content are handled, guaranteeing even stronger resistance to security vulnerabilities and attacks.
+- 🔏 **Correct Model Usage with Ollama Proxy Prefixes**: Enhanced model reference handling so proxied models in Ollama always download and run correctly—even when using custom prefixes.
+- 📥 **Video File Upload Handling**: Prevented video files from being misclassified as text, fixing bugs with uploads and ensuring media files work as expected.
+- 🔄 **No More Dependent WebSocket Sequential Delays**: Streamlined WebSocket operation to prevent delays and maintain snappy real-time collaboration, especially in multi-user environments.
+- 🛠️ **More Robust Action Module Execution**: Multiple actions in a module now trigger as designed, increasing automation and scripting flexibility.
+- 📧 **Notification Webhooks**: Ensured that notification webhooks are always sent for user events, even when the user isn’t currently active.
+- 🗂️ **Smarter Knowledge Base Reindexing**: Knowledge reindexing continues even when corrupt or missing collections are encountered, keeping your search features running reliably.
+- 🏷️ **User Import with Profile Images**: When importing users, their profile images now come along—making onboarding and collaboration visually clearer from day one.
+- 💬 **OpenAI o-Series Universal Support**: All OpenAI o-series models are now seamlessly recognized and supported, unlocking more advanced capabilities and model choices for every workflow.
+
+### Changed
+
+- 📜 **Custom License Update & Contributor Agreement**: Open WebUI now operates under a custom license with Contributor License Agreement required by default—see https://docs.openwebui.com/license/ for details, ensuring sustainable open innovation for the community.
+- 🔨 **CUDA Docker Images Updated to 12.8**: Upgraded CUDA image support for faster, more compatible model inference and futureproof GPU performance in your AI infrastructure.
+- 🧱 **General Backend Refactoring for Reliability**: Continuous stability improvements streamline backend logic, reduce errors, and lay a stronger foundation for the next wave of feature releases—all under the hood for a more dependable WebUI.
+
+## [0.6.5] - 2025-04-14
+
+### Added
+
+- 🛂 **Granular Voice Feature Permissions Per User Group**: Admins can now separately manage access to Speech-to-Text (record voice), Text-to-Speech (read aloud), and Tool Calls for each user group—giving teams tighter control over voice features and enhanced governance across roles.
+- 🗣️ **Toggle Voice Activity Detection (VAD) for Whisper STT**: New environment variable lets you enable/disable VAD filtering with built-in Whisper speech-to-text, giving you flexibility to optimize for different audio quality and response accuracy levels.
+- 📋 **Copy Formatted Response Mode**: You can now enable “Copy Formatted” in Settings > Interface to copy AI responses exactly as styled (with rich formatting, links, and structure preserved), making it faster and cleaner to paste into documents, emails, or reports.
+- ⚙️ **Backend Stability and Performance Enhancements**: General backend refactoring improves system resilience, consistency, and overall reliability—offering smoother performance across workflows whether chatting, generating media, or using external tools.
+- 🌎 **Translation Refinements Across Multiple Languages**: Updated translations deliver smoother language localization, clearer labels, and improved international usability throughout the UI—ensuring a better experience for non-English speakers.
+
+### Fixed
+
+- 🛠️ **LDAP Login Reliability Restored**: Resolved a critical issue where some LDAP setups failed due to attribute parsing—ensuring consistent, secure, and seamless user authentication across enterprise deployments.
+- 🖼️ **Image Generation in Temporary Chats Now Works Properly**: Fixed a bug where image outputs weren’t generated during temporary chats—visual content can now be used reliably in all chat modes without interruptions.
+
+## [0.6.4] - 2025-04-12
+
+### Fixed
+
+- 🛠️ **RAG_TEMPLATE Display Issue Resolved**: Fixed a formatting problem where the custom RAG_TEMPLATE wasn't correctly rendered in the interface—ensuring that custom retrieval prompts now appear exactly as intended for more reliable prompt engineering.
+
+## [0.6.3] - 2025-04-12
+
+### Added
+
+- 🧪 **Auto-Artifact Detection Toggle**: Automatically detects artifacts in results—but now you can disable this behavior under advanced settings for full control.
+- 🖼️ **Widescreen Mode for Shared Chats**: Shared link conversations now support widescreen layouts—perfect for presentations or easier review across wider displays.
+- 🔁 **Reindex Knowledge Files on Demand**: Admins can now trigger reindexing of all knowledge files after changing embeddings—ensuring immediate alignment with new models for optimal RAG performance.
+- 📄 **OpenAPI YAML Format Support**: External tools can now use YAML-format OpenAPI specs—making integration simpler for developers familiar with YAML-based configurations.
+- 💬 **Message Content Copy Behavior**: Copy action now excludes 'details' tags—streamlining clipboard content when sharing or pasting summaries elsewhere.
+- 🧭 **Sougou Web Search Integration**: New search engine option added—enhancing global relevance and diversity of search sources for multilingual users.
+- 🧰 **Frontend Web Loader Engine Configuration**: Admins can now set preferred web loader engine for RAG workflows directly from the frontend—offering more control across setups.
+- 👥 **Multi-Model Chat Permission Control**: Admins can manage access to multi-model chats per user group—allowing tighter governance in team environments.
+- 🧱 **Persistent Configuration Can Be Disabled**: New environment variable lets advanced users and hosts turn off persistent configs—ideal for volatile or stateless deployments.
+- 🧠 **Elixir Code Highlighting Support**: Elixir syntax is now beautifully rendered in code blocks—perfect for developers using this language in AI or automation projects.
+- 🌐 **PWA External Manifest URL Support**: You can now define an external manifest.json—integrate Open WebUI seamlessly in managed or proxy-based PWA environments like Cloudflare Zero Trust.
+- 🧪 **Azure AI Speech-to-Text Provider Integration**: Easily transcribe large audio files (up to 200MB) with high accuracy using Microsoft's Azure STT—fully configurable in Audio Settings.
+- 🔏 **PKCE (Code Challenge Method) Support for OIDC**: Enhance your OIDC login security with Proof Key for Code Exchange—ideal for zero-trust and native client apps.
+- ✨ **General UI/UX Enhancements**: Numerous refinements across layout, styling, and tool interactions—reducing visual noise and improving overall usability across key workflows.
+- 🌍 **Translation Updates Across Multiple Languages**: Refined Catalan, Russian, Chinese (Simplified & Traditional), Hungarian, and Spanish translations for clearer navigation and instructions globally.
+
+### Fixed
+
+- 💥 **Chat Completion Error with Missing Models Resolved**: Fixed internal server error when referencing a model that doesn’t exist—ensuring graceful fallback and clear error guidance.
+- 🔧 **Correct Knowledge Base Citations Restored**: Citations generated by RAG workflows now show accurate references—ensuring verifiability in outputs from sourced content.
+- 🎙️ **Broken OGG/WebM Audio Upload Handling for OpenAI Fixed**: Uploading OGG or WebM files now converts properly to WAV before transcription—restoring accurate AI speech recognition workflows.
+- 🔐 **Tool Server 'Session' Authentication Restored**: Previously broken session auth on external tool servers is now fully functional—ensuring secure and seamless access to connected tools.
+- 🌐 **Folder-Based Chat Rename Now Updates Correctly**: Renaming chats in folders now reflects instantly everywhere—improving chat organization and clarity.
+- 📜 **KaTeX Overflow Displays Fixed**: Math expressions now stay neatly within message bounds—preserving layout consistency even with long formulas.
+- 🚫 **Stopping Ongoing Chat Fixed**: You can now return to an active (ongoing) chat and stop generation at any time—ensuring full control over sessions.
+- 🔧 **TOOL_SERVERS / TOOL_SERVER_CONNECTIONS Indexing Issue Fixed**: Fixed a mismatch between tool lists and their access paths—restoring full function and preventing confusion in tool management.
+- 🔐 **LDAP Login Handles Multiple Emails**: When LDAP returns multiple email attributes, the first valid one is now used—ensuring login success and account consistency.
+- 🧩 **Model Visibility Toggle Fix**: Toggling model visibility now works even for untouched models—letting admins smoothly manage user access across base models.
+- ⚙️ **Cross-Origin manifest.json Now Loads Properly**: Compatibility issues with Cloudflare Zero Trust (and others) resolved, allowing manifest.json to load behind authenticated proxies.
+
+### Changed
+
+- 🔒 **Default Access Scopes Set to Private for All Resources**: Models, tools, and knowledge are now private by default when created—ensuring better baseline security and visibility controls.
+- 🧱 **General Backend Refactoring for Stability**: Numerous invisible improvements enhance backend scalability, security, and maintainability—powering upcoming features with a stronger foundation.
+- 🧩 **Stable Dependency Upgrades**: Updated key platform libraries—Chromadb (0.6.3), pgvector (0.4.0), Azure Identity (1.21.0), and Youtube Transcript API (1.0.3)—for improved compatibility, functionality, and security.
+
 ## [0.6.2] - 2025-04-06
 
 ### Added

CODE_OF_CONDUCT.md

@@ -2,13 +2,13 @@
 
 ## Our Pledge
 
-As members, contributors, and leaders of this community, we pledge to make participation in our open-source project a harassment-free experience for everyone, regardless of age, body size, visible or invisible disability, ethnicity, sex characteristics, gender identity and expression, level of experience, education, socioeconomic status, nationality, personal appearance, race, religion, or sexual identity and orientation.
+As members, contributors, and leaders of this community, we pledge to make participation in our project a harassment-free experience for everyone, regardless of age, body size, visible or invisible disability, ethnicity, sex characteristics, gender identity and expression, level of experience, education, socioeconomic status, nationality, personal appearance, race, religion, or sexual identity and orientation.
 
 We are committed to creating and maintaining an open, respectful, and professional environment where positive contributions and meaningful discussions can flourish. By participating in this project, you agree to uphold these values and align your behavior to the standards outlined in this Code of Conduct.
 
 ## Why These Standards Are Important
 
-Open-source projects rely on a community of volunteers dedicating their time, expertise, and effort toward a shared goal. These projects are inherently collaborative but also fragile, as the success of the project depends on the goodwill, energy, and productivity of those involved.
+Projects rely on a community of volunteers dedicating their time, expertise, and effort toward a shared goal. These projects are inherently collaborative but also fragile, as the success of the project depends on the goodwill, energy, and productivity of those involved.
 
 Maintaining a positive and respectful environment is essential to safeguarding the integrity of this project and protecting contributors' efforts. Behavior that disrupts this atmosphere—whether through hostility, entitlement, or unprofessional conduct—can severely harm the morale and productivity of the community. **Strict enforcement of these standards ensures a safe and supportive space for meaningful collaboration.**
 
@@ -79,7 +79,7 @@ This approach ensures that disruptive behaviors are addressed swiftly and decisi
 
 ## Why Zero Tolerance Is Necessary
 
-Open-source projects thrive on collaboration, goodwill, and mutual respect. Toxic behaviors—such as entitlement, hostility, or persistent negativity—threaten not just individual contributors but the health of the project as a whole. Allowing such behaviors to persist robs contributors of their time, energy, and enthusiasm for the work they do.
+Projects thrive on collaboration, goodwill, and mutual respect. Toxic behaviors—such as entitlement, hostility, or persistent negativity—threaten not just individual contributors but the health of the project as a whole. Allowing such behaviors to persist robs contributors of their time, energy, and enthusiasm for the work they do.
 
 By enforcing a zero-tolerance policy, we ensure that the community remains a safe, welcoming space for all participants. These measures are not about harshness—they are about protecting contributors and fostering a productive environment where innovation can thrive.
 

CONTRIBUTOR_LICENSE_AGREEMENT

@@ -0,0 +1,7 @@
+# Open WebUI Contributor License Agreement
+
+By submitting my contributions to Open WebUI, I grant Open WebUI full freedom to use my work in any way they choose, under any terms they like, both now and in the future. This approach helps ensure the project remains unified, flexible, and easy to maintain, while empowering Open WebUI to respond quickly to the needs of its users and the wider community.
+
+Taking part in this process means my work can be seamlessly integrated and combined with others, ensuring longevity and adaptability for everyone who benefits from the Open WebUI project. This collaborative approach strengthens the project’s future and helps guarantee that improvements can always be shared and distributed in the most effective way possible.
+
+**_To the fullest extent permitted by law, my contributions are provided on an “as is” basis, with no warranties or guarantees of any kind, and I disclaim any liability for any issues or damages arising from their use or incorporation into the project, regardless of the type of legal claim._**

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

@@ -4,7 +4,7 @@
 ARG USE_CUDA=false
 ARG USE_OLLAMA=false
 # Tested with cu117 for CUDA 11 and cu121 for CUDA 12 (default)
-ARG USE_CUDA_VER=cu121
+ARG USE_CUDA_VER=cu128
 # any sentence transformer model; models to use can be found at https://huggingface.co/models?library=sentence-transformers
 # Leaderboard: https://huggingface.co/spaces/mteb/leaderboard 
 # for better performance and multilangauge support use "intfloat/multilingual-e5-large" (~2.5GB) or "intfloat/multilingual-e5-base" (~1.5GB)
@@ -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
 

LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2023-2025 Timothy Jaeryang Baek
+Copyright (c) 2023-2025 Timothy Jaeryang Baek (Open WebUI)
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -15,6 +15,12 @@ modification, are permitted provided that the following conditions are met:
    contributors may be used to endorse or promote products derived from
    this software without specific prior written permission.
 
+4. Notwithstanding any other provision of this License, and as a material condition of the rights granted herein, licensees are strictly prohibited from altering, removing, obscuring, or replacing any "Open WebUI" branding, including but not limited to the name, logo, or any visual, textual, or symbolic identifiers that distinguish the software and its interfaces, in any deployment or distribution, regardless of the number of users, except as explicitly set forth in Clauses 5 and 6 below.
+
+5. The branding restriction enumerated in Clause 4 shall not apply in the following limited circumstances: (i) deployments or distributions where the total number of end users (defined as individual natural persons with direct access to the application) does not exceed fifty (50) within any rolling thirty (30) day period; (ii) cases in which the licensee is an official contributor to the codebase—with a substantive code change successfully merged into the main branch of the official codebase maintained by the copyright holder—who has obtained specific prior written permission for branding adjustment from the copyright holder; or (iii) where the licensee has obtained a duly executed enterprise license expressly permitting such modification. For all other cases, any removal or alteration of the "Open WebUI" branding shall constitute a material breach of license.
+
+6. All code, modifications, or derivative works incorporated into this project prior to the incorporation of this branding clause remain licensed under the BSD 3-Clause License, and prior contributors retain all BSD-3 rights therein; if any such contributor requests the removal of their BSD-3-licensed code, the copyright holder will do so, and any replacement code will be licensed under the project's primary license then in effect. By contributing after this clause's adoption, you agree to the project's Contributor License Agreement (CLA) and to these updated terms for all new contributions.
+
 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE

README.md

@@ -7,7 +7,6 @@
 ![GitHub language count](https://img.shields.io/github/languages/count/open-webui/open-webui)
 ![GitHub top language](https://img.shields.io/github/languages/top/open-webui/open-webui)
 ![GitHub last commit](https://img.shields.io/github/last-commit/open-webui/open-webui?color=red)
-![Hits](https://hits.seeyoufarm.com/api/count/incr/badge.svg?url=https%3A%2F%2Fgithub.com%2Follama-webui%2Follama-wbui&count_bg=%2379C83D&title_bg=%23555555&icon=&icon_color=%23E7E7E7&title=hits&edge_flat=false)
 [![Discord](https://img.shields.io/badge/Discord-Open_WebUI-blue?logo=discord&logoColor=white)](https://discord.gg/5rJgQTnV4s)
 [![](https://img.shields.io/static/v1?label=Sponsor&message=%E2%9D%A4&logo=GitHub&color=%23fe8e86)](https://github.com/sponsors/tjbck)
 
@@ -62,9 +61,46 @@ For more information, be sure to check out our [Open WebUI Documentation](https:
 
 Want to learn more about Open WebUI's features? Check out our [Open WebUI documentation](https://docs.openwebui.com/features) for a comprehensive overview!
 
-## 🔗 Also Check Out Open WebUI Community!
+## Sponsors 🙌
 
-Don't forget to explore our sibling project, [Open WebUI Community](https://openwebui.com/), where you can discover, download, and explore customized Modelfiles. Open WebUI Community offers a wide range of exciting possibilities for enhancing your chat interactions with Open WebUI! 🚀
+#### Emerald
+
+<table>
+  <tr>
+    <td>
+      <a href="https://n8n.io/" target="_blank">
+        <img src="https://docs.openwebui.com/sponsors/logos/n8n.png" alt="n8n" style="width: 8rem; height: 8rem; border-radius: .75rem;" />
+      </a>
+    </td>
+    <td>
+      <a href="https://n8n.io/">n8n</a> • Does your interface have a backend yet?<br>Try <a href="https://n8n.io/">n8n</a>
+    </td>
+  </tr>
+  <tr>
+    <td>
+      <a href="https://warp.dev/open-webui" target="_blank">
+        <img src="https://docs.openwebui.com/sponsors/logos/warp.png" alt="Warp" style="width: 8rem; height: 8rem; border-radius: .75rem;" />
+      </a>
+    </td>
+    <td>
+      <a href="https://warp.dev/open-webui">Warp</a> • The intelligent terminal for developers
+    </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>
+
+---
+
+We are incredibly grateful for the generous support of our sponsors. Their contributions help us to maintain and improve our project, ensuring we can continue to deliver quality work to our community. Thank you!
 
 ## How to Install 🚀
 
@@ -155,6 +191,8 @@ 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).
@@ -206,7 +244,7 @@ Discover upcoming features on our roadmap in the [Open WebUI Documentation](http
 
 ## License 📜
 
-This project is licensed under the [BSD-3-Clause License](LICENSE) - see the [LICENSE](LICENSE) file for details. 📄
+This project is licensed under the [Open WebUI License](LICENSE), a revised BSD-3-Clause license. You receive all the same rights as the classic BSD-3 license: you can use, modify, and distribute the software, including in proprietary and commercial products, with minimal restrictions. The only additional requirement is to preserve the "Open WebUI" branding, as detailed in the LICENSE file. For full terms, see the [LICENSE](LICENSE) document. 📄
 
 ## Support 💬
 

backend/open_webui/__init__.py

@@ -73,8 +73,15 @@ def serve(
             os.environ["LD_LIBRARY_PATH"] = ":".join(LD_LIBRARY_PATH)
 
     import open_webui.main  # we need set environment variables before importing main
+    from open_webui.env import UVICORN_WORKERS  # Import the workers setting
 
-    uvicorn.run(open_webui.main.app, host=host, port=port, forwarded_allow_ips="*")
+    uvicorn.run(
+        "open_webui.main:app",
+        host=host,
+        port=port,
+        forwarded_allow_ips="*",
+        workers=UVICORN_WORKERS,
+    )
 
 
 @app.command()

backend/open_webui/constants.py

@@ -31,6 +31,7 @@ class ERROR_MESSAGES(str, Enum):
     USERNAME_TAKEN = (
         "Uh-oh! This username is already registered. Please choose another username."
     )
+    PASSWORD_TOO_LONG = "Uh-oh! The password you entered is too long. Please make sure your password is less than 72 bytes long."
     COMMAND_TAKEN = "Uh-oh! This command is already registered. Please choose another command string."
     FILE_EXISTS = "Uh-oh! This file is already registered. Please choose another file."
 
@@ -110,6 +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
@@ -326,6 +328,20 @@ REDIS_URL = os.environ.get("REDIS_URL", "")
 REDIS_SENTINEL_HOSTS = os.environ.get("REDIS_SENTINEL_HOSTS", "")
 REDIS_SENTINEL_PORT = os.environ.get("REDIS_SENTINEL_PORT", "26379")
 
+####################################
+# UVICORN WORKERS
+####################################
+
+# Number of uvicorn worker processes for handling requests
+UVICORN_WORKERS = os.environ.get("UVICORN_WORKERS", "1")
+try:
+    UVICORN_WORKERS = int(UVICORN_WORKERS)
+    if UVICORN_WORKERS < 1:
+        UVICORN_WORKERS = 1
+except ValueError:
+    UVICORN_WORKERS = 1
+    log.info(f"Invalid UVICORN_WORKERS value, defaulting to {UVICORN_WORKERS}")
+
 ####################################
 # WEBUI_AUTH (Required for security)
 ####################################
@@ -335,11 +351,19 @@ 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"
 )
 
+WEBUI_AUTH_SIGNOUT_REDIRECT_URL = os.environ.get(
+    "WEBUI_AUTH_SIGNOUT_REDIRECT_URL", None
+)
+
 ####################################
 # WEBUI_SECRET_KEY
 ####################################
@@ -395,6 +419,11 @@ else:
     except Exception:
         AIOHTTP_CLIENT_TIMEOUT = 300
 
+
+AIOHTTP_CLIENT_SESSION_SSL = (
+    os.environ.get("AIOHTTP_CLIENT_SESSION_SSL", "True").lower() == "true"
+)
+
 AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = os.environ.get(
     "AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST",
     os.environ.get("AIOHTTP_CLIENT_TIMEOUT_OPENAI_MODEL_LIST", "10"),
@@ -408,6 +437,71 @@ else:
     except Exception:
         AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = 10
 
+
+AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = os.environ.get(
+    "AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA", "10"
+)
+
+if AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA == "":
+    AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = None
+else:
+    try:
+        AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = int(
+            AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA
+        )
+    except Exception:
+        AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = 10
+
+
+AIOHTTP_CLIENT_SESSION_TOOL_SERVER_SSL = (
+    os.environ.get("AIOHTTP_CLIENT_SESSION_TOOL_SERVER_SSL", "True").lower() == "true"
+)
+
+
+####################################
+# SENTENCE TRANSFORMERS
+####################################
+
+
+SENTENCE_TRANSFORMERS_BACKEND = os.environ.get("SENTENCE_TRANSFORMERS_BACKEND", "")
+if SENTENCE_TRANSFORMERS_BACKEND == "":
+    SENTENCE_TRANSFORMERS_BACKEND = "torch"
+
+
+SENTENCE_TRANSFORMERS_MODEL_KWARGS = os.environ.get(
+    "SENTENCE_TRANSFORMERS_MODEL_KWARGS", ""
+)
+if SENTENCE_TRANSFORMERS_MODEL_KWARGS == "":
+    SENTENCE_TRANSFORMERS_MODEL_KWARGS = None
+else:
+    try:
+        SENTENCE_TRANSFORMERS_MODEL_KWARGS = json.loads(
+            SENTENCE_TRANSFORMERS_MODEL_KWARGS
+        )
+    except Exception:
+        SENTENCE_TRANSFORMERS_MODEL_KWARGS = None
+
+
+SENTENCE_TRANSFORMERS_CROSS_ENCODER_BACKEND = os.environ.get(
+    "SENTENCE_TRANSFORMERS_CROSS_ENCODER_BACKEND", ""
+)
+if SENTENCE_TRANSFORMERS_CROSS_ENCODER_BACKEND == "":
+    SENTENCE_TRANSFORMERS_CROSS_ENCODER_BACKEND = "torch"
+
+
+SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS = os.environ.get(
+    "SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS", ""
+)
+if SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS == "":
+    SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS = None
+else:
+    try:
+        SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS = json.loads(
+            SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS
+        )
+    except Exception:
+        SENTENCE_TRANSFORMERS_CROSS_ENCODER_MODEL_KWARGS = None
+
 ####################################
 # OFFLINE_MODE
 ####################################
@@ -417,6 +511,7 @@ OFFLINE_MODE = os.environ.get("OFFLINE_MODE", "false").lower() == "true"
 if OFFLINE_MODE:
     os.environ["HF_HUB_OFFLINE"] = "1"
 
+
 ####################################
 # AUDIT LOGGING
 ####################################
@@ -438,6 +533,7 @@ AUDIT_EXCLUDED_PATHS = os.getenv("AUDIT_EXCLUDED_PATHS", "/chats,/chat,/folders"
 AUDIT_EXCLUDED_PATHS = [path.strip() for path in AUDIT_EXCLUDED_PATHS]
 AUDIT_EXCLUDED_PATHS = [path.lstrip("/") for path in AUDIT_EXCLUDED_PATHS]
 
+
 ####################################
 # OPENTELEMETRY
 ####################################
@@ -460,3 +556,10 @@ OTEL_TRACES_SAMPLER = os.environ.get(
 
 PIP_OPTIONS = os.getenv("PIP_OPTIONS", "").split()
 PIP_PACKAGE_INDEX_OPTIONS = os.getenv("PIP_PACKAGE_INDEX_OPTIONS", "").split()
+
+
+####################################
+# PROGRESSIVE WEB APP OPTIONS
+####################################
+
+EXTERNAL_PWA_MANIFEST_URL = os.environ.get("EXTERNAL_PWA_MANIFEST_URL")

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/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
@@ -17,7 +19,9 @@ from sqlalchemy import text
 from typing import Optional
 from aiocache import cached
 import aiohttp
+import anyio.to_thread
 import requests
+from redis import Redis
 
 
 from fastapi import (
@@ -36,9 +40,11 @@ from fastapi import (
 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
@@ -63,6 +69,7 @@ from open_webui.routers import (
     auths,
     channels,
     chats,
+    notes,
     folders,
     configs,
     groups,
@@ -100,11 +107,15 @@ from open_webui.config import (
     # OpenAI
     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,
+    # Thread pool size for FastAPI/AnyIO
+    THREAD_POOL_SIZE,
     # Tool Server Configs
     TOOL_SERVER_CONNECTIONS,
     # Code Execution
@@ -148,6 +159,11 @@ from open_webui.config import (
     AUDIO_STT_MODEL,
     AUDIO_STT_OPENAI_API_BASE_URL,
     AUDIO_STT_OPENAI_API_KEY,
+    AUDIO_STT_AZURE_API_KEY,
+    AUDIO_STT_AZURE_REGION,
+    AUDIO_STT_AZURE_LOCALES,
+    AUDIO_STT_AZURE_BASE_URL,
+    AUDIO_STT_AZURE_MAX_SPEAKERS,
     AUDIO_TTS_API_KEY,
     AUDIO_TTS_ENGINE,
     AUDIO_TTS_MODEL,
@@ -156,13 +172,16 @@ from open_webui.config import (
     AUDIO_TTS_SPLIT_ON,
     AUDIO_TTS_VOICE,
     AUDIO_TTS_AZURE_SPEECH_REGION,
+    AUDIO_TTS_AZURE_SPEECH_BASE_URL,
     AUDIO_TTS_AZURE_SPEECH_OUTPUT_FORMAT,
-    PLAYWRIGHT_WS_URI,
+    PLAYWRIGHT_WS_URL,
     PLAYWRIGHT_TIMEOUT,
     FIRECRAWL_API_BASE_URL,
     FIRECRAWL_API_KEY,
-    RAG_WEB_LOADER_ENGINE,
+    WEB_LOADER_ENGINE,
     WHISPER_MODEL,
+    WHISPER_VAD_FILTER,
+    WHISPER_LANGUAGE,
     DEEPGRAM_API_KEY,
     WHISPER_MODEL_AUTO_UPDATE,
     WHISPER_MODEL_DIR,
@@ -174,46 +193,76 @@ 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,
     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,
     YOUTUBE_LOADER_LANGUAGE,
     YOUTUBE_LOADER_PROXY_URL,
     # Retrieval (Web Search)
-    RAG_WEB_SEARCH_ENGINE,
+    ENABLE_WEB_SEARCH,
+    WEB_SEARCH_ENGINE,
     BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL,
-    RAG_WEB_SEARCH_RESULT_COUNT,
-    RAG_WEB_SEARCH_CONCURRENT_REQUESTS,
-    RAG_WEB_SEARCH_TRUST_ENV,
-    RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
+    BYPASS_WEB_SEARCH_WEB_LOADER,
+    WEB_SEARCH_RESULT_COUNT,
+    WEB_SEARCH_CONCURRENT_REQUESTS,
+    WEB_SEARCH_TRUST_ENV,
+    WEB_SEARCH_DOMAIN_FILTER_LIST,
     JINA_API_KEY,
     SEARCHAPI_API_KEY,
     SEARCHAPI_ENGINE,
     SERPAPI_API_KEY,
     SERPAPI_ENGINE,
     SEARXNG_QUERY_URL,
+    YACY_QUERY_URL,
+    YACY_USERNAME,
+    YACY_PASSWORD,
     SERPER_API_KEY,
     SERPLY_API_KEY,
     SERPSTACK_API_KEY,
@@ -225,6 +274,10 @@ 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,
     MOJEEK_SEARCH_API_KEY,
     BOCHA_SEARCH_API_KEY,
@@ -233,13 +286,18 @@ from open_webui.config import (
     GOOGLE_DRIVE_CLIENT_ID,
     GOOGLE_DRIVE_API_KEY,
     ONEDRIVE_CLIENT_ID,
+    ONEDRIVE_SHAREPOINT_URL,
+    ONEDRIVE_SHAREPOINT_TENANT_ID,
     ENABLE_RAG_HYBRID_SEARCH,
     ENABLE_RAG_LOCAL_WEB_FETCH,
-    ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
-    ENABLE_RAG_WEB_SEARCH,
+    ENABLE_WEB_LOADER_SSL_VERIFICATION,
     ENABLE_GOOGLE_DRIVE_INTEGRATION,
     ENABLE_ONEDRIVE_INTEGRATION,
     UPLOAD_DIR,
+    EXTERNAL_WEB_SEARCH_URL,
+    EXTERNAL_WEB_SEARCH_API_KEY,
+    EXTERNAL_WEB_LOADER_URL,
+    EXTERNAL_WEB_LOADER_API_KEY,
     # WebUI
     WEBUI_AUTH,
     WEBUI_NAME,
@@ -254,12 +312,15 @@ from open_webui.config import (
     ENABLE_API_KEY_ENDPOINT_RESTRICTIONS,
     API_KEY_ALLOWED_ENDPOINTS,
     ENABLE_CHANNELS,
+    ENABLE_NOTES,
     ENABLE_COMMUNITY_SHARING,
     ENABLE_MESSAGE_RATING,
     ENABLE_USER_WEBHOOKS,
     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,
@@ -286,6 +347,7 @@ from open_webui.config import (
     LDAP_APP_PASSWORD,
     LDAP_USE_TLS,
     LDAP_CA_CERT_FILE,
+    LDAP_VALIDATE_CERT,
     LDAP_CIPHERS,
     # Misc
     ENV,
@@ -296,6 +358,7 @@ from open_webui.config import (
     DEFAULT_LOCALE,
     OAUTH_PROVIDERS,
     WEBUI_URL,
+    RESPONSE_WATERMARK,
     # Admin
     ENABLE_ADMIN_CHAT_ACCESS,
     ENABLE_ADMIN_EXPORT,
@@ -304,10 +367,12 @@ 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,
@@ -329,17 +394,21 @@ from open_webui.env import (
     SAFE_MODE,
     SRC_LOG_LEVELS,
     VERSION,
+    INSTANCE_ID,
     WEBUI_BUILD_HASH,
     WEBUI_SECRET_KEY,
     WEBUI_SESSION_COOKIE_SAME_SITE,
     WEBUI_SESSION_COOKIE_SECURE,
     WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
     WEBUI_AUTH_TRUSTED_NAME_HEADER,
+    WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
     ENABLE_WEBSOCKET_SUPPORT,
     BYPASS_MODEL_ACCESS_CONTROL,
     RESET_CONFIG_ON_START,
     OFFLINE_MODE,
     ENABLE_OTEL,
+    EXTERNAL_PWA_MANIFEST_URL,
+    AIOHTTP_CLIENT_SESSION_SSL,
 )
 
 
@@ -353,6 +422,7 @@ 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
 
@@ -363,10 +433,17 @@ from open_webui.utils.auth import (
     get_admin_user,
     get_verified_user,
 )
+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 stop_task, list_tasks  # Import from tasks.py
+from open_webui.tasks import (
+    redis_task_command_listener,
+    list_task_ids_by_chat_id,
+    stop_task,
+    list_tasks,
+)  # Import from tasks.py
 
 from open_webui.utils.redis import get_sentinels_from_env
 
@@ -405,7 +482,7 @@ print(
  ╚═════╝ ╚═╝     ╚══════╝╚═╝  ╚═══╝     ╚══╝╚══╝ ╚══════╝╚═════╝  ╚═════╝ ╚═╝
 
 
-v{VERSION} - building the best open-source AI user interface.
+v{VERSION} - building the best AI user interface.
 {f"Commit: {WEBUI_BUILD_HASH}" if WEBUI_BUILD_HASH != "dev-build" else ""}
 https://github.com/open-webui/open-webui
 """
@@ -414,18 +491,47 @@ 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()
 
     if LICENSE_KEY:
         get_license_data(app, LICENSE_KEY)
 
+    # This should be blocking (sync) so functions are not deactivated on first /get_models calls
+    # when the first user lands on the / route.
+    log.info("Installing external dependencies of functions and tools...")
+    install_tool_and_function_dependencies()
+
+    app.state.redis = get_redis_connection(
+        redis_url=REDIS_URL,
+        redis_sentinels=get_sentinels_from_env(
+            REDIS_SENTINEL_HOSTS, REDIS_SENTINEL_PORT
+        ),
+        async_mode=True,
+    )
+
+    if app.state.redis is not None:
+        app.state.redis_task_command_listener = asyncio.create_task(
+            redis_task_command_listener(app)
+        )
+
+    if THREAD_POOL_SIZE and THREAD_POOL_SIZE > 0:
+        limiter = anyio.to_thread.current_default_thread_limiter()
+        limiter.total_tokens = THREAD_POOL_SIZE
+
     asyncio.create_task(periodic_usage_pool_cleanup())
+
     yield
 
+    if hasattr(app.state, "redis_task_command_listener"):
+        app.state.redis_task_command_listener.cancel()
+
 
 app = FastAPI(
+    title="Open WebUI",
     docs_url="/docs" if ENV == "dev" else None,
     openapi_url="/openapi.json" if ENV == "dev" else None,
     redoc_url=None,
@@ -434,10 +540,12 @@ 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),
 )
+app.state.redis = None
 
 app.state.WEBUI_NAME = WEBUI_NAME
 app.state.LICENSE_METADATA = None
@@ -524,6 +632,11 @@ 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
@@ -531,6 +644,7 @@ app.state.config.MODEL_ORDER_LIST = MODEL_ORDER_LIST
 
 
 app.state.config.ENABLE_CHANNELS = ENABLE_CHANNELS
+app.state.config.ENABLE_NOTES = ENABLE_NOTES
 app.state.config.ENABLE_COMMUNITY_SHARING = ENABLE_COMMUNITY_SHARING
 app.state.config.ENABLE_MESSAGE_RATING = ENABLE_MESSAGE_RATING
 app.state.config.ENABLE_USER_WEBHOOKS = ENABLE_USER_WEBHOOKS
@@ -559,15 +673,22 @@ 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
 
 
 app.state.AUTH_TRUSTED_EMAIL_HEADER = WEBUI_AUTH_TRUSTED_EMAIL_HEADER
 app.state.AUTH_TRUSTED_NAME_HEADER = WEBUI_AUTH_TRUSTED_NAME_HEADER
+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 = {}
 
 ########################################
 #
@@ -579,6 +700,8 @@ 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
 
@@ -586,13 +709,30 @@ app.state.config.FILE_MAX_COUNT = RAG_FILE_MAX_COUNT
 app.state.config.RAG_FULL_CONTEXT = RAG_FULL_CONTEXT
 app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL = BYPASS_EMBEDDING_AND_RETRIEVAL
 app.state.config.ENABLE_RAG_HYBRID_SEARCH = ENABLE_RAG_HYBRID_SEARCH
-app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION = (
-    ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION
-)
+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
@@ -606,12 +746,21 @@ 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
 
@@ -621,16 +770,24 @@ app.state.config.YOUTUBE_LOADER_LANGUAGE = YOUTUBE_LOADER_LANGUAGE
 app.state.config.YOUTUBE_LOADER_PROXY_URL = YOUTUBE_LOADER_PROXY_URL
 
 
-app.state.config.ENABLE_RAG_WEB_SEARCH = ENABLE_RAG_WEB_SEARCH
-app.state.config.RAG_WEB_SEARCH_ENGINE = RAG_WEB_SEARCH_ENGINE
+app.state.config.ENABLE_WEB_SEARCH = ENABLE_WEB_SEARCH
+app.state.config.WEB_SEARCH_ENGINE = WEB_SEARCH_ENGINE
+app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST = WEB_SEARCH_DOMAIN_FILTER_LIST
+app.state.config.WEB_SEARCH_RESULT_COUNT = WEB_SEARCH_RESULT_COUNT
+app.state.config.WEB_SEARCH_CONCURRENT_REQUESTS = WEB_SEARCH_CONCURRENT_REQUESTS
+app.state.config.WEB_LOADER_ENGINE = WEB_LOADER_ENGINE
+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.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST = RAG_WEB_SEARCH_DOMAIN_FILTER_LIST
+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
 app.state.config.SEARXNG_QUERY_URL = SEARXNG_QUERY_URL
+app.state.config.YACY_QUERY_URL = YACY_QUERY_URL
+app.state.config.YACY_USERNAME = YACY_USERNAME
+app.state.config.YACY_PASSWORD = YACY_PASSWORD
 app.state.config.GOOGLE_PSE_API_KEY = GOOGLE_PSE_API_KEY
 app.state.config.GOOGLE_PSE_ENGINE_ID = GOOGLE_PSE_ENGINE_ID
 app.state.config.BRAVE_SEARCH_API_KEY = BRAVE_SEARCH_API_KEY
@@ -651,12 +808,17 @@ 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
+app.state.config.EXTERNAL_WEB_SEARCH_API_KEY = EXTERNAL_WEB_SEARCH_API_KEY
+app.state.config.EXTERNAL_WEB_LOADER_URL = EXTERNAL_WEB_LOADER_URL
+app.state.config.EXTERNAL_WEB_LOADER_API_KEY = EXTERNAL_WEB_LOADER_API_KEY
 
-app.state.config.RAG_WEB_SEARCH_RESULT_COUNT = RAG_WEB_SEARCH_RESULT_COUNT
-app.state.config.RAG_WEB_SEARCH_CONCURRENT_REQUESTS = RAG_WEB_SEARCH_CONCURRENT_REQUESTS
-app.state.config.RAG_WEB_LOADER_ENGINE = RAG_WEB_LOADER_ENGINE
-app.state.config.RAG_WEB_SEARCH_TRUST_ENV = RAG_WEB_SEARCH_TRUST_ENV
-app.state.config.PLAYWRIGHT_WS_URI = PLAYWRIGHT_WS_URI
+
+app.state.config.PLAYWRIGHT_WS_URL = PLAYWRIGHT_WS_URL
 app.state.config.PLAYWRIGHT_TIMEOUT = PLAYWRIGHT_TIMEOUT
 app.state.config.FIRECRAWL_API_BASE_URL = FIRECRAWL_API_BASE_URL
 app.state.config.FIRECRAWL_API_KEY = FIRECRAWL_API_KEY
@@ -677,7 +839,10 @@ 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:
@@ -692,14 +857,27 @@ app.state.EMBEDDING_FUNCTION = get_embedding_function(
     (
         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
+        )
     ),
     (
         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,
+    azure_api_version=(
+        app.state.config.RAG_AZURE_OPENAI_API_VERSION
+        if app.state.config.RAG_EMBEDDING_ENGINE == "azure_openai"
+        else None
+    ),
 )
 
 ########################################
@@ -776,8 +954,15 @@ app.state.config.STT_ENGINE = AUDIO_STT_ENGINE
 app.state.config.STT_MODEL = AUDIO_STT_MODEL
 
 app.state.config.WHISPER_MODEL = WHISPER_MODEL
+app.state.config.WHISPER_VAD_FILTER = WHISPER_VAD_FILTER
 app.state.config.DEEPGRAM_API_KEY = DEEPGRAM_API_KEY
 
+app.state.config.AUDIO_STT_AZURE_API_KEY = AUDIO_STT_AZURE_API_KEY
+app.state.config.AUDIO_STT_AZURE_REGION = AUDIO_STT_AZURE_REGION
+app.state.config.AUDIO_STT_AZURE_LOCALES = AUDIO_STT_AZURE_LOCALES
+app.state.config.AUDIO_STT_AZURE_BASE_URL = AUDIO_STT_AZURE_BASE_URL
+app.state.config.AUDIO_STT_AZURE_MAX_SPEAKERS = AUDIO_STT_AZURE_MAX_SPEAKERS
+
 app.state.config.TTS_OPENAI_API_BASE_URL = AUDIO_TTS_OPENAI_API_BASE_URL
 app.state.config.TTS_OPENAI_API_KEY = AUDIO_TTS_OPENAI_API_KEY
 app.state.config.TTS_ENGINE = AUDIO_TTS_ENGINE
@@ -788,6 +973,7 @@ app.state.config.TTS_SPLIT_ON = AUDIO_TTS_SPLIT_ON
 
 
 app.state.config.TTS_AZURE_SPEECH_REGION = AUDIO_TTS_AZURE_SPEECH_REGION
+app.state.config.TTS_AZURE_SPEECH_BASE_URL = AUDIO_TTS_AZURE_SPEECH_BASE_URL
 app.state.config.TTS_AZURE_SPEECH_OUTPUT_FORMAT = AUDIO_TTS_AZURE_SPEECH_OUTPUT_FORMAT
 
 
@@ -812,6 +998,7 @@ 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
@@ -819,6 +1006,9 @@ 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
@@ -850,7 +1040,8 @@ class RedirectMiddleware(BaseHTTPMiddleware):
 
             # Check for the specific watch path and the presence of 'v' parameter
             if path.endswith("/watch") and "v" in query_params:
-                video_id = query_params["v"][0]  # Extract the first 'v' parameter
+                # Extract the first 'v' parameter
+                video_id = query_params["v"][0]
                 encoded_video_id = urlencode({"youtube": video_id})
                 redirect_url = f"/?{encoded_video_id}"
                 return RedirectResponse(url=redirect_url)
@@ -861,6 +1052,7 @@ class RedirectMiddleware(BaseHTTPMiddleware):
 
 
 # Add the middleware to the app
+app.add_middleware(CompressMiddleware)
 app.add_middleware(RedirectMiddleware)
 app.add_middleware(SecurityHeadersMiddleware)
 
@@ -936,6 +1128,8 @@ app.include_router(users.router, prefix="/api/v1/users", tags=["users"])
 
 app.include_router(channels.router, prefix="/api/v1/channels", tags=["channels"])
 app.include_router(chats.router, prefix="/api/v1/chats", tags=["chats"])
+app.include_router(notes.router, prefix="/api/v1/notes", tags=["notes"])
+
 
 app.include_router(models.router, prefix="/api/v1/models", tags=["models"])
 app.include_router(knowledge.router, prefix="/api/v1/knowledge", tags=["knowledge"])
@@ -1006,14 +1200,19 @@ async def get_models(request: Request, user=Depends(get_verified_user)):
         if "pipeline" in model and model["pipeline"].get("type", None) == "filter":
             continue
 
-        model_tags = [
-            tag.get("name")
-            for tag in model.get("info", {}).get("meta", {}).get("tags", [])
-        ]
-        tags = [tag.get("name") for tag in model.get("tags", [])]
+        try:
+            model_tags = [
+                tag.get("name")
+                for tag in model.get("info", {}).get("meta", {}).get("tags", [])
+            ]
+            tags = [tag.get("name") for tag in model.get("tags", [])]
 
-        tags = list(set(model_tags + tags))
-        model["tags"] = [{"name": tag} for tag in tags]
+            tags = list(set(model_tags + tags))
+            model["tags"] = [{"name": tag} for tag in tags]
+        except Exception as e:
+            log.debug(f"Error processing model tags: {e}")
+            model["tags"] = []
+            pass
 
         models.append(model)
 
@@ -1041,6 +1240,37 @@ 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,
@@ -1053,6 +1283,7 @@ async def chat_completion(
     model_item = form_data.pop("model_item", {})
     tasks = form_data.pop("background_tasks", None)
 
+    metadata = {}
     try:
         if not model_item.get("direct", False):
             model_id = form_data.get("model", None)
@@ -1080,11 +1311,12 @@ 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),
             **(
@@ -1108,13 +1340,15 @@ async def chat_completion(
 
     except Exception as e:
         log.debug(f"Error processing chat payload: {e}")
-        Chats.upsert_message_to_chat_by_id_and_message_id(
-            metadata["chat_id"],
-            metadata["message_id"],
-            {
-                "error": {"content": str(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,
@@ -1178,17 +1412,33 @@ 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)  # Use the function from tasks.py
+        result = await stop_task(request, 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()}  # Use the function from tasks.py
+async def list_tasks_endpoint(request: Request, user=Depends(get_verified_user)):
+    return {"tasks": await list_tasks(request)}
+
+
+@app.get("/api/tasks/chat/{chat_id}")
+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 = await list_task_ids_by_chat_id(request, chat_id)
+
+    print(f"Task IDs for chat {chat_id}: {task_ids}")
+    return {"task_ids": task_ids}
 
 
 ##################################
@@ -1244,7 +1494,8 @@ async def get_app_config(request: Request):
                 {
                     "enable_direct_connections": app.state.config.ENABLE_DIRECT_CONNECTIONS,
                     "enable_channels": app.state.config.ENABLE_CHANNELS,
-                    "enable_web_search": app.state.config.ENABLE_RAG_WEB_SEARCH,
+                    "enable_notes": app.state.config.ENABLE_NOTES,
+                    "enable_web_search": app.state.config.ENABLE_WEB_SEARCH,
                     "enable_code_execution": app.state.config.ENABLE_CODE_EXECUTION,
                     "enable_code_interpreter": app.state.config.ENABLE_CODE_INTERPRETER,
                     "enable_image_generation": app.state.config.ENABLE_IMAGE_GENERATION,
@@ -1288,7 +1539,16 @@ async def get_app_config(request: Request):
                     "client_id": GOOGLE_DRIVE_CLIENT_ID.value,
                     "api_key": GOOGLE_DRIVE_API_KEY.value,
                 },
-                "onedrive": {"client_id": ONEDRIVE_CLIENT_ID.value},
+                "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,
                 **(
                     {
@@ -1340,7 +1600,8 @@ async def get_app_latest_release_version(user=Depends(get_verified_user)):
         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()
@@ -1390,29 +1651,32 @@ async def oauth_callback(provider: str, request: Request, response: Response):
 
 @app.get("/manifest.json")
 async def get_manifest_json():
-    return {
-        "name": app.state.WEBUI_NAME,
-        "short_name": app.state.WEBUI_NAME,
-        "description": "Open WebUI is an open, extensible, user-friendly interface for AI that adapts to your workflow.",
-        "start_url": "/",
-        "display": "standalone",
-        "background_color": "#343541",
-        "orientation": "natural",
-        "icons": [
-            {
-                "src": "/static/logo.png",
-                "type": "image/png",
-                "sizes": "500x500",
-                "purpose": "any",
-            },
-            {
-                "src": "/static/logo.png",
-                "type": "image/png",
-                "sizes": "500x500",
-                "purpose": "maskable",
-            },
-        ],
-    }
+    if app.state.EXTERNAL_PWA_MANIFEST_URL:
+        return requests.get(app.state.EXTERNAL_PWA_MANIFEST_URL).json()
+    else:
+        return {
+            "name": app.state.WEBUI_NAME,
+            "short_name": app.state.WEBUI_NAME,
+            "description": "Open WebUI is an open, extensible, user-friendly interface for AI that adapts to your workflow.",
+            "start_url": "/",
+            "display": "standalone",
+            "background_color": "#343541",
+            "orientation": "any",
+            "icons": [
+                {
+                    "src": "/static/logo.png",
+                    "type": "image/png",
+                    "sizes": "500x500",
+                    "purpose": "any",
+                },
+                {
+                    "src": "/static/logo.png",
+                    "type": "image/png",
+                    "sizes": "500x500",
+                    "purpose": "maskable",
+                },
+            ],
+        }
 
 
 @app.get("/opensearch.xml")
@@ -1442,7 +1706,20 @@ 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/9f0c9cd09105_add_note_table.py

@@ -0,0 +1,33 @@
+"""Add note table
+
+Revision ID: 9f0c9cd09105
+Revises: 3781e22d8b01
+Create Date: 2025-05-03 03:00:00.000000
+
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "9f0c9cd09105"
+down_revision = "3781e22d8b01"
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    op.create_table(
+        "note",
+        sa.Column("id", sa.Text(), nullable=False, primary_key=True, unique=True),
+        sa.Column("user_id", sa.Text(), nullable=True),
+        sa.Column("title", sa.Text(), nullable=True),
+        sa.Column("data", sa.JSON(), nullable=True),
+        sa.Column("meta", sa.JSON(), nullable=True),
+        sa.Column("access_control", sa.JSON(), nullable=True),
+        sa.Column("created_at", sa.BigInteger(), nullable=True),
+        sa.Column("updated_at", sa.BigInteger(), nullable=True),
+    )
+
+
+def downgrade():
+    op.drop_table("note")

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

@@ -377,22 +377,47 @@ 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 +426,23 @@ 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)
@@ -436,7 +477,7 @@ class ChatTable:
 
             all_chats = query.all()
 
-            # result has to be destrctured from sqlalchemy `row` and mapped to a dict since the `ChatModel`is not the returned dataclass.
+            # result has to be destructured from sqlalchemy `row` and mapped to a dict since the `ChatModel`is not the returned dataclass.
             return [
                 ChatTitleIdResponse.model_validate(
                     {
@@ -542,7 +583,9 @@ class ChatTable:
         search_text = search_text.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(" ")
 

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 +207,43 @@ class GroupTable:
             except Exception:
                 return False
 
+    def sync_user_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/memories.py

@@ -63,14 +63,15 @@ class MemoriesTable:
             else:
                 return None
 
-    def update_memory_by_id(
+    def update_memory_by_id_and_user_id(
         self,
         id: str,
+        user_id: str,
         content: str,
     ) -> Optional[MemoryModel]:
         with get_db() as db:
             try:
-                db.query(Memory).filter_by(id=id).update(
+                db.query(Memory).filter_by(id=id, user_id=user_id).update(
                     {"content": content, "updated_at": int(time.time())}
                 )
                 db.commit()

backend/open_webui/models/notes.py

@@ -0,0 +1,135 @@
+import json
+import time
+import uuid
+from typing import Optional
+
+from open_webui.internal.db import Base, get_db
+from open_webui.utils.access_control import has_access
+from open_webui.models.users import Users, UserResponse
+
+
+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
+
+####################
+# Note DB Schema
+####################
+
+
+class Note(Base):
+    __tablename__ = "note"
+
+    id = Column(Text, primary_key=True)
+    user_id = Column(Text)
+
+    title = Column(Text)
+    data = Column(JSON, nullable=True)
+    meta = Column(JSON, nullable=True)
+
+    access_control = Column(JSON, nullable=True)
+
+    created_at = Column(BigInteger)
+    updated_at = Column(BigInteger)
+
+
+class NoteModel(BaseModel):
+    model_config = ConfigDict(from_attributes=True)
+
+    id: str
+    user_id: str
+
+    title: str
+    data: Optional[dict] = None
+    meta: Optional[dict] = None
+
+    access_control: Optional[dict] = None
+
+    created_at: int  # timestamp in epoch
+    updated_at: int  # timestamp in epoch
+
+
+####################
+# Forms
+####################
+
+
+class NoteForm(BaseModel):
+    title: str
+    data: Optional[dict] = None
+    meta: Optional[dict] = None
+    access_control: Optional[dict] = None
+
+
+class NoteUserResponse(NoteModel):
+    user: Optional[UserResponse] = None
+
+
+class NoteTable:
+    def insert_new_note(
+        self,
+        form_data: NoteForm,
+        user_id: str,
+    ) -> Optional[NoteModel]:
+        with get_db() as db:
+            note = NoteModel(
+                **{
+                    "id": str(uuid.uuid4()),
+                    "user_id": user_id,
+                    **form_data.model_dump(),
+                    "created_at": int(time.time_ns()),
+                    "updated_at": int(time.time_ns()),
+                }
+            )
+
+            new_note = Note(**note.model_dump())
+
+            db.add(new_note)
+            db.commit()
+            return note
+
+    def get_notes(self) -> list[NoteModel]:
+        with get_db() as db:
+            notes = db.query(Note).order_by(Note.updated_at.desc()).all()
+            return [NoteModel.model_validate(note) for note in notes]
+
+    def get_notes_by_user_id(
+        self, user_id: str, permission: str = "write"
+    ) -> list[NoteModel]:
+        notes = self.get_notes()
+        return [
+            note
+            for note in notes
+            if note.user_id == user_id
+            or has_access(user_id, permission, note.access_control)
+        ]
+
+    def get_note_by_id(self, id: str) -> Optional[NoteModel]:
+        with get_db() as db:
+            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]:
+        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
+            note.updated_at = int(time.time_ns())
+
+            db.commit()
+            return NoteModel.model_validate(note) if note else None
+
+    def delete_note_by_id(self, id: str):
+        with get_db() as db:
+            db.query(Note).filter(Note.id == id).delete()
+            db.commit()
+            return True
+
+
+Notes = NoteTable()

backend/open_webui/models/users.py

@@ -10,6 +10,8 @@ from open_webui.models.groups import Groups
 
 from pydantic import BaseModel, ConfigDict
 from sqlalchemy import BigInteger, Column, String, Text
+from sqlalchemy import or_
+
 
 ####################
 # User DB Schema
@@ -67,6 +69,11 @@ class UserModel(BaseModel):
 ####################
 
 
+class UserListResponse(BaseModel):
+    users: list[UserModel]
+    total: int
+
+
 class UserResponse(BaseModel):
     id: str
     name: str
@@ -88,6 +95,7 @@ class UserRoleUpdateForm(BaseModel):
 
 
 class UserUpdateForm(BaseModel):
+    role: str
     name: str
     email: str
     profile_image_url: str
@@ -160,11 +168,63 @@ class UsersTable:
             return None
 
     def get_users(
-        self, skip: Optional[int] = None, limit: Optional[int] = None
-    ) -> list[UserModel]:
+        self,
+        filter: Optional[dict] = None,
+        skip: Optional[int] = None,
+        limit: Optional[int] = None,
+    ) -> UserListResponse:
         with get_db() as db:
+            query = db.query(User)
 
-            query = db.query(User).order_by(User.created_at.desc())
+            if filter:
+                query_key = filter.get("query")
+                if query_key:
+                    query = query.filter(
+                        or_(
+                            User.name.ilike(f"%{query_key}%"),
+                            User.email.ilike(f"%{query_key}%"),
+                        )
+                    )
+
+                order_by = filter.get("order_by")
+                direction = filter.get("direction")
+
+                if order_by == "name":
+                    if direction == "asc":
+                        query = query.order_by(User.name.asc())
+                    else:
+                        query = query.order_by(User.name.desc())
+                elif order_by == "email":
+                    if direction == "asc":
+                        query = query.order_by(User.email.asc())
+                    else:
+                        query = query.order_by(User.email.desc())
+
+                elif order_by == "created_at":
+                    if direction == "asc":
+                        query = query.order_by(User.created_at.asc())
+                    else:
+                        query = query.order_by(User.created_at.desc())
+
+                elif order_by == "last_active_at":
+                    if direction == "asc":
+                        query = query.order_by(User.last_active_at.asc())
+                    else:
+                        query = query.order_by(User.last_active_at.desc())
+
+                elif order_by == "updated_at":
+                    if direction == "asc":
+                        query = query.order_by(User.updated_at.asc())
+                    else:
+                        query = query.order_by(User.updated_at.desc())
+                elif order_by == "role":
+                    if direction == "asc":
+                        query = query.order_by(User.role.asc())
+                    else:
+                        query = query.order_by(User.role.desc())
+
+            else:
+                query = query.order_by(User.created_at.desc())
 
             if skip:
                 query = query.offset(skip)
@@ -172,8 +232,10 @@ class UsersTable:
                 query = query.limit(limit)
 
             users = query.all()
-
-            return [UserModel.model_validate(user) for user in users]
+            return {
+                "users": [UserModel.model_validate(user) for user in users],
+                "total": db.query(User).count(),
+            }
 
     def get_users_by_user_ids(self, user_ids: list[str]) -> list[UserModel]:
         with get_db() as db:
@@ -308,7 +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})
@@ -330,5 +392,13 @@ class UsersTable:
             users = db.query(User).filter(User.id.in_(user_ids)).all()
             return [user.id for user in users]
 
+    def get_super_admin_user(self) -> Optional[UserModel]:
+        with get_db() as db:
+            user = db.query(User).filter_by(role="admin").first()
+            if user:
+                return UserModel.model_validate(user)
+            else:
+                return None
+
 
 Users = UsersTable()

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 +1,58 @@
+import requests
+import logging
+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}"
+
+        url = self.url
+        if url.endswith("/"):
+            url = url[:-1]
+
+        r = requests.put(f"{url}/process", data=data, headers=headers)
+
+        if r.ok:
+            res = r.json()
+
+            if res:
+                return [
+                    Document(
+                        page_content=res.get("page_content"),
+                        metadata=res.get("metadata"),
+                    )
+                ]
+            else:
+                raise Exception("Error loading document: No content returned")
+        else:
+            raise Exception(f"Error loading document: {r.status_code} {r.text}")

backend/open_webui/retrieval/loaders/external_web.py

@@ -0,0 +1,53 @@
+import requests
+import logging
+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 ExternalWebLoader(BaseLoader):
+    def __init__(
+        self,
+        web_paths: Union[str, List[str]],
+        external_url: str,
+        external_api_key: str,
+        continue_on_failure: bool = True,
+        **kwargs,
+    ) -> None:
+        self.external_url = external_url
+        self.external_api_key = external_api_key
+        self.urls = web_paths if isinstance(web_paths, list) else [web_paths]
+        self.continue_on_failure = continue_on_failure
+
+    def lazy_load(self) -> Iterator[Document]:
+        batch_size = 20
+        for i in range(0, len(self.urls), batch_size):
+            urls = self.urls[i : i + batch_size]
+            try:
+                response = requests.post(
+                    self.external_url,
+                    headers={
+                        "User-Agent": "Open WebUI (https://github.com/open-webui/open-webui) External Web Loader",
+                        "Authorization": f"Bearer {self.external_api_key}",
+                    },
+                    json={
+                        "urls": urls,
+                    },
+                )
+                response.raise_for_status()
+                results = response.json()
+                for result in results:
+                    yield Document(
+                        page_content=result.get("page_content", ""),
+                        metadata=result.get("metadata", {}),
+                    )
+            except Exception as e:
+                if self.continue_on_failure:
+                    log.error(f"Error extracting content from batch {urls}: {e}")
+                else:
+                    raise e

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,
@@ -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",
@@ -85,11 +89,13 @@ known_source_ext = [
 
 
 class TikaLoader:
-    def __init__(self, url, file_path, mime_type=None):
+    def __init__(self, url, file_path, mime_type=None, extract_images=None):
         self.url = url
         self.file_path = file_path
         self.mime_type = mime_type
 
+        self.extract_images = extract_images
+
     def load(self) -> list[Document]:
         with open(self.file_path, "rb") as f:
             data = f.read()
@@ -99,6 +105,9 @@ class TikaLoader:
         else:
             headers = {}
 
+        if self.extract_images == True:
+            headers["X-Tika-PDFextractInlineImages"] = "true"
+
         endpoint = self.url
         if not endpoint.endswith("/"):
             endpoint += "/"
@@ -121,11 +130,13 @@ class TikaLoader:
 
 
 class DoclingLoader:
-    def __init__(self, url, file_path=None, mime_type=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.params = params or {}
+
     def load(self) -> list[Document]:
         with open(self.file_path, "rb") as f:
             files = {
@@ -136,10 +147,39 @@ class DoclingLoader:
                 )
             }
 
-            params = {
-                "image_export_mode": "placeholder",
-                "table_mode": "accurate",
-            }
+            params = {"image_export_mode": "placeholder", "table_mode": "accurate"}
+
+            if self.params:
+                if self.params.get("do_picture_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"] = self.params.get(
+                            "picture_description_local", {}
+                        )
+
+                    elif picture_description_mode == "api" and self.params.get(
+                        "picture_description_api", {}
+                    ):
+                        params["picture_description_api"] = 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)
@@ -192,7 +232,18 @@ class Loader:
     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:
@@ -200,15 +251,69 @@ class Loader:
                     url=self.kwargs.get("TIKA_SERVER_URL"),
                     file_path=file_path,
                     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,
+                    params=params,
                 )
         elif (
             self.engine == "document_intelligence"
@@ -240,6 +345,15 @@ 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(

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 +388,217 @@ 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
+        ) 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 +610,30 @@ 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 +644,124 @@ 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
@@ -62,12 +63,17 @@ class YoutubeLoader:
         _video_id = _parse_video_id(video_id)
         self.video_id = _video_id if _video_id is not None else video_id
         self._metadata = {"source": video_id}
-        self.language = language
         self.proxy_url = proxy_url
+
+        # Ensure language is a list
         if isinstance(language, str):
             self.language = [language]
         else:
-            self.language = language
+            self.language = list(language)
+
+        # Add English as fallback if not already in the list
+        if "en" not in self.language:
+            self.language.append("en")
 
     def load(self) -> List[Document]:
         """Load YouTube transcripts into `Document` objects."""
@@ -88,7 +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
@@ -101,17 +106,55 @@ class YoutubeLoader:
             log.exception("Loading YouTube transcript failed")
             return []
 
-        try:
-            transcript = transcript_list.find_transcript(self.language)
-        except NoTranscriptFound:
-            transcript = transcript_list.find_transcript(["en"])
+        # Try each language in order of priority
+        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
 
-        transcript_pieces: List[Dict[str, Any]] = transcript.fetch()
+                log.debug(f"Found transcript for language '{lang}'")
+                try:
+                    transcript_pieces: List[Dict[str, Any]] = transcript.fetch()
+                except ParseError:
+                    log.debug(f"Empty or invalid transcript for language '{lang}'")
+                    continue
 
-        transcript = " ".join(
-            map(
-                lambda transcript_piece: transcript_piece["text"].strip(" "),
-                transcript_pieces,
-            )
+                if not transcript_pieces:
+                    log.debug(f"Empty transcript for language '{lang}'")
+                    continue
+
+                transcript_text = " ".join(
+                    map(
+                        lambda transcript_piece: (
+                            transcript_piece.text.strip(" ")
+                            if hasattr(transcript_piece, "text")
+                            else ""
+                        ),
+                        transcript_pieces,
+                    )
+                )
+                return [Document(page_content=transcript_text, metadata=self._metadata)]
+            except NoTranscriptFound:
+                log.debug(f"No transcript found for language '{lang}'")
+                continue
+            except Exception as e:
+                log.info(f"Error finding transcript for language '{lang}'")
+                raise e
+
+        # If we get here, all languages failed
+        languages_tried = ", ".join(self.language)
+        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."
         )
-        return [Document(page_content=transcript, metadata=self._metadata)]
+        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 +1,60 @@
+import logging
+import requests
+from typing import Optional, List, Tuple
+
+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 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]]) -> 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}",
+                },
+                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

@@ -5,6 +5,7 @@ from typing import Optional, Union
 import requests
 import hashlib
 from concurrent.futures import ThreadPoolExecutor
+import time
 
 from huggingface_hub import snapshot_download
 from langchain.retrievers import ContextualCompressionRetriever, EnsembleRetriever
@@ -12,7 +13,7 @@ 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
@@ -77,6 +78,7 @@ def query_doc(
     collection_name: str, query_embedding: list[float], k: int, user: UserModel = None
 ):
     try:
+        log.debug(f"query_doc:doc {collection_name}")
         result = VECTOR_DB_CLIENT.search(
             collection_name=collection_name,
             vectors=[query_embedding],
@@ -94,6 +96,7 @@ def query_doc(
 
 def get_doc(collection_name: str, user: UserModel = None):
     try:
+        log.debug(f"get_doc:doc {collection_name}")
         result = VECTOR_DB_CLIENT.get(collection_name=collection_name)
 
         if result:
@@ -114,8 +117,10 @@ 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}")
         bm25_retriever = BM25Retriever.from_texts(
             texts=collection_result.documents[0],
             metadatas=collection_result.metadatas[0],
@@ -128,9 +133,20 @@ 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,
@@ -168,6 +184,7 @@ def query_doc_with_hybrid_search(
         )
         return result
     except Exception as e:
+        log.exception(f"Error querying doc {collection_name} with hybrid search: {e}")
         raise e
 
 
@@ -203,7 +220,7 @@ def merge_and_sort_query_results(query_results: list[dict], k: int) -> dict:
 
         for distance, document, metadata in zip(distances, documents, metadatas):
             if isinstance(document, str):
-                doc_hash = hashlib.md5(
+                doc_hash = hashlib.sha256(
                     document.encode()
                 ).hexdigest()  # Compute a hash for uniqueness
 
@@ -256,22 +273,47 @@ def query_collection(
     k: int,
 ) -> dict:
     results = []
-    for query in queries:
-        query_embedding = embedding_function(query, prefix=RAG_EMBEDDING_QUERY_PREFIX)
-        for collection_name in collection_names:
+    error = False
+
+    def process_query_collection(collection_name, query_embedding):
+        try:
             if collection_name:
-                try:
-                    result = query_doc(
-                        collection_name=collection_name,
-                        k=k,
-                        query_embedding=query_embedding,
-                    )
-                    if result is not None:
-                        results.append(result.model_dump())
-                except Exception as e:
-                    log.exception(f"Error when querying the collection: {e}")
-            else:
-                pass
+                result = query_doc(
+                    collection_name=collection_name,
+                    k=k,
+                    query_embedding=query_embedding,
+                )
+                if result is not None:
+                    return result.model_dump(), None
+            return None, None
+        except Exception as e:
+            log.exception(f"Error when querying the collection: {e}")
+            return None, e
+
+    # Generate all query embeddings (in one call)
+    query_embeddings = embedding_function(queries, prefix=RAG_EMBEDDING_QUERY_PREFIX)
+    log.debug(
+        f"query_collection: processing {len(queries)} queries across {len(collection_names)} collections"
+    )
+
+    with ThreadPoolExecutor() as executor:
+        future_results = []
+        for query_embedding in query_embeddings:
+            for collection_name in collection_names:
+                result = executor.submit(
+                    process_query_collection, collection_name, query_embedding
+                )
+                future_results.append(result)
+        task_results = [future.result() for future in future_results]
+
+    for result, err in task_results:
+        if err is not None:
+            error = True
+        elif result is not None:
+            results.append(result)
+
+    if error and not results:
+        log.warning("All collection queries failed. No results returned.")
 
     return merge_and_sort_query_results(results, k=k)
 
@@ -284,6 +326,7 @@ def query_collection_with_hybrid_search(
     reranking_function,
     k_reranker: int,
     r: float,
+    hybrid_bm25_weight: float,
 ) -> dict:
     results = []
     error = False
@@ -292,6 +335,9 @@ def query_collection_with_hybrid_search(
     collection_results = {}
     for collection_name in collection_names:
         try:
+            log.debug(
+                f"query_collection_with_hybrid_search:VECTOR_DB_CLIENT.get:collection {collection_name}"
+            )
             collection_results[collection_name] = VECTOR_DB_CLIENT.get(
                 collection_name=collection_name
             )
@@ -314,6 +360,7 @@ 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:
@@ -354,12 +401,13 @@ 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,
@@ -368,6 +416,7 @@ def get_embedding_function(
             url=url,
             key=key,
             user=user,
+            azure_api_version=azure_api_version,
         )
 
         def generate_multiple(query, prefix, user, func):
@@ -401,6 +450,7 @@ def get_sources_from_files(
     reranking_function,
     k_reranker,
     r,
+    hybrid_bm25_weight,
     hybrid_search,
     full_context=False,
 ):
@@ -518,6 +568,7 @@ def get_sources_from_files(
                                     reranking_function=reranking_function,
                                     k_reranker=k_reranker,
                                     r=r,
+                                    hybrid_bm25_weight=hybrid_bm25_weight,
                                 )
                             except Exception as e:
                                 log.debug(
@@ -613,6 +664,9 @@ def generate_openai_batch_embeddings(
     user: UserModel = None,
 ) -> Optional[list[list[float]]]:
     try:
+        log.debug(
+            f"generate_openai_batch_embeddings:model {model} batch size: {len(texts)}"
+        )
         json_data = {"input": texts, "model": model}
         if isinstance(RAG_EMBEDDING_PREFIX_FIELD_NAME, str) and isinstance(prefix, str):
             json_data[RAG_EMBEDDING_PREFIX_FIELD_NAME] = prefix
@@ -646,6 +700,60 @@ 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": 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 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],
@@ -655,6 +763,9 @@ def generate_ollama_batch_embeddings(
     user: UserModel = None,
 ) -> Optional[list[list[float]]]:
     try:
+        log.debug(
+            f"generate_ollama_batch_embeddings:model {model} batch size: {len(texts)}"
+        )
         json_data = {"input": texts, "model": model}
         if isinstance(RAG_EMBEDDING_PREFIX_FIELD_NAME, str) and isinstance(prefix, str):
             json_data[RAG_EMBEDDING_PREFIX_FIELD_NAME] = prefix
@@ -707,38 +818,33 @@ 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
 
 
@@ -780,7 +886,9 @@ 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

@@ -1,26 +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()
-else:
-    from open_webui.retrieval.vector.dbs.chroma import ChromaClient
-
-    VECTOR_DB_CLIENT = ChromaClient()

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

@@ -5,7 +5,12 @@ from chromadb.utils.batch_utils import create_batches
 
 from typing import Optional
 
-from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
 from open_webui.config import (
     CHROMA_DATA_PATH,
     CHROMA_HTTP_HOST,
@@ -23,7 +28,7 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
-class ChromaClient:
+class ChromaClient(VectorDBBase):
     def __init__(self):
         settings_dict = {
             "allow_reset": True,

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

@@ -2,7 +2,12 @@ from elasticsearch import Elasticsearch, BadRequestError
 from typing import Optional
 import ssl
 from elasticsearch.helpers import bulk, scan
-from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
 from open_webui.config import (
     ELASTICSEARCH_URL,
     ELASTICSEARCH_CA_CERTS,
@@ -15,7 +20,7 @@ from open_webui.config import (
 )
 
 
-class ElasticsearchClient:
+class ElasticsearchClient(VectorDBBase):
     """
     Important:
     in order to reduce the number of indexes and since the embedding vector length is fixed, we avoid creating

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

@@ -3,12 +3,21 @@ from pymilvus import FieldSchema, DataType
 import json
 import logging
 from typing import Optional
-
-from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
 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
 
@@ -16,7 +25,7 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
-class MilvusClient:
+class MilvusClient(VectorDBBase):
     def __init__(self):
         self.collection_prefix = "open_webui"
         if MILVUS_TOKEN is None:
@@ -28,7 +37,6 @@ class MilvusClient:
         ids = []
         documents = []
         metadatas = []
-
         for match in result:
             _ids = []
             _documents = []
@@ -37,11 +45,9 @@ class MilvusClient:
                 _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,
@@ -55,13 +61,11 @@ class MilvusClient:
         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
@@ -70,12 +74,10 @@ class MilvusClient:
                 _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,
@@ -108,11 +110,39 @@ class MilvusClient:
         )
 
         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(
@@ -120,6 +150,9 @@ class MilvusClient:
             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.
@@ -140,84 +173,113 @@ class MilvusClient:
     ) -> 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.
@@ -225,10 +287,23 @@ class MilvusClient:
         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=[
@@ -248,10 +323,23 @@ class MilvusClient:
         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=[
@@ -271,30 +359,55 @@ class MilvusClient:
         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

@@ -2,7 +2,12 @@ from opensearchpy import OpenSearch
 from opensearchpy.helpers import bulk
 from typing import Optional
 
-from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
 from open_webui.config import (
     OPENSEARCH_URI,
     OPENSEARCH_SSL,
@@ -12,7 +17,7 @@ from open_webui.config import (
 )
 
 
-class OpenSearchClient:
+class OpenSearchClient(VectorDBBase):
     def __init__(self):
         self.index_prefix = "open_webui"
         self.client = OpenSearch(

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,
@@ -22,8 +26,18 @@ from pgvector.sqlalchemy import Vector
 from sqlalchemy.ext.mutable import MutableDict
 from sqlalchemy.exc import NoSuchTableError
 
-from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
-from open_webui.config import PGVECTOR_DB_URL, PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
+from open_webui.config import (
+    PGVECTOR_DB_URL,
+    PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH,
+    PGVECTOR_PGCRYPTO,
+    PGVECTOR_PGCRYPTO_KEY,
+)
 
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -34,17 +48,30 @@ 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:
+class PgvectorClient(VectorDBBase):
     def __init__(self) -> None:
 
         # if no pgvector uri, use the existing database connection
@@ -65,6 +92,15 @@ class PgvectorClient:
             # 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()
 
@@ -136,51 +172,45 @@ class PgvectorClient:
             # Pad the vector with zeros
             vector += [0.0] * (VECTOR_LENGTH - current_length)
         elif current_length > VECTOR_LENGTH:
-            raise Exception(
-                f"Vector length {current_length} not supported. Max length must be <= {VECTOR_LENGTH}"
-            )
+            # Truncate the vector to VECTOR_LENGTH
+            vector = vector[:VECTOR_LENGTH]
         return vector
 
     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,
@@ -188,11 +218,78 @@ class PgvectorClient:
                         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}")
@@ -226,16 +323,32 @@ class PgvectorClient:
                 .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))
@@ -295,17 +408,43 @@ class PgvectorClient:
         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
@@ -327,20 +466,38 @@ class PgvectorClient:
         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:
@@ -354,17 +511,33 @@ class PgvectorClient:
         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

@@ -0,0 +1,581 @@
+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,
+    SearchResult,
+    GetResult,
+)
+from open_webui.config import (
+    PINECONE_API_KEY,
+    PINECONE_ENVIRONMENT,
+    PINECONE_INDEX_NAME,
+    PINECONE_DIMENSION,
+    PINECONE_METRIC,
+    PINECONE_CLOUD,
+)
+from open_webui.env import SRC_LOG_LEVELS
+
+NO_LIMIT = 10000  # Reasonable limit to avoid overwhelming the system
+BATCH_SIZE = 100  # Recommended batch size for Pinecone operations
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class PineconeClient(VectorDBBase):
+    def __init__(self):
+        self.collection_prefix = "open-webui"
+
+        # Validate required configuration
+        self._validate_config()
+
+        # Store configuration values
+        self.api_key = PINECONE_API_KEY
+        self.environment = PINECONE_ENVIRONMENT
+        self.index_name = PINECONE_INDEX_NAME
+        self.dimension = PINECONE_DIMENSION
+        self.metric = PINECONE_METRIC
+        self.cloud = PINECONE_CLOUD
+
+        # 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()
+
+    def _validate_config(self) -> None:
+        """Validate that all required configuration variables are set."""
+        missing_vars = []
+        if not PINECONE_API_KEY:
+            missing_vars.append("PINECONE_API_KEY")
+        if not PINECONE_ENVIRONMENT:
+            missing_vars.append("PINECONE_ENVIRONMENT")
+        if not PINECONE_INDEX_NAME:
+            missing_vars.append("PINECONE_INDEX_NAME")
+        if not PINECONE_DIMENSION:
+            missing_vars.append("PINECONE_DIMENSION")
+        if not PINECONE_CLOUD:
+            missing_vars.append("PINECONE_CLOUD")
+
+        if missing_vars:
+            raise ValueError(
+                f"Required configuration missing: {', '.join(missing_vars)}"
+            )
+
+    def _initialize_index(self) -> None:
+        """Initialize the Pinecone index."""
+        try:
+            # Check if index exists
+            if self.index_name not in self.client.list_indexes().names():
+                log.info(f"Creating Pinecone index '{self.index_name}'...")
+                self.client.create_index(
+                    name=self.index_name,
+                    dimension=self.dimension,
+                    metric=self.metric,
+                    spec=ServerlessSpec(cloud=self.cloud, region=self.environment),
+                )
+                log.info(f"Successfully created Pinecone index '{self.index_name}'")
+            else:
+                log.info(f"Using existing Pinecone index '{self.index_name}'")
+
+            # Connect to the index
+            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]]:
+        """Convert VectorItem objects to Pinecone point format."""
+        points = []
+        for item in items:
+            # Start with any existing metadata or an empty dict
+            metadata = item.get("metadata", {}).copy() if item.get("metadata") else {}
+
+            # Add text to metadata if available
+            if "text" in item:
+                metadata["text"] = item["text"]
+
+            # Always add collection_name to metadata for filtering
+            metadata["collection_name"] = collection_name_with_prefix
+
+            point = {
+                "id": item["id"],
+                "values": item["vector"],
+                "metadata": metadata,
+            }
+            points.append(point)
+        return points
+
+    def _get_collection_name_with_prefix(self, collection_name: str) -> str:
+        """Get the collection name with prefix."""
+        return f"{self.collection_prefix}_{collection_name}"
+
+    def _normalize_distance(self, score: float) -> float:
+        """Normalize distance score based on the metric used."""
+        if self.metric.lower() == "cosine":
+            # Cosine similarity ranges from -1 to 1, normalize to 0 to 1
+            return (score + 1.0) / 2.0
+        elif self.metric.lower() in ["euclidean", "dotproduct"]:
+            # These are already suitable for ranking (smaller is better for Euclidean)
+            return score
+        else:
+            # For other metrics, use as is
+            return score
+
+    def _result_to_get_result(self, matches: list) -> GetResult:
+        """Convert Pinecone matches to GetResult format."""
+        ids = []
+        documents = []
+        metadatas = []
+
+        for match in matches:
+            metadata = getattr(match, "metadata", {}) or {}
+            ids.append(match.id if hasattr(match, "id") else match["id"])
+            documents.append(metadata.get("text", ""))
+            metadatas.append(metadata)
+
+        return GetResult(
+            **{
+                "ids": [ids],
+                "documents": [documents],
+                "metadatas": [metadatas],
+            }
+        )
+
+    def has_collection(self, collection_name: str) -> bool:
+        """Check if a collection exists by searching for at least one item."""
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+
+        try:
+            # Search for at least 1 item with this collection name in metadata
+            response = self.index.query(
+                vector=[0.0] * self.dimension,  # dummy vector
+                top_k=1,
+                filter={"collection_name": collection_name_with_prefix},
+                include_metadata=False,
+            )
+            matches = getattr(response, "matches", []) or []
+            return len(matches) > 0
+        except Exception as e:
+            log.exception(
+                f"Error checking collection '{collection_name_with_prefix}': {e}"
+            )
+            return False
+
+    def delete_collection(self, collection_name: str) -> None:
+        """Delete a collection by removing all vectors with the collection name in metadata."""
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+        try:
+            self.index.delete(filter={"collection_name": collection_name_with_prefix})
+            log.info(
+                f"Collection '{collection_name_with_prefix}' deleted (all vectors removed)."
+            )
+        except Exception as e:
+            log.warning(
+                f"Failed to delete collection '{collection_name_with_prefix}': {e}"
+            )
+            raise
+
+    def insert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """Insert vectors into a collection."""
+        if not items:
+            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)
+
+        # 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:
+                future.result()
+            except Exception as 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(points)} vectors in parallel batches "
+            f"into '{collection_name_with_prefix}'"
+        )
+
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """Upsert (insert or update) vectors into a collection."""
+        if not items:
+            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)
+
+        # 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 async upserted {len(points)} vectors in batches "
+            f"into '{collection_name_with_prefix}'"
+        )
+
+    def search(
+        self, collection_name: str, vectors: List[List[Union[float, int]]], limit: int
+    ) -> Optional[SearchResult]:
+        """Search for similar vectors in a collection."""
+        if not vectors or not vectors[0]:
+            log.warning("No vectors provided for search")
+            return None
+
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+
+        if limit is None or limit <= 0:
+            limit = NO_LIMIT
+
+        try:
+            # Search using the first vector (assuming this is the intended behavior)
+            query_vector = vectors[0]
+
+            # Perform the search
+            query_response = self.index.query(
+                vector=query_vector,
+                top_k=limit,
+                include_metadata=True,
+                filter={"collection_name": collection_name_with_prefix},
+            )
+
+            matches = getattr(query_response, "matches", []) or []
+            if not matches:
+                # Return empty result if no matches
+                return SearchResult(
+                    ids=[[]],
+                    documents=[[]],
+                    metadatas=[[]],
+                    distances=[[]],
+                )
+
+            # Convert to GetResult format
+            get_result = self._result_to_get_result(matches)
+
+            # Calculate normalized distances based on metric
+            distances = [
+                [
+                    self._normalize_distance(getattr(match, "score", 0.0))
+                    for match in matches
+                ]
+            ]
+
+            return SearchResult(
+                ids=get_result.ids,
+                documents=get_result.documents,
+                metadatas=get_result.metadatas,
+                distances=distances,
+            )
+        except Exception as e:
+            log.error(f"Error searching in '{collection_name_with_prefix}': {e}")
+            return None
+
+    def query(
+        self, collection_name: str, filter: Dict, limit: Optional[int] = None
+    ) -> Optional[GetResult]:
+        """Query vectors by metadata filter."""
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+
+        if limit is None or limit <= 0:
+            limit = NO_LIMIT
+
+        try:
+            # Create a zero vector for the dimension as Pinecone requires a vector
+            zero_vector = [0.0] * self.dimension
+
+            # Combine user filter with collection_name
+            pinecone_filter = {"collection_name": collection_name_with_prefix}
+            if filter:
+                pinecone_filter.update(filter)
+
+            # Perform metadata-only query
+            query_response = self.index.query(
+                vector=zero_vector,
+                filter=pinecone_filter,
+                top_k=limit,
+                include_metadata=True,
+            )
+
+            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}")
+            return None
+
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        """Get all vectors in a collection."""
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+
+        try:
+            # Use a zero vector for fetching all entries
+            zero_vector = [0.0] * self.dimension
+
+            # Add filter to only get vectors for this collection
+            query_response = self.index.query(
+                vector=zero_vector,
+                top_k=NO_LIMIT,
+                include_metadata=True,
+                filter={"collection_name": collection_name_with_prefix},
+            )
+
+            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}")
+            return None
+
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[List[str]] = None,
+        filter: Optional[Dict] = None,
+    ) -> None:
+        """Delete vectors by IDs or filter."""
+        collection_name_with_prefix = self._get_collection_name_with_prefix(
+            collection_name
+        )
+
+        try:
+            if ids:
+                # Delete by IDs (in batches for large deletions)
+                for i in range(0, len(ids), BATCH_SIZE):
+                    batch_ids = ids[i : i + BATCH_SIZE]
+                    # Note: When deleting by ID, we can't filter by collection_name
+                    # 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 "
+                        f"from '{collection_name_with_prefix}'"
+                    )
+                log.info(
+                    f"Successfully deleted {len(ids)} vectors by ID "
+                    f"from '{collection_name_with_prefix}'"
+                )
+
+            elif filter:
+                # Combine user filter with collection_name
+                pinecone_filter = {"collection_name": collection_name_with_prefix}
+                if filter:
+                    pinecone_filter.update(filter)
+                # Delete by metadata filter
+                self.index.delete(filter=pinecone_filter)
+                log.info(
+                    f"Successfully deleted vectors by filter from '{collection_name_with_prefix}'"
+                )
+
+            else:
+                log.warning("No ids or filter provided for delete operation")
+
+        except Exception as e:
+            log.error(f"Error deleting from collection '{collection_name}': {e}")
+            raise
+
+    def reset(self) -> None:
+        """Reset the database by deleting all collections."""
+        try:
+            self.index.delete(delete_all=True)
+            log.info("All vectors successfully deleted from the index.")
+        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

@@ -1,12 +1,24 @@
 from typing import Optional
 import logging
+from urllib.parse import urlparse
 
 from qdrant_client import QdrantClient as Qclient
 from qdrant_client.http.models import PointStruct
 from qdrant_client.models import models
 
-from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
-from open_webui.config import QDRANT_URI, QDRANT_API_KEY
+from open_webui.retrieval.vector.main import (
+    VectorDBBase,
+    VectorItem,
+    SearchResult,
+    GetResult,
+)
+from open_webui.config import (
+    QDRANT_URI,
+    QDRANT_API_KEY,
+    QDRANT_ON_DISK,
+    QDRANT_GRPC_PORT,
+    QDRANT_PREFER_GRPC,
+)
 from open_webui.env import SRC_LOG_LEVELS
 
 NO_LIMIT = 999999999
@@ -15,16 +27,34 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
-class QdrantClient:
+class QdrantClient(VectorDBBase):
     def __init__(self):
         self.collection_prefix = "open-webui"
         self.QDRANT_URI = QDRANT_URI
         self.QDRANT_API_KEY = QDRANT_API_KEY
-        self.client = (
-            Qclient(url=self.QDRANT_URI, api_key=self.QDRANT_API_KEY)
-            if self.QDRANT_URI
-            else None
-        )
+        self.QDRANT_ON_DISK = QDRANT_ON_DISK
+        self.PREFER_GRPC = QDRANT_PREFER_GRPC
+        self.GRPC_PORT = QDRANT_GRPC_PORT
+
+        if not self.QDRANT_URI:
+            self.client = None
+            return
+
+        # Unified handling for either scheme
+        parsed = urlparse(self.QDRANT_URI)
+        host = parsed.hostname or self.QDRANT_URI
+        http_port = parsed.port or 6333  # default REST port
+
+        if self.PREFER_GRPC:
+            self.client = Qclient(
+                host=host,
+                port=http_port,
+                grpc_port=self.GRPC_PORT,
+                prefer_grpc=self.PREFER_GRPC,
+                api_key=self.QDRANT_API_KEY,
+            )
+        else:
+            self.client = Qclient(url=self.QDRANT_URI, api_key=self.QDRANT_API_KEY)
 
     def _result_to_get_result(self, points) -> GetResult:
         ids = []
@@ -50,7 +80,9 @@ class QdrantClient:
         self.client.create_collection(
             collection_name=collection_name_with_prefix,
             vectors_config=models.VectorParams(
-                size=dimension, distance=models.Distance.COSINE
+                size=dimension,
+                distance=models.Distance.COSINE,
+                on_disk=self.QDRANT_ON_DISK,
             ),
         )
 

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

@@ -0,0 +1,712 @@
+import logging
+from typing import Optional, Tuple
+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,
+)
+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
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class QdrantClient(VectorDBBase):
+    def __init__(self):
+        self.collection_prefix = "open-webui"
+        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:
+            self.client = None
+            return
+
+        # Unified handling for either scheme
+        parsed = urlparse(self.QDRANT_URI)
+        host = parsed.hostname or self.QDRANT_URI
+        http_port = parsed.port or 6333  # default REST port
+
+        if self.PREFER_GRPC:
+            self.client = Qclient(
+                host=host,
+                port=http_port,
+                grpc_port=self.GRPC_PORT,
+                prefer_grpc=self.PREFER_GRPC,
+                api_key=self.QDRANT_API_KEY,
+            )
+        else:
+            self.client = 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 _extract_error_message(self, exception):
+        """
+        Extract error message from either HTTP or gRPC exceptions
+
+        Returns:
+            tuple: (status_code, error_message)
+        """
+        # Check if it's an HTTP exception
+        if isinstance(exception, UnexpectedResponse):
+            try:
+                error_data = exception.structured()
+                error_msg = error_data.get("status", {}).get("error", "")
+                return exception.status_code, error_msg
+            except Exception as inner_e:
+                log.error(f"Failed to parse HTTP error: {inner_e}")
+                return exception.status_code, str(exception)
+
+        # Check if it's a gRPC exception
+        elif isinstance(exception, grpc.RpcError):
+            # Extract status code from gRPC error
+            status_code = None
+            if hasattr(exception, "code") and callable(exception.code):
+                status_code = exception.code().value[0]
+
+            # Extract error message
+            error_msg = str(exception)
+            if "details =" in error_msg:
+                # Parse the details line which contains the actual error message
+                try:
+                    details_line = [
+                        line.strip()
+                        for line in error_msg.split("\n")
+                        if "details =" in line
+                    ][0]
+                    error_msg = details_line.split("details =")[1].strip(' "')
+                except (IndexError, AttributeError):
+                    # Fall back to full message if parsing fails
+                    pass
+
+            return status_code, error_msg
+
+        # For any other type of exception
+        return None, str(exception)
+
+    def _is_collection_not_found_error(self, exception):
+        """
+        Check if the exception is due to collection not found, supporting both HTTP and gRPC
+        """
+        status_code, error_msg = self._extract_error_message(exception)
+
+        # HTTP error (404)
+        if (
+            status_code == 404
+            and "Collection" in error_msg
+            and "doesn't exist" in error_msg
+        ):
+            return True
+
+        # gRPC error (NOT_FOUND status)
+        if (
+            isinstance(exception, grpc.RpcError)
+            and exception.code() == grpc.StatusCode.NOT_FOUND
+        ):
+            return True
+
+        return False
+
+    def _is_dimension_mismatch_error(self, exception):
+        """
+        Check if the exception is due to dimension mismatch, supporting both HTTP and gRPC
+        """
+        status_code, error_msg = self._extract_error_message(exception)
+
+        # Common patterns in both HTTP and gRPC
+        return (
+            "Vector dimension error" in error_msg
+            or "dimensions mismatch" in error_msg
+            or "invalid vector size" in error_msg
+        )
+
+    def _create_multi_tenant_collection_if_not_exists(
+        self, mt_collection_name: str, dimension: int = 384
+    ):
+        """
+        Creates a collection with multi-tenancy configuration if it doesn't exist.
+        Default dimension is set to 384 which corresponds to 'sentence-transformers/all-MiniLM-L6-v2'.
+        When creating collections dynamically (insert/upsert), the actual vector dimensions will be used.
+        """
+        try:
+            # Try to create the collection directly - will fail if it already exists
+            self.client.create_collection(
+                collection_name=mt_collection_name,
+                vectors_config=models.VectorParams(
+                    size=dimension,
+                    distance=models.Distance.COSINE,
+                    on_disk=self.QDRANT_ON_DISK,
+                ),
+                hnsw_config=models.HnswConfigDiff(
+                    payload_m=16,  # Enable per-tenant indexing
+                    m=0,
+                    on_disk=self.QDRANT_ON_DISK,
+                ),
+            )
+
+            # Create tenant ID payload index
+            self.client.create_payload_index(
+                collection_name=mt_collection_name,
+                field_name="tenant_id",
+                field_schema=models.KeywordIndexParams(
+                    type=models.KeywordIndexType.KEYWORD,
+                    is_tenant=True,
+                    on_disk=self.QDRANT_ON_DISK,
+                ),
+                wait=True,
+            )
+
+            log.info(
+                f"Multi-tenant collection {mt_collection_name} created with dimension {dimension}!"
+            )
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            # Check for the specific error indicating collection already exists
+            status_code, error_msg = self._extract_error_message(e)
+
+            # HTTP status code 409 or gRPC ALREADY_EXISTS
+            if (isinstance(e, UnexpectedResponse) and status_code == 409) or (
+                isinstance(e, grpc.RpcError)
+                and e.code() == grpc.StatusCode.ALREADY_EXISTS
+            ):
+                if "already exists" in error_msg:
+                    log.debug(f"Collection {mt_collection_name} already exists")
+                    return
+            # If it's not an already exists error, re-raise
+            raise e
+        except Exception as e:
+            raise e
+
+    def _create_points(self, items: list[VectorItem], tenant_id: str):
+        """
+        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": tenant_id,
+                },
+            )
+            for item in items
+        ]
+
+    def has_collection(self, collection_name: str) -> bool:
+        """
+        Check if a logical collection exists by checking for any points with the tenant ID.
+        """
+        if not self.client:
+            return False
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Create tenant filter
+        tenant_filter = models.FieldCondition(
+            key="tenant_id", match=models.MatchValue(value=tenant_id)
+        )
+
+        try:
+            # Try directly querying - most of the time collection should exist
+            response = self.client.query_points(
+                collection_name=mt_collection,
+                query_filter=models.Filter(must=[tenant_filter]),
+                limit=1,
+            )
+
+            # Collection exists with this tenant ID if there are points
+            return len(response.points) > 0
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            if self._is_collection_not_found_error(e):
+                log.debug(f"Collection {mt_collection} doesn't exist")
+                return False
+            else:
+                # For other API errors, log and return False
+                _, error_msg = self._extract_error_message(e)
+                log.warning(f"Unexpected Qdrant error: {error_msg}")
+                return False
+        except Exception as e:
+            # For any other errors, log and return False
+            log.debug(f"Error checking collection {mt_collection}: {e}")
+            return False
+
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[list[str]] = None,
+        filter: Optional[dict] = None,
+    ):
+        """
+        Delete vectors by ID or filter from a collection with tenant isolation.
+        """
+        if not self.client:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Create tenant filter
+        tenant_filter = models.FieldCondition(
+            key="tenant_id", match=models.MatchValue(value=tenant_id)
+        )
+
+        must_conditions = [tenant_filter]
+        should_conditions = []
+
+        if ids:
+            for id_value in ids:
+                should_conditions.append(
+                    models.FieldCondition(
+                        key="metadata.id",
+                        match=models.MatchValue(value=id_value),
+                    ),
+                )
+        elif filter:
+            for key, value in filter.items():
+                must_conditions.append(
+                    models.FieldCondition(
+                        key=f"metadata.{key}",
+                        match=models.MatchValue(value=value),
+                    ),
+                )
+
+        try:
+            # Try to delete directly - most of the time collection should exist
+            update_result = self.client.delete(
+                collection_name=mt_collection,
+                points_selector=models.FilterSelector(
+                    filter=models.Filter(must=must_conditions, should=should_conditions)
+                ),
+            )
+
+            return update_result
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            if self._is_collection_not_found_error(e):
+                log.debug(
+                    f"Collection {mt_collection} doesn't exist, nothing to delete"
+                )
+                return None
+            else:
+                # For other API errors, log and re-raise
+                _, error_msg = self._extract_error_message(e)
+                log.warning(f"Unexpected Qdrant error: {error_msg}")
+                raise
+        except Exception as e:
+            # For non-Qdrant exceptions, re-raise
+            raise
+
+    def search(
+        self, collection_name: str, vectors: list[list[float | int]], limit: int
+    ) -> Optional[SearchResult]:
+        """
+        Search for the nearest neighbor items based on the vectors with tenant isolation.
+        """
+        if not self.client:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Get the vector dimension from the query vector
+        dimension = len(vectors[0]) if vectors and len(vectors) > 0 else None
+
+        try:
+            # Try the search operation directly - most of the time collection should exist
+
+            # Create tenant filter
+            tenant_filter = models.FieldCondition(
+                key="tenant_id", match=models.MatchValue(value=tenant_id)
+            )
+
+            # Ensure vector dimensions match the collection
+            collection_dim = self.client.get_collection(
+                mt_collection
+            ).config.params.vectors.size
+
+            if collection_dim != dimension:
+                if collection_dim < dimension:
+                    vectors = [vector[:collection_dim] for vector in vectors]
+                else:
+                    vectors = [
+                        vector + [0] * (collection_dim - dimension)
+                        for vector in vectors
+                    ]
+
+            # Search with tenant filter
+            prefetch_query = models.Prefetch(
+                filter=models.Filter(must=[tenant_filter]),
+                limit=NO_LIMIT,
+            )
+            query_response = self.client.query_points(
+                collection_name=mt_collection,
+                query=vectors[0],
+                prefetch=prefetch_query,
+                limit=limit,
+            )
+
+            get_result = self._result_to_get_result(query_response.points)
+            return SearchResult(
+                ids=get_result.ids,
+                documents=get_result.documents,
+                metadatas=get_result.metadatas,
+                # qdrant distance is [-1, 1], normalize to [0, 1]
+                distances=[
+                    [(point.score + 1.0) / 2.0 for point in query_response.points]
+                ],
+            )
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            if self._is_collection_not_found_error(e):
+                log.debug(
+                    f"Collection {mt_collection} doesn't exist, search returns None"
+                )
+                return None
+            else:
+                # For other API errors, log and re-raise
+                _, error_msg = self._extract_error_message(e)
+                log.warning(f"Unexpected Qdrant error during search: {error_msg}")
+                raise
+        except Exception as e:
+            # For non-Qdrant exceptions, log and return None
+            log.exception(f"Error searching collection '{collection_name}': {e}")
+            return None
+
+    def query(self, collection_name: str, filter: dict, limit: Optional[int] = None):
+        """
+        Query points with filters and tenant isolation.
+        """
+        if not self.client:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Set default limit if not provided
+        if limit is None:
+            limit = NO_LIMIT
+
+        # Create tenant filter
+        tenant_filter = models.FieldCondition(
+            key="tenant_id", match=models.MatchValue(value=tenant_id)
+        )
+
+        # Create metadata filters
+        field_conditions = []
+        for key, value in filter.items():
+            field_conditions.append(
+                models.FieldCondition(
+                    key=f"metadata.{key}", match=models.MatchValue(value=value)
+                )
+            )
+
+        # Combine tenant filter with metadata filters
+        combined_filter = models.Filter(must=[tenant_filter, *field_conditions])
+
+        try:
+            # Try the query directly - most of the time collection should exist
+            points = self.client.query_points(
+                collection_name=mt_collection,
+                query_filter=combined_filter,
+                limit=limit,
+            )
+
+            return self._result_to_get_result(points.points)
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            if self._is_collection_not_found_error(e):
+                log.debug(
+                    f"Collection {mt_collection} doesn't exist, query returns None"
+                )
+                return None
+            else:
+                # For other API errors, log and re-raise
+                _, error_msg = self._extract_error_message(e)
+                log.warning(f"Unexpected Qdrant error during query: {error_msg}")
+                raise
+        except Exception as e:
+            # For non-Qdrant exceptions, log and re-raise
+            log.exception(f"Error querying collection '{collection_name}': {e}")
+            return None
+
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        """
+        Get all items in a collection with tenant isolation.
+        """
+        if not self.client:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Create tenant filter
+        tenant_filter = models.FieldCondition(
+            key="tenant_id", match=models.MatchValue(value=tenant_id)
+        )
+
+        try:
+            # Try to get points directly - most of the time collection should exist
+            points = self.client.query_points(
+                collection_name=mt_collection,
+                query_filter=models.Filter(must=[tenant_filter]),
+                limit=NO_LIMIT,
+            )
+
+            return self._result_to_get_result(points.points)
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            if self._is_collection_not_found_error(e):
+                log.debug(f"Collection {mt_collection} doesn't exist, get returns None")
+                return None
+            else:
+                # For other API errors, log and re-raise
+                _, error_msg = self._extract_error_message(e)
+                log.warning(f"Unexpected Qdrant error during get: {error_msg}")
+                raise
+        except Exception as e:
+            # For non-Qdrant exceptions, log and return None
+            log.exception(f"Error getting collection '{collection_name}': {e}")
+            return None
+
+    def _handle_operation_with_error_retry(
+        self, operation_name, mt_collection, points, dimension
+    ):
+        """
+        Private helper to handle common error cases for insert and upsert operations.
+
+        Args:
+            operation_name: 'insert' or 'upsert'
+            mt_collection: The multi-tenant collection name
+            points: The vector points to insert/upsert
+            dimension: The dimension of the vectors
+
+        Returns:
+            The operation result (for upsert) or None (for insert)
+        """
+        try:
+            if operation_name == "insert":
+                self.client.upload_points(mt_collection, points)
+                return None
+            else:  # upsert
+                return self.client.upsert(mt_collection, points)
+        except (UnexpectedResponse, grpc.RpcError) as e:
+            # Handle collection not found
+            if self._is_collection_not_found_error(e):
+                log.info(
+                    f"Collection {mt_collection} doesn't exist. Creating it with dimension {dimension}."
+                )
+                # Create collection with correct dimensions from our vectors
+                self._create_multi_tenant_collection_if_not_exists(
+                    mt_collection_name=mt_collection, dimension=dimension
+                )
+                # Try operation again - no need for dimension adjustment since we just created with correct dimensions
+                if operation_name == "insert":
+                    self.client.upload_points(mt_collection, points)
+                    return None
+                else:  # upsert
+                    return self.client.upsert(mt_collection, points)
+
+            # Handle dimension mismatch
+            elif self._is_dimension_mismatch_error(e):
+                # For dimension errors, the collection must exist, so get its configuration
+                mt_collection_info = self.client.get_collection(mt_collection)
+                existing_size = mt_collection_info.config.params.vectors.size
+
+                log.info(
+                    f"Dimension mismatch: Collection {mt_collection} expects {existing_size}, got {dimension}"
+                )
+
+                if existing_size < dimension:
+                    # Truncate vectors to fit
+                    log.info(
+                        f"Truncating vectors from {dimension} to {existing_size} dimensions"
+                    )
+                    points = [
+                        PointStruct(
+                            id=point.id,
+                            vector=point.vector[:existing_size],
+                            payload=point.payload,
+                        )
+                        for point in points
+                    ]
+                elif existing_size > dimension:
+                    # Pad vectors with zeros
+                    log.info(
+                        f"Padding vectors from {dimension} to {existing_size} dimensions with zeros"
+                    )
+                    points = [
+                        PointStruct(
+                            id=point.id,
+                            vector=point.vector
+                            + [0] * (existing_size - len(point.vector)),
+                            payload=point.payload,
+                        )
+                        for point in points
+                    ]
+                # Try operation again with adjusted dimensions
+                if operation_name == "insert":
+                    self.client.upload_points(mt_collection, points)
+                    return None
+                else:  # upsert
+                    return self.client.upsert(mt_collection, points)
+            else:
+                # Not a known error we can handle, log and re-raise
+                _, error_msg = self._extract_error_message(e)
+                log.warning(f"Unhandled Qdrant error: {error_msg}")
+                raise
+        except Exception as e:
+            # For non-Qdrant exceptions, re-raise
+            raise
+
+    def insert(self, collection_name: str, items: list[VectorItem]):
+        """
+        Insert items with tenant ID.
+        """
+        if not self.client or not items:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Get dimensions from the actual vectors
+        dimension = len(items[0]["vector"]) if items else None
+
+        # Create points with tenant ID
+        points = self._create_points(items, tenant_id)
+
+        # Handle the operation with error retry
+        return self._handle_operation_with_error_retry(
+            "insert", mt_collection, points, dimension
+        )
+
+    def upsert(self, collection_name: str, items: list[VectorItem]):
+        """
+        Upsert items with tenant ID.
+        """
+        if not self.client or not items:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        # Get dimensions from the actual vectors
+        dimension = len(items[0]["vector"]) if items else None
+
+        # Create points with tenant ID
+        points = self._create_points(items, tenant_id)
+
+        # Handle the operation with error retry
+        return self._handle_operation_with_error_retry(
+            "upsert", mt_collection, points, dimension
+        )
+
+    def reset(self):
+        """
+        Reset the database by deleting all collections.
+        """
+        if not self.client:
+            return None
+
+        collection_names = self.client.get_collections().collections
+        for collection_name in collection_names:
+            if collection_name.name.startswith(self.collection_prefix):
+                self.client.delete_collection(collection_name=collection_name.name)
+
+    def delete_collection(self, collection_name: str):
+        """
+        Delete a collection.
+        """
+        if not self.client:
+            return None
+
+        # Map to multi-tenant collection and tenant ID
+        mt_collection, tenant_id = self._get_collection_and_tenant_id(collection_name)
+
+        tenant_filter = models.FieldCondition(
+            key="tenant_id", match=models.MatchValue(value=tenant_id)
+        )
+
+        field_conditions = [tenant_filter]
+
+        update_result = self.client.delete(
+            collection_name=mt_collection,
+            points_selector=models.FilterSelector(
+                filter=models.Filter(must=field_conditions)
+            ),
+        )
+
+        if self.client.get_collection(mt_collection).points_count == 0:
+            self.client.delete_collection(mt_collection)
+
+        return update_result

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/main.py

@@ -1,5 +1,6 @@
 from pydantic import BaseModel
-from typing import Optional, List, Any
+from abc import ABC, abstractmethod
+from typing import Any, Dict, List, Optional, Union
 
 
 class VectorItem(BaseModel):
@@ -17,3 +18,69 @@ class GetResult(BaseModel):
 
 class SearchResult(GetResult):
     distances: Optional[List[List[float | int]]]
+
+
+class VectorDBBase(ABC):
+    """
+    Abstract base class for all vector database backends.
+
+    Implementations of this class provide methods for collection management,
+    vector insertion, deletion, similarity search, and metadata filtering.
+
+    Any custom vector database integration must inherit from this class and
+    implement all abstract methods.
+    """
+
+    @abstractmethod
+    def has_collection(self, collection_name: str) -> bool:
+        """Check if the collection exists in the vector DB."""
+        pass
+
+    @abstractmethod
+    def delete_collection(self, collection_name: str) -> None:
+        """Delete a collection from the vector DB."""
+        pass
+
+    @abstractmethod
+    def insert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """Insert a list of vector items into a collection."""
+        pass
+
+    @abstractmethod
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        """Insert or update vector items in a collection."""
+        pass
+
+    @abstractmethod
+    def search(
+        self, collection_name: str, vectors: List[List[Union[float, int]]], limit: int
+    ) -> Optional[SearchResult]:
+        """Search for similar vectors in a collection."""
+        pass
+
+    @abstractmethod
+    def query(
+        self, collection_name: str, filter: Dict, limit: Optional[int] = None
+    ) -> Optional[GetResult]:
+        """Query vectors from a collection using metadata filter."""
+        pass
+
+    @abstractmethod
+    def get(self, collection_name: str) -> Optional[GetResult]:
+        """Retrieve all vectors from a collection."""
+        pass
+
+    @abstractmethod
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[List[str]] = None,
+        filter: Optional[Dict] = None,
+    ) -> None:
+        """Delete vectors by ID or filter from a collection."""
+        pass
+
+    @abstractmethod
+    def reset(self) -> None:
+        """Reset the vector database by removing all collections or those matching a condition."""
+        pass

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/duckduckgo.py

@@ -3,6 +3,7 @@ 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 open_webui.env import SRC_LOG_LEVELS
 
 log = logging.getLogger(__name__)
@@ -22,16 +23,15 @@ def search_duckduckgo(
         list[SearchResult]: A list of search results
     """
     # Use the DDGS context manager to create a DDGS object
+    search_results = []
     with DDGS() as ddgs:
         # Use the ddgs.text() method to perform the search
-        ddgs_gen = ddgs.text(
-            query, safesearch="moderate", max_results=count, backend="api"
-        )
-        # Check if there are search results
-        if ddgs_gen:
-            # Convert the search results into a list
-            search_results = [r for r in ddgs_gen]
-
+        try:
+            search_results = ddgs.text(
+                query, safesearch="moderate", max_results=count, backend="lite"
+            )
+        except RatelimitException as e:
+            log.error(f"RatelimitException: {e}")
     if filter_list:
         search_results = get_filtered_results(search_results, filter_list)
 

backend/open_webui/retrieval/web/external.py

@@ -0,0 +1,47 @@
+import logging
+from typing import Optional, List
+
+import requests
+from open_webui.retrieval.web.main import SearchResult, get_filtered_results
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+def search_external(
+    external_url: str,
+    external_api_key: str,
+    query: str,
+    count: int,
+    filter_list: Optional[List[str]] = None,
+) -> List[SearchResult]:
+    try:
+        response = requests.post(
+            external_url,
+            headers={
+                "User-Agent": "Open WebUI (https://github.com/open-webui/open-webui) RAG Bot",
+                "Authorization": f"Bearer {external_api_key}",
+            },
+            json={
+                "query": query,
+                "count": count,
+            },
+        )
+        response.raise_for_status()
+        results = response.json()
+        if filter_list:
+            results = get_filtered_results(results, filter_list)
+        results = [
+            SearchResult(
+                link=result.get("link"),
+                title=result.get("title"),
+                snippet=result.get("snippet"),
+            )
+            for result in results[:count]
+        ]
+        log.info(f"External search results: {results}")
+        return results
+    except Exception as e:
+        log.error(f"Error in External search: {e}")
+        return []

backend/open_webui/retrieval/web/firecrawl.py

@@ -0,0 +1,49 @@
+import logging
+from typing import Optional, List
+from urllib.parse import urljoin
+
+import requests
+from open_webui.retrieval.web.main import SearchResult, get_filtered_results
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+def search_firecrawl(
+    firecrawl_url: str,
+    firecrawl_api_key: str,
+    query: str,
+    count: int,
+    filter_list: Optional[List[str]] = None,
+) -> List[SearchResult]:
+    try:
+        firecrawl_search_url = urljoin(firecrawl_url, "/v1/search")
+        response = requests.post(
+            firecrawl_search_url,
+            headers={
+                "User-Agent": "Open WebUI (https://github.com/open-webui/open-webui) RAG Bot",
+                "Authorization": f"Bearer {firecrawl_api_key}",
+            },
+            json={
+                "query": query,
+                "limit": count,
+            },
+        )
+        response.raise_for_status()
+        results = response.json().get("data", [])
+        if filter_list:
+            results = get_filtered_results(results, filter_list)
+        results = [
+            SearchResult(
+                link=result.get("url"),
+                title=result.get("title"),
+                snippet=result.get("description"),
+            )
+            for result in results[:count]
+        ]
+        log.info(f"External search results: {results}")
+        return results
+    except Exception as e:
+        log.error(f"Error in External search: {e}")
+        return []

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/sougou.py

@@ -0,0 +1,60 @@
+import logging
+import json
+from typing import Optional, List
+
+
+from open_webui.retrieval.web.main import SearchResult, get_filtered_results
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+def search_sougou(
+    sougou_api_sid: str,
+    sougou_api_sk: str,
+    query: str,
+    count: int,
+    filter_list: Optional[List[str]] = None,
+) -> List[SearchResult]:
+    from tencentcloud.common.common_client import CommonClient
+    from tencentcloud.common import credential
+    from tencentcloud.common.exception.tencent_cloud_sdk_exception import (
+        TencentCloudSDKException,
+    )
+    from tencentcloud.common.profile.client_profile import ClientProfile
+    from tencentcloud.common.profile.http_profile import HttpProfile
+
+    try:
+        cred = credential.Credential(sougou_api_sid, sougou_api_sk)
+        http_profile = HttpProfile()
+        http_profile.endpoint = "tms.tencentcloudapi.com"
+        client_profile = ClientProfile()
+        client_profile.http_profile = http_profile
+        params = json.dumps({"Query": query, "Cnt": 20})
+        common_client = CommonClient(
+            "tms", "2020-12-29", cred, "", profile=client_profile
+        )
+        results = [
+            json.loads(page)
+            for page in common_client.call_json("SearchPro", json.loads(params))[
+                "Response"
+            ]["Pages"]
+        ]
+        sorted_results = sorted(
+            results, key=lambda x: x.get("scour", 0.0), reverse=True
+        )
+        if filter_list:
+            sorted_results = get_filtered_results(sorted_results, filter_list)
+
+        return [
+            SearchResult(
+                link=result.get("url"),
+                title=result.get("title"),
+                snippet=result.get("passage"),
+            )
+            for result in sorted_results[:count]
+        ]
+    except TencentCloudSDKException as err:
+        log.error(f"Error in Sougou search: {err}")
+        return []

backend/open_webui/retrieval/web/tavily.py

@@ -2,7 +2,7 @@ import logging
 from typing import Optional
 
 import requests
-from open_webui.retrieval.web.main import SearchResult
+from open_webui.retrieval.web.main import SearchResult, get_filtered_results
 from open_webui.env import SRC_LOG_LEVELS
 
 log = logging.getLogger(__name__)
@@ -21,18 +21,25 @@ def search_tavily(
     Args:
         api_key (str): A Tavily Search API key
         query (str): The query to search for
+        count (int): The maximum number of results to return
 
     Returns:
         list[SearchResult]: A list of search results
     """
     url = "https://api.tavily.com/search"
-    data = {"query": query, "api_key": api_key}
-    response = requests.post(url, json=data)
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": f"Bearer {api_key}",
+    }
+    data = {"query": query, "max_results": count}
+    response = requests.post(url, headers=headers, json=data)
     response.raise_for_status()
 
     json_response = response.json()
 
-    raw_search_results = json_response.get("results", [])
+    results = json_response.get("results", [])
+    if filter_list:
+        results = get_filtered_results(results, filter_list)
 
     return [
         SearchResult(
@@ -40,5 +47,5 @@ def search_tavily(
             title=result.get("title", ""),
             snippet=result.get("content"),
         )
-        for result in raw_search_results[:count]
+        for result in results
     ]

backend/open_webui/retrieval/web/utils.py

@@ -25,18 +25,21 @@ 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_web import ExternalWebLoader
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.config import (
     ENABLE_RAG_LOCAL_WEB_FETCH,
-    PLAYWRIGHT_WS_URI,
+    PLAYWRIGHT_WS_URL,
     PLAYWRIGHT_TIMEOUT,
-    RAG_WEB_LOADER_ENGINE,
+    WEB_LOADER_ENGINE,
     FIRECRAWL_API_BASE_URL,
     FIRECRAWL_API_KEY,
     TAVILY_API_KEY,
     TAVILY_EXTRACT_DEPTH,
+    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"])
@@ -167,7 +170,7 @@ class SafeFireCrawlLoader(BaseLoader, RateLimitMixin, URLProcessingMixin):
         continue_on_failure: bool = True,
         api_key: Optional[str] = None,
         api_url: Optional[str] = None,
-        mode: Literal["crawl", "scrape", "map"] = "crawl",
+        mode: Literal["crawl", "scrape", "map"] = "scrape",
         proxy: Optional[Dict[str, str]] = None,
         params: Optional[Dict] = None,
     ):
@@ -225,7 +228,10 @@ class SafeFireCrawlLoader(BaseLoader, RateLimitMixin, URLProcessingMixin):
                     mode=self.mode,
                     params=self.params,
                 )
-                yield from loader.lazy_load()
+                for document in loader.lazy_load():
+                    if not document.metadata.get("source"):
+                        document.metadata["source"] = document.metadata.get("sourceURL")
+                    yield document
             except Exception as e:
                 if self.continue_on_failure:
                     log.exception(f"Error loading {url}: {e}")
@@ -245,6 +251,8 @@ class SafeFireCrawlLoader(BaseLoader, RateLimitMixin, URLProcessingMixin):
                     params=self.params,
                 )
                 async for document in loader.alazy_load():
+                    if not document.metadata.get("source"):
+                        document.metadata["source"] = document.metadata.get("sourceURL")
                     yield document
             except Exception as e:
                 if self.continue_on_failure:
@@ -507,7 +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()
@@ -584,13 +593,6 @@ class SafeWebBaseLoader(WebBaseLoader):
         return [document async for document in self.alazy_load()]
 
 
-RAG_WEB_LOADER_ENGINES = defaultdict(lambda: SafeWebBaseLoader)
-RAG_WEB_LOADER_ENGINES["playwright"] = SafePlaywrightURLLoader
-RAG_WEB_LOADER_ENGINES["safe_web"] = SafeWebBaseLoader
-RAG_WEB_LOADER_ENGINES["firecrawl"] = SafeFireCrawlLoader
-RAG_WEB_LOADER_ENGINES["tavily"] = SafeTavilyLoader
-
-
 def get_web_loader(
     urls: Union[str, Sequence[str]],
     verify_ssl: bool = True,
@@ -608,27 +610,41 @@ def get_web_loader(
         "trust_env": trust_env,
     }
 
-    if RAG_WEB_LOADER_ENGINE.value == "playwright":
+    if WEB_LOADER_ENGINE.value == "" or WEB_LOADER_ENGINE.value == "safe_web":
+        WebLoaderClass = SafeWebBaseLoader
+    if WEB_LOADER_ENGINE.value == "playwright":
+        WebLoaderClass = SafePlaywrightURLLoader
         web_loader_args["playwright_timeout"] = PLAYWRIGHT_TIMEOUT.value * 1000
-        if PLAYWRIGHT_WS_URI.value:
-            web_loader_args["playwright_ws_url"] = PLAYWRIGHT_WS_URI.value
+        if PLAYWRIGHT_WS_URL.value:
+            web_loader_args["playwright_ws_url"] = PLAYWRIGHT_WS_URL.value
 
-    if RAG_WEB_LOADER_ENGINE.value == "firecrawl":
+    if WEB_LOADER_ENGINE.value == "firecrawl":
+        WebLoaderClass = SafeFireCrawlLoader
         web_loader_args["api_key"] = FIRECRAWL_API_KEY.value
         web_loader_args["api_url"] = FIRECRAWL_API_BASE_URL.value
 
-    if RAG_WEB_LOADER_ENGINE.value == "tavily":
+    if WEB_LOADER_ENGINE.value == "tavily":
+        WebLoaderClass = SafeTavilyLoader
         web_loader_args["api_key"] = TAVILY_API_KEY.value
         web_loader_args["extract_depth"] = TAVILY_EXTRACT_DEPTH.value
 
-    # Create the appropriate WebLoader based on the configuration
-    WebLoaderClass = RAG_WEB_LOADER_ENGINES[RAG_WEB_LOADER_ENGINE.value]
-    web_loader = WebLoaderClass(**web_loader_args)
+    if WEB_LOADER_ENGINE.value == "external":
+        WebLoaderClass = ExternalWebLoader
+        web_loader_args["external_url"] = EXTERNAL_WEB_LOADER_URL.value
+        web_loader_args["external_api_key"] = EXTERNAL_WEB_LOADER_API_KEY.value
 
-    log.debug(
-        "Using RAG_WEB_LOADER_ENGINE %s for %s URLs",
-        web_loader.__class__.__name__,
-        len(safe_urls),
-    )
+    if WebLoaderClass:
+        web_loader = WebLoaderClass(**web_loader_args)
 
-    return web_loader
+        log.debug(
+            "Using WEB_LOADER_ENGINE %s for %s URLs",
+            web_loader.__class__.__name__,
+            len(safe_urls),
+        )
+
+        return web_loader
+    else:
+        raise ValueError(
+            f"Invalid WEB_LOADER_ENGINE: {WEB_LOADER_ENGINE.value}. "
+            "Please set it to 'safe_web', 'playwright', 'firecrawl', or 'tavily'."
+        )

backend/open_webui/retrieval/web/yacy.py

@@ -0,0 +1,87 @@
+import logging
+from typing import Optional
+
+import requests
+from requests.auth import HTTPDigestAuth
+from open_webui.retrieval.web.main import SearchResult, get_filtered_results
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+def search_yacy(
+    query_url: str,
+    username: Optional[str],
+    password: Optional[str],
+    query: str,
+    count: int,
+    filter_list: Optional[list[str]] = None,
+) -> list[SearchResult]:
+    """
+    Search a Yacy instance for a given query and return the results as a list of SearchResult objects.
+
+    The function accepts username and password for authenticating to Yacy.
+
+    Args:
+        query_url (str): The base URL of the Yacy server.
+        username (str): Optional YaCy username.
+        password (str): Optional YaCy password.
+        query (str): The search term or question to find in the Yacy database.
+        count (int): The maximum number of results to retrieve from the search.
+
+    Returns:
+        list[SearchResult]: A list of SearchResults sorted by relevance score in descending order.
+
+    Raise:
+        requests.exceptions.RequestException: If a request error occurs during the search process.
+    """
+
+    # Use authentication if either username or password is set
+    yacy_auth = None
+    if username or password:
+        yacy_auth = HTTPDigestAuth(username, password)
+
+    params = {
+        "query": query,
+        "contentdom": "text",
+        "resource": "global",
+        "maximumRecords": count,
+        "nav": "none",
+    }
+
+    # Check if provided a json API URL
+    if not query_url.endswith("yacysearch.json"):
+        # Strip all query parameters from the URL
+        query_url = query_url.rstrip("/") + "/yacysearch.json"
+
+    log.debug(f"searching {query_url}")
+
+    response = requests.get(
+        query_url,
+        auth=yacy_auth,
+        headers={
+            "User-Agent": "Open WebUI (https://github.com/open-webui/open-webui) RAG Bot",
+            "Accept": "text/html",
+            "Accept-Encoding": "gzip, deflate",
+            "Accept-Language": "en-US,en;q=0.5",
+            "Connection": "keep-alive",
+        },
+        params=params,
+    )
+
+    response.raise_for_status()  # Raise an exception for HTTP errors.
+
+    json_response = response.json()
+    results = json_response.get("channels", [{}])[0].get("items", [])
+    sorted_results = sorted(results, key=lambda x: x.get("ranking", 0), reverse=True)
+    if filter_list:
+        sorted_results = get_filtered_results(sorted_results, filter_list)
+    return [
+        SearchResult(
+            link=result["link"],
+            title=result.get("title"),
+            snippet=result.get("description"),
+        )
+        for result in sorted_results[:count]
+    ]

backend/open_webui/routers/audio.py

@@ -7,6 +7,9 @@ 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
+
 
 import aiohttp
 import aiofiles
@@ -17,6 +20,7 @@ from fastapi import (
     Depends,
     FastAPI,
     File,
+    Form,
     HTTPException,
     Request,
     UploadFile,
@@ -33,10 +37,12 @@ from open_webui.config import (
     WHISPER_MODEL_AUTO_UPDATE,
     WHISPER_MODEL_DIR,
     CACHE_DIR,
+    WHISPER_LANGUAGE,
 )
 
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import (
+    AIOHTTP_CLIENT_SESSION_SSL,
     AIOHTTP_CLIENT_TIMEOUT,
     ENV,
     SRC_LOG_LEVELS,
@@ -48,8 +54,10 @@ 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
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["AUDIO"])
@@ -68,27 +76,47 @@ from pydub import AudioSegment
 from pydub.utils import mediainfo
 
 
-def is_mp4_audio(file_path):
-    """Check if the given file is an MP4 audio file."""
+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"
-    ):
+    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
-    return False
+    except Exception as e:
+        log.error(f"Error getting audio format: {e}")
+        return False
 
 
-def convert_mp4_to_wav(file_path, output_path):
-    """Convert MP4 audio file to WAV format."""
-    audio = AudioSegment.from_file(file_path, format="mp4")
-    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):
@@ -131,6 +159,7 @@ class TTSConfigForm(BaseModel):
     VOICE: str
     SPLIT_ON: str
     AZURE_SPEECH_REGION: str
+    AZURE_SPEECH_BASE_URL: str
     AZURE_SPEECH_OUTPUT_FORMAT: str
 
 
@@ -141,6 +170,11 @@ class STTConfigForm(BaseModel):
     MODEL: str
     WHISPER_MODEL: str
     DEEPGRAM_API_KEY: str
+    AZURE_API_KEY: str
+    AZURE_REGION: str
+    AZURE_LOCALES: str
+    AZURE_BASE_URL: str
+    AZURE_MAX_SPEAKERS: str
 
 
 class AudioConfigUpdateForm(BaseModel):
@@ -160,6 +194,7 @@ async def get_audio_config(request: Request, user=Depends(get_admin_user)):
             "VOICE": request.app.state.config.TTS_VOICE,
             "SPLIT_ON": request.app.state.config.TTS_SPLIT_ON,
             "AZURE_SPEECH_REGION": request.app.state.config.TTS_AZURE_SPEECH_REGION,
+            "AZURE_SPEECH_BASE_URL": request.app.state.config.TTS_AZURE_SPEECH_BASE_URL,
             "AZURE_SPEECH_OUTPUT_FORMAT": request.app.state.config.TTS_AZURE_SPEECH_OUTPUT_FORMAT,
         },
         "stt": {
@@ -169,6 +204,11 @@ async def get_audio_config(request: Request, user=Depends(get_admin_user)):
             "MODEL": request.app.state.config.STT_MODEL,
             "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,
+            "AZURE_REGION": request.app.state.config.AUDIO_STT_AZURE_REGION,
+            "AZURE_LOCALES": request.app.state.config.AUDIO_STT_AZURE_LOCALES,
+            "AZURE_BASE_URL": request.app.state.config.AUDIO_STT_AZURE_BASE_URL,
+            "AZURE_MAX_SPEAKERS": request.app.state.config.AUDIO_STT_AZURE_MAX_SPEAKERS,
         },
     }
 
@@ -185,6 +225,9 @@ async def update_audio_config(
     request.app.state.config.TTS_VOICE = form_data.tts.VOICE
     request.app.state.config.TTS_SPLIT_ON = form_data.tts.SPLIT_ON
     request.app.state.config.TTS_AZURE_SPEECH_REGION = form_data.tts.AZURE_SPEECH_REGION
+    request.app.state.config.TTS_AZURE_SPEECH_BASE_URL = (
+        form_data.tts.AZURE_SPEECH_BASE_URL
+    )
     request.app.state.config.TTS_AZURE_SPEECH_OUTPUT_FORMAT = (
         form_data.tts.AZURE_SPEECH_OUTPUT_FORMAT
     )
@@ -195,6 +238,13 @@ async def update_audio_config(
     request.app.state.config.STT_MODEL = form_data.stt.MODEL
     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
+    request.app.state.config.AUDIO_STT_AZURE_REGION = form_data.stt.AZURE_REGION
+    request.app.state.config.AUDIO_STT_AZURE_LOCALES = form_data.stt.AZURE_LOCALES
+    request.app.state.config.AUDIO_STT_AZURE_BASE_URL = form_data.stt.AZURE_BASE_URL
+    request.app.state.config.AUDIO_STT_AZURE_MAX_SPEAKERS = (
+        form_data.stt.AZURE_MAX_SPEAKERS
+    )
 
     if request.app.state.config.STT_ENGINE == "":
         request.app.state.faster_whisper_model = set_faster_whisper_model(
@@ -211,6 +261,7 @@ async def update_audio_config(
             "VOICE": request.app.state.config.TTS_VOICE,
             "SPLIT_ON": request.app.state.config.TTS_SPLIT_ON,
             "AZURE_SPEECH_REGION": request.app.state.config.TTS_AZURE_SPEECH_REGION,
+            "AZURE_SPEECH_BASE_URL": request.app.state.config.TTS_AZURE_SPEECH_BASE_URL,
             "AZURE_SPEECH_OUTPUT_FORMAT": request.app.state.config.TTS_AZURE_SPEECH_OUTPUT_FORMAT,
         },
         "stt": {
@@ -220,6 +271,11 @@ async def update_audio_config(
             "MODEL": request.app.state.config.STT_MODEL,
             "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,
+            "AZURE_REGION": request.app.state.config.AUDIO_STT_AZURE_REGION,
+            "AZURE_LOCALES": request.app.state.config.AUDIO_STT_AZURE_LOCALES,
+            "AZURE_BASE_URL": request.app.state.config.AUDIO_STT_AZURE_BASE_URL,
+            "AZURE_MAX_SPEAKERS": request.app.state.config.AUDIO_STT_AZURE_MAX_SPEAKERS,
         },
     }
 
@@ -287,6 +343,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                             else {}
                         ),
                     },
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as r:
                     r.raise_for_status()
 
@@ -312,7 +369,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500),
+                status_code=getattr(r, "status", 500) if r else 500,
                 detail=detail if detail else "Open WebUI: Server Connection Error",
             )
 
@@ -342,6 +399,7 @@ 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()
 
@@ -366,7 +424,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500),
+                status_code=getattr(r, "status", 500) if r else 500,
                 detail=detail if detail else "Open WebUI: Server Connection Error",
             )
 
@@ -377,7 +435,8 @@ async def speech(request: Request, user=Depends(get_verified_user)):
             log.exception(e)
             raise HTTPException(status_code=400, detail="Invalid JSON payload")
 
-        region = request.app.state.config.TTS_AZURE_SPEECH_REGION
+        region = request.app.state.config.TTS_AZURE_SPEECH_REGION or "eastus"
+        base_url = request.app.state.config.TTS_AZURE_SPEECH_BASE_URL
         language = request.app.state.config.TTS_VOICE
         locale = "-".join(request.app.state.config.TTS_VOICE.split("-")[:1])
         output_format = request.app.state.config.TTS_AZURE_SPEECH_OUTPUT_FORMAT
@@ -391,13 +450,15 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 timeout=timeout, trust_env=True
             ) as session:
                 async with session.post(
-                    f"https://{region}.tts.speech.microsoft.com/cognitiveservices/v1",
+                    (base_url or f"https://{region}.tts.speech.microsoft.com")
+                    + "/cognitiveservices/v1",
                     headers={
                         "Ocp-Apim-Subscription-Key": request.app.state.config.TTS_API_KEY,
                         "Content-Type": "application/ssml+xml",
                         "X-Microsoft-OutputFormat": output_format,
                     },
                     data=data,
+                    ssl=AIOHTTP_CLIENT_SESSION_SSL,
                 ) as r:
                     r.raise_for_status()
 
@@ -422,7 +483,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500),
+                status_code=getattr(r, "status", 500) if r else 500,
                 detail=detail if detail else "Open WebUI: Server Connection Error",
             )
 
@@ -466,12 +527,13 @@ 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(
@@ -479,7 +541,12 @@ def transcribe(request: Request, file_path):
             )
 
         model = request.app.state.faster_whisper_model
-        segments, info = model.transcribe(file_path, beam_size=5)
+        segments, info = model.transcribe(
+            file_path,
+            beam_size=5,
+            vad_filter=request.app.state.config.WHISPER_VAD_FILTER,
+            language=metadata.get("language") or WHISPER_LANGUAGE,
+        )
         log.info(
             "Detected language '%s' with probability %f"
             % (info.language, info.language_probability)
@@ -496,11 +563,6 @@ def transcribe(request: Request, file_path):
         log.debug(data)
         return data
     elif request.app.state.config.STT_ENGINE == "openai":
-        if is_mp4_audio(file_path):
-            os.rename(file_path, file_path.replace(".wav", ".mp4"))
-            # Convert MP4 audio file to WAV format
-            convert_mp4_to_wav(file_path.replace(".wav", ".mp4"), file_path)
-
         r = None
         try:
             r = requests.post(
@@ -509,7 +571,14 @@ 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()
@@ -598,36 +667,254 @@ def transcribe(request: Request, file_path):
                     detail = f"External: {e}"
             raise Exception(detail if detail else "Open WebUI: Server Connection Error")
 
+    elif request.app.state.config.STT_ENGINE == "azure":
+        # Check file exists and size
+        if not os.path.exists(file_path):
+            raise HTTPException(status_code=400, detail="Audio file not found")
+
+        # Check file size (Azure has a larger limit of 200MB)
+        file_size = os.path.getsize(file_path)
+        if file_size > AZURE_MAX_FILE_SIZE:
+            raise HTTPException(
+                status_code=400,
+                detail=f"File size exceeds Azure's limit of {AZURE_MAX_FILE_SIZE_MB}MB",
+            )
+
+        api_key = request.app.state.config.AUDIO_STT_AZURE_API_KEY
+        region = request.app.state.config.AUDIO_STT_AZURE_REGION or "eastus"
+        locales = request.app.state.config.AUDIO_STT_AZURE_LOCALES
+        base_url = request.app.state.config.AUDIO_STT_AZURE_BASE_URL
+        max_speakers = request.app.state.config.AUDIO_STT_AZURE_MAX_SPEAKERS or 3
+
+        # IF NO LOCALES, USE DEFAULTS
+        if len(locales) < 2:
+            locales = [
+                "en-US",
+                "es-ES",
+                "es-MX",
+                "fr-FR",
+                "hi-IN",
+                "it-IT",
+                "de-DE",
+                "en-GB",
+                "en-IN",
+                "ja-JP",
+                "ko-KR",
+                "pt-BR",
+                "zh-CN",
+            ]
+            locales = ",".join(locales)
+
+        if not api_key or not region:
+            raise HTTPException(
+                status_code=400,
+                detail="Azure API key is required for Azure STT",
+            )
+
+        r = None
+        try:
+            # Prepare the request
+            data = {
+                "definition": json.dumps(
+                    {
+                        "locales": locales.split(","),
+                        "diarization": {"maxSpeakers": max_speakers, "enabled": True},
+                    }
+                    if locales
+                    else {}
+                )
+            }
+
+            url = (
+                base_url or f"https://{region}.api.cognitive.microsoft.com"
+            ) + "/speechtotext/transcriptions:transcribe?api-version=2024-11-15"
+
+            # Use context manager to ensure file is properly closed
+            with open(file_path, "rb") as audio_file:
+                r = requests.post(
+                    url=url,
+                    files={"audio": audio_file},
+                    data=data,
+                    headers={
+                        "Ocp-Apim-Subscription-Key": api_key,
+                    },
+                )
+
+            r.raise_for_status()
+            response = r.json()
+
+            # Extract transcript from response
+            if not response.get("combinedPhrases"):
+                raise ValueError("No transcription found in response")
+
+            # Get the full transcript from combinedPhrases
+            transcript = response["combinedPhrases"][0].get("text", "").strip()
+            if not transcript:
+                raise ValueError("Empty transcript in response")
+
+            data = {"text": transcript}
+
+            # Save transcript to json file (consistent with other providers)
+            transcript_file = f"{file_dir}/{id}.json"
+            with open(transcript_file, "w") as f:
+                json.dump(data, f)
+
+            log.debug(data)
+            return data
+
+        except (KeyError, IndexError, ValueError) as e:
+            log.exception("Error parsing Azure response")
+            raise HTTPException(
+                status_code=500,
+                detail=f"Failed to parse Azure response: {str(e)}",
+            )
+        except requests.exceptions.RequestException as e:
+            log.exception(e)
+            detail = None
+
+            try:
+                if r is not None and r.status_code != 200:
+                    res = r.json()
+                    if "error" in res:
+                        detail = f"External: {res['error'].get('message', '')}"
+            except Exception:
+                detail = f"External: {e}"
+
+            raise HTTPException(
+                status_code=getattr(r, "status_code", 500) if r else 500,
+                detail=detail if detail else "Open WebUI: Server Connection Error",
+            )
+
+
+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")
-
-    if not file.content_type.startswith(supported_filetypes):
+    SUPPORTED_CONTENT_TYPES = {"video/webm"}  # Extend if you add more video types!
+    if not (
+        file.content_type.startswith("audio/")
+        or file.content_type in SUPPORTED_CONTENT_TYPES
+    ):
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,
             detail=ERROR_MESSAGES.FILE_NOT_SUPPORTED,
@@ -648,19 +935,18 @@ 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)
 
@@ -769,7 +1055,10 @@ def get_available_voices(request) -> dict:
     elif request.app.state.config.TTS_ENGINE == "azure":
         try:
             region = request.app.state.config.TTS_AZURE_SPEECH_REGION
-            url = f"https://{region}.tts.speech.microsoft.com/cognitiveservices/voices/list"
+            base_url = request.app.state.config.TTS_AZURE_SPEECH_BASE_URL
+            url = (
+                base_url or f"https://{region}.tts.speech.microsoft.com"
+            ) + "/cognitiveservices/voices/list"
             headers = {
                 "Ocp-Apim-Subscription-Key": request.app.state.config.TTS_API_KEY
             }

backend/open_webui/routers/auths.py

@@ -19,35 +19,41 @@ 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
+
 from open_webui.utils.misc import parse_duration, validate_email_format
 from open_webui.utils.auth import (
+    decode_token,
     create_api_key,
     create_token,
     get_admin_user,
     get_verified_user,
     get_current_user,
     get_password_hash,
+    get_http_authorization_cred,
 )
 from open_webui.utils.webhook import post_webhook
 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
@@ -72,31 +78,36 @@ class SessionUserResponse(Token, UserResponse):
 async def get_session_user(
     request: Request, response: Response, user=Depends(get_current_user)
 ):
-    expires_delta = parse_duration(request.app.state.config.JWT_EXPIRES_IN)
+
+    auth_header = request.headers.get("Authorization")
+    auth_token = get_http_authorization_cred(auth_header)
+    token = auth_token.credentials
+    data = decode_token(token)
+
     expires_at = None
-    if expires_delta:
-        expires_at = int(time.time()) + int(expires_delta.total_seconds())
 
-    token = create_token(
-        data={"id": user.id},
-        expires_delta=expires_delta,
-    )
+    if data:
+        expires_at = data.get("exp")
 
-    datetime_expires_at = (
-        datetime.datetime.fromtimestamp(expires_at, datetime.timezone.utc)
-        if expires_at
-        else None
-    )
+        if (expires_at is not None) and int(time.time()) > expires_at:
+            raise HTTPException(
+                status_code=status.HTTP_401_UNAUTHORIZED,
+                detail=ERROR_MESSAGES.INVALID_TOKEN,
+            )
 
-    # Set the cookie token
-    response.set_cookie(
-        key="token",
-        value=token,
-        expires=datetime_expires_at,
-        httponly=True,  # Ensures the cookie is not accessible via JavaScript
-        samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
-        secure=WEBUI_AUTH_COOKIE_SECURE,
-    )
+        # Set the cookie token
+        response.set_cookie(
+            key="token",
+            value=token,
+            expires=(
+                datetime.datetime.fromtimestamp(expires_at, datetime.timezone.utc)
+                if expires_at
+                else None
+            ),
+            httponly=True,  # Ensures the cookie is not accessible via JavaScript
+            samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+            secure=WEBUI_AUTH_COOKIE_SECURE,
+        )
 
     user_permissions = get_permissions(
         user.id, request.app.state.config.USER_PERMISSIONS
@@ -177,6 +188,9 @@ 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
@@ -188,7 +202,7 @@ 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,
@@ -225,16 +239,22 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
             ],
         )
 
-        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]
         username = str(entry[f"{LDAP_ATTRIBUTE_FOR_USERNAME}"]).lower()
-        email = str(entry[f"{LDAP_ATTRIBUTE_FOR_MAIL}"])
-        if not email or email == "" or email == "[]":
+        email = entry[
+            f"{LDAP_ATTRIBUTE_FOR_MAIL}"
+        ].value  # retrieve the Attribute value
+        if not email:
             raise HTTPException(400, "User does not have a valid email address.")
-        else:
+        elif isinstance(email, str):
             email = email.lower()
+        elif isinstance(email, list):
+            email = email[0].lower()
+        else:
+            email = str(email).lower()
 
         cn = str(entry["cn"])
         user_dn = entry.entry_dn
@@ -281,21 +301,33 @@ 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)
+                expires_at = None
+                if expires_delta:
+                    expires_at = int(time.time()) + int(expires_delta.total_seconds())
+
                 token = create_token(
                     data={"id": user.id},
-                    expires_delta=parse_duration(
-                        request.app.state.config.JWT_EXPIRES_IN
-                    ),
+                    expires_delta=expires_delta,
                 )
 
                 # Set the cookie token
                 response.set_cookie(
                     key="token",
                     value=token,
+                    expires=(
+                        datetime.datetime.fromtimestamp(
+                            expires_at, datetime.timezone.utc
+                        )
+                        if expires_at
+                        else None
+                    ),
                     httponly=True,  # Ensures the cookie is not accessible via JavaScript
+                    samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+                    secure=WEBUI_AUTH_COOKIE_SECURE,
                 )
 
                 user_permissions = get_permissions(
@@ -305,6 +337,7 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                 return {
                     "token": token,
                     "token_type": "Bearer",
+                    "expires_at": expires_at,
                     "id": user.id,
                     "email": user.email,
                     "name": user.name,
@@ -332,21 +365,29 @@ 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_user_groups_by_group_names(user.id, group_names)
+
     elif WEBUI_AUTH == False:
         admin_email = "admin@localhost"
         admin_password = "admin"
@@ -450,9 +491,12 @@ 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(
+                status.HTTP_400_BAD_REQUEST,
+                detail=ERROR_MESSAGES.PASSWORD_TOO_LONG,
+            )
 
         hashed = get_password_hash(form_data.password)
         user = Auths.insert_new_auth(
@@ -506,6 +550,10 @@ 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",
@@ -539,9 +587,14 @@ async def signout(request: Request, response: Response):
                             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}",
+                                    },
                                     headers=response.headers,
-                                    url=f"{logout_url}?id_token_hint={oauth_id_token}",
                                 )
                         else:
                             raise HTTPException(
@@ -555,7 +608,19 @@ async def signout(request: Request, response: Response):
                     detail="Failed to sign out from the OpenID provider.",
                 )
 
-    return {"status": True}
+    if WEBUI_AUTH_SIGNOUT_REDIRECT_URL:
+        return JSONResponse(
+            status_code=200,
+            content={
+                "status": True,
+                "redirect_url": WEBUI_AUTH_SIGNOUT_REDIRECT_URL,
+            },
+            headers=response.headers,
+        )
+
+    return JSONResponse(
+        status_code=200, content={"status": True}, headers=response.headers
+    )
 
 
 ############################
@@ -653,7 +718,11 @@ async def get_admin_config(request: Request, user=Depends(get_admin_user)):
         "ENABLE_COMMUNITY_SHARING": request.app.state.config.ENABLE_COMMUNITY_SHARING,
         "ENABLE_MESSAGE_RATING": request.app.state.config.ENABLE_MESSAGE_RATING,
         "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,
     }
 
 
@@ -669,7 +738,11 @@ class AdminConfig(BaseModel):
     ENABLE_COMMUNITY_SHARING: bool
     ENABLE_MESSAGE_RATING: bool
     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")
@@ -689,6 +762,7 @@ async def update_admin_config(
     )
 
     request.app.state.config.ENABLE_CHANNELS = form_data.ENABLE_CHANNELS
+    request.app.state.config.ENABLE_NOTES = form_data.ENABLE_NOTES
 
     if form_data.DEFAULT_USER_ROLE in ["pending", "user", "admin"]:
         request.app.state.config.DEFAULT_USER_ROLE = form_data.DEFAULT_USER_ROLE
@@ -706,6 +780,15 @@ 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,
@@ -713,12 +796,16 @@ async def update_admin_config(
         "ENABLE_API_KEY": request.app.state.config.ENABLE_API_KEY,
         "ENABLE_API_KEY_ENDPOINT_RESTRICTIONS": request.app.state.config.ENABLE_API_KEY_ENDPOINT_RESTRICTIONS,
         "API_KEY_ALLOWED_ENDPOINTS": request.app.state.config.API_KEY_ALLOWED_ENDPOINTS,
-        "ENABLE_CHANNELS": request.app.state.config.ENABLE_CHANNELS,
         "DEFAULT_USER_ROLE": request.app.state.config.DEFAULT_USER_ROLE,
         "JWT_EXPIRES_IN": request.app.state.config.JWT_EXPIRES_IN,
         "ENABLE_COMMUNITY_SHARING": request.app.state.config.ENABLE_COMMUNITY_SHARING,
         "ENABLE_MESSAGE_RATING": request.app.state.config.ENABLE_MESSAGE_RATING,
+        "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,
     }
 
 
@@ -734,6 +821,7 @@ class LdapServerConfig(BaseModel):
     search_filters: str = ""
     use_tls: bool = True
     certificate_path: Optional[str] = None
+    validate_cert: bool = True
     ciphers: Optional[str] = "ALL"
 
 
@@ -751,6 +839,7 @@ 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,
     }
 
@@ -786,6 +875,7 @@ 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 {
@@ -800,6 +890,7 @@ 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/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
 
 
 ############################
@@ -579,7 +624,12 @@ async def clone_chat_by_id(
 
 @router.post("/{id}/clone/shared", response_model=Optional[ChatResponse])
 async def clone_shared_chat_by_id(id: str, user=Depends(get_verified_user)):
-    chat = Chats.get_chat_by_share_id(id)
+
+    if user.role == "admin":
+        chat = Chats.get_chat_by_id(id)
+    else:
+        chat = Chats.get_chat_by_share_id(id)
+
     if chat:
         updated_chat = {
             **chat.chat,
@@ -633,8 +683,17 @@ 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(id: str, user=Depends(get_verified_user)):
+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
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
     chat = Chats.get_chat_by_id_and_user_id(id, user.id)
+
     if chat:
         if chat.share_id:
             shared_chat = Chats.update_shared_chat_by_chat_id(chat.id)

backend/open_webui/routers/evaluations.py

@@ -56,7 +56,7 @@ async def update_config(
     }
 
 
-class FeedbackUserReponse(BaseModel):
+class UserResponse(BaseModel):
     id: str
     name: str
     email: str
@@ -68,21 +68,23 @@ class FeedbackUserReponse(BaseModel):
 
 
 class FeedbackUserResponse(FeedbackResponse):
-    user: Optional[FeedbackUserReponse] = None
+    user: Optional[UserResponse] = None
 
 
 @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=FeedbackUserReponse(
-                **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")
@@ -94,12 +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 +1,8 @@
 import logging
 import os
 import uuid
+import json
+from fnmatch import fnmatch
 from pathlib import Path
 from typing import Optional
 from urllib.parse import quote
@@ -9,6 +11,7 @@ from fastapi import (
     APIRouter,
     Depends,
     File,
+    Form,
     HTTPException,
     Request,
     UploadFile,
@@ -18,6 +21,8 @@ from fastapi import (
 from fastapi.responses import FileResponse, StreamingResponse
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import SRC_LOG_LEVELS
+
+from open_webui.models.users import Users
 from open_webui.models.files import (
     FileForm,
     FileModel,
@@ -81,20 +86,55 @@ def has_access_to_file(
 def upload_file(
     request: Request,
     file: UploadFile = File(...),
-    user=Depends(get_verified_user),
-    file_metadata: dict = {},
+    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}")
+
+    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
         filename = f"{id}_{filename}"
-        contents, file_path = Storage.upload_file(file.file, filename)
+        tags = {
+            "OpenWebUI-User-Email": user.email,
+            "OpenWebUI-User-Id": user.id,
+            "OpenWebUI-User-Name": user.name,
+            "OpenWebUI-File-Id": id,
+        }
+        contents, file_path = Storage.upload_file(file.file, filename, tags)
 
         file_item = Files.insert_new_file(
             user.id,
@@ -114,21 +154,26 @@ def upload_file(
         )
         if process:
             try:
-                if file.content_type in [
-                    "audio/mpeg",
-                    "audio/wav",
-                    "audio/ogg",
-                    "audio/x-m4a",
-                ]:
-                    file_path = Storage.get_file(file_path)
-                    result = transcribe(request, file_path)
+                if file.content_type:
+                    if file.content_type.startswith("audio/") or file.content_type in {
+                        "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,
+                        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"]:
                     process_file(request, ProcessFileForm(file_id=id), user=user)
 
                 file_item = Files.get_file_by_id(id=id)
@@ -154,7 +199,7 @@ 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"),
         )
 
 
@@ -172,11 +217,54 @@ async def list_files(user=Depends(get_verified_user), content: bool = Query(True
 
     if not content:
         for file in files:
-            del file.data["content"]
+            if "content" in file.data:
+                del file.data["content"]
 
     return files
 
 
+############################
+# Search Files
+############################
+
+
+@router.get("/search", response_model=list[FileModelResponse])
+async def search_files(
+    filename: str = Query(
+        ...,
+        description="Filename pattern to search for. Supports wildcards such as '*.txt'",
+    ),
+    content: bool = Query(True),
+    user=Depends(get_verified_user),
+):
+    """
+    Search for files by filename with support for wildcard patterns.
+    """
+    # Get files according to user role
+    if user.role == "admin":
+        files = Files.get_files()
+    else:
+        files = Files.get_files_by_user_id(user.id)
+
+    # Get matching files
+    matching_files = [
+        file for file in files if fnmatch(file.filename.lower(), filename.lower())
+    ]
+
+    if not matching_files:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="No files found matching the pattern.",
+        )
+
+    if not content:
+        for file in matching_files:
+            if "content" in file.data:
+                del file.data["content"]
+
+    return matching_files
+
+
 ############################
 # Delete All Files
 ############################
@@ -389,6 +477,13 @@ async def get_html_file_content_by_id(id: str, user=Depends(get_verified_user)):
             detail=ERROR_MESSAGES.NOT_FOUND,
         )
 
+    file_user = Users.get_user_by_id(file.user_id)
+    if not file_user.role == "admin":
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
     if (
         file.user_id == user.id
         or user.role == "admin"

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() 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

@@ -333,10 +333,11 @@ 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 +420,7 @@ 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 +428,7 @@ 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 +451,7 @@ 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 +460,7 @@ 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
 
@@ -500,7 +501,11 @@ async def image_generations(
                     if form_data.size
                     else request.app.state.config.IMAGE_SIZE
                 ),
-                "response_format": "b64_json",
+                **(
+                    {}
+                    if "gpt-image-1" in request.app.state.config.IMAGE_GENERATION_MODEL
+                    else {"response_format": "b64_json"}
+                ),
             }
 
             # Use asyncio.to_thread for the requests.post call
@@ -522,7 +527,7 @@ 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
 
@@ -556,7 +561,7 @@ 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
@@ -607,9 +612,9 @@ 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})
@@ -619,7 +624,7 @@ async def image_generations(
             or request.app.state.config.IMAGE_GENERATION_ENGINE == ""
         ):
             if form_data.model:
-                set_image_model(form_data.model)
+                set_image_model(request, form_data.model)
 
             data = {
                 "prompt": form_data.prompt,
@@ -660,9 +665,9 @@ 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

@@ -9,8 +9,8 @@ from open_webui.models.knowledge import (
     KnowledgeResponse,
     KnowledgeUserResponse,
 )
-from open_webui.models.files import Files, FileModel
-from open_webui.retrieval.vector.connector import VECTOR_DB_CLIENT
+from open_webui.models.files import Files, FileModel, FileMetadataResponse
+from open_webui.retrieval.vector.factory import VECTOR_DB_CLIENT
 from open_webui.routers.retrieval import (
     process_file,
     ProcessFileForm,
@@ -161,13 +161,94 @@ async def create_new_knowledge(
         )
 
 
+############################
+# ReindexKnowledgeFiles
+############################
+
+
+@router.post("/reindex", response_model=bool)
+async def reindex_knowledge_files(request: Request, user=Depends(get_verified_user)):
+    if user.role != "admin":
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    knowledge_bases = Knowledges.get_knowledge_bases()
+
+    log.info(f"Starting reindexing for {len(knowledge_bases)} knowledge bases")
+
+    deleted_knowledge_bases = []
+
+    for knowledge_base in knowledge_bases:
+        # -- Robust error handling for missing or invalid data
+        if not knowledge_base.data or not isinstance(knowledge_base.data, dict):
+            log.warning(
+                f"Knowledge base {knowledge_base.id} has no data or invalid data ({knowledge_base.data!r}). Deleting."
+            )
+            try:
+                Knowledges.delete_knowledge_by_id(id=knowledge_base.id)
+                deleted_knowledge_bases.append(knowledge_base.id)
+            except Exception as e:
+                log.error(
+                    f"Failed to delete invalid knowledge base {knowledge_base.id}: {e}"
+                )
+            continue
+
+        try:
+            file_ids = knowledge_base.data.get("file_ids", [])
+            files = Files.get_files_by_ids(file_ids)
+            try:
+                if VECTOR_DB_CLIENT.has_collection(collection_name=knowledge_base.id):
+                    VECTOR_DB_CLIENT.delete_collection(
+                        collection_name=knowledge_base.id
+                    )
+            except Exception as e:
+                log.error(f"Error deleting collection {knowledge_base.id}: {str(e)}")
+                continue  # Skip, don't raise
+
+            failed_files = []
+            for file in files:
+                try:
+                    process_file(
+                        request,
+                        ProcessFileForm(
+                            file_id=file.id, collection_name=knowledge_base.id
+                        ),
+                        user=user,
+                    )
+                except Exception as e:
+                    log.error(
+                        f"Error processing file {file.filename} (ID: {file.id}): {str(e)}"
+                    )
+                    failed_files.append({"file_id": file.id, "error": str(e)})
+                    continue
+
+        except Exception as e:
+            log.error(f"Error processing knowledge base {knowledge_base.id}: {str(e)}")
+            # Don't raise, just continue
+            continue
+
+        if failed_files:
+            log.warning(
+                f"Failed to process {len(failed_files)} files in knowledge base {knowledge_base.id}"
+            )
+            for failed in failed_files:
+                log.warning(f"File ID: {failed['file_id']}, Error: {failed['error']}")
+
+    log.info(
+        f"Reindexing completed. Deleted {len(deleted_knowledge_bases)} invalid knowledge bases: {deleted_knowledge_bases}"
+    )
+    return True
+
+
 ############################
 # GetKnowledgeById
 ############################
 
 
 class KnowledgeFilesResponse(KnowledgeResponse):
-    files: list[FileModel]
+    files: list[FileMetadataResponse]
 
 
 @router.get("/{id}", response_model=Optional[KnowledgeFilesResponse])
@@ -183,7 +264,7 @@ async def get_knowledge_by_id(id: str, user=Depends(get_verified_user)):
         ):
 
             file_ids = knowledge.data.get("file_ids", []) if knowledge.data else []
-            files = Files.get_files_by_ids(file_ids)
+            files = Files.get_file_metadatas_by_ids(file_ids)
 
             return KnowledgeFilesResponse(
                 **knowledge.model_dump(),
@@ -311,7 +392,7 @@ def add_file_to_knowledge_by_id(
             knowledge = Knowledges.update_knowledge_data_by_id(id=id, data=data)
 
             if knowledge:
-                files = Files.get_files_by_ids(file_ids)
+                files = Files.get_file_metadatas_by_ids(file_ids)
 
                 return KnowledgeFilesResponse(
                     **knowledge.model_dump(),
@@ -388,7 +469,7 @@ def update_file_from_knowledge_by_id(
         data = knowledge.data or {}
         file_ids = data.get("file_ids", [])
 
-        files = Files.get_files_by_ids(file_ids)
+        files = Files.get_file_metadatas_by_ids(file_ids)
 
         return KnowledgeFilesResponse(
             **knowledge.model_dump(),
@@ -470,7 +551,7 @@ def remove_file_from_knowledge_by_id(
             knowledge = Knowledges.update_knowledge_data_by_id(id=id, data=data)
 
             if knowledge:
-                files = Files.get_files_by_ids(file_ids)
+                files = Files.get_file_metadatas_by_ids(file_ids)
 
                 return KnowledgeFilesResponse(
                     **knowledge.model_dump(),
@@ -666,7 +747,7 @@ def add_files_to_knowledge_batch(
         error_details = [f"{err.file_id}: {err.error}" for err in result.errors]
         return KnowledgeFilesResponse(
             **knowledge.model_dump(),
-            files=Files.get_files_by_ids(existing_file_ids),
+            files=Files.get_file_metadatas_by_ids(existing_file_ids),
             warnings={
                 "message": "Some files failed to process",
                 "errors": error_details,
@@ -674,5 +755,6 @@ def add_files_to_knowledge_batch(
         )
 
     return KnowledgeFilesResponse(
-        **knowledge.model_dump(), files=Files.get_files_by_ids(existing_file_ids)
+        **knowledge.model_dump(),
+        files=Files.get_file_metadatas_by_ids(existing_file_ids),
     )

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
 
@@ -153,7 +153,9 @@ async def update_memory_by_id(
     form_data: MemoryUpdateModel,
     user=Depends(get_verified_user),
 ):
-    memory = Memories.update_memory_by_id(memory_id, form_data.content)
+    memory = Memories.update_memory_by_id_and_user_id(
+        memory_id, user.id, form_data.content
+    )
     if memory is None:
         raise HTTPException(status_code=404, detail="Memory not found")
 

backend/open_webui/routers/notes.py

@@ -0,0 +1,215 @@
+import json
+import logging
+from typing import Optional
+
+
+from fastapi import APIRouter, Depends, HTTPException, Request, status, BackgroundTasks
+from pydantic import BaseModel
+
+from open_webui.models.users import Users, UserResponse
+from open_webui.models.notes import Notes, NoteModel, NoteForm, NoteUserResponse
+
+from open_webui.config import ENABLE_ADMIN_CHAT_ACCESS, ENABLE_ADMIN_EXPORT
+from open_webui.constants import ERROR_MESSAGES
+from open_webui.env import SRC_LOG_LEVELS
+
+
+from open_webui.utils.auth import get_admin_user, get_verified_user
+from open_webui.utils.access_control import has_access, has_permission
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["MODELS"])
+
+router = APIRouter()
+
+############################
+# GetNotes
+############################
+
+
+@router.get("/", response_model=list[NoteUserResponse])
+async def get_notes(request: Request, user=Depends(get_verified_user)):
+
+    if user.role != "admin" and not has_permission(
+        user.id, "features.notes", request.app.state.config.USER_PERMISSIONS
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    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, "write")
+    ]
+
+    return notes
+
+
+@router.get("/list", response_model=list[NoteUserResponse])
+async def get_note_list(request: Request, user=Depends(get_verified_user)):
+
+    if user.role != "admin" and not has_permission(
+        user.id, "features.notes", request.app.state.config.USER_PERMISSIONS
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    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")
+    ]
+
+    return notes
+
+
+############################
+# CreateNewNote
+############################
+
+
+@router.post("/create", response_model=Optional[NoteModel])
+async def create_new_note(
+    request: Request, form_data: NoteForm, user=Depends(get_verified_user)
+):
+
+    if user.role != "admin" and not has_permission(
+        user.id, "features.notes", request.app.state.config.USER_PERMISSIONS
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    try:
+        note = Notes.insert_new_note(form_data, user.id)
+        return note
+    except Exception as e:
+        log.exception(e)
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.DEFAULT()
+        )
+
+
+############################
+# GetNoteById
+############################
+
+
+@router.get("/{id}", response_model=Optional[NoteModel])
+async def get_note_by_id(request: Request, id: str, user=Depends(get_verified_user)):
+    if user.role != "admin" and not has_permission(
+        user.id, "features.notes", request.app.state.config.USER_PERMISSIONS
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    note = Notes.get_note_by_id(id)
+    if not note:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
+        )
+
+    if user.role != "admin" or (
+        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()
+        )
+
+    return note
+
+
+############################
+# UpdateNoteById
+############################
+
+
+@router.post("/{id}/update", response_model=Optional[NoteModel])
+async def update_note_by_id(
+    request: Request, id: str, form_data: NoteForm, user=Depends(get_verified_user)
+):
+    if user.role != "admin" and not has_permission(
+        user.id, "features.notes", request.app.state.config.USER_PERMISSIONS
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    note = Notes.get_note_by_id(id)
+    if not note:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
+        )
+
+    if user.role != "admin" or (
+        user.id != note.user_id
+        and not has_access(user.id, type="write", access_control=note.access_control)
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
+        )
+
+    try:
+        note = Notes.update_note_by_id(id, form_data)
+        return note
+    except Exception as e:
+        log.exception(e)
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.DEFAULT()
+        )
+
+
+############################
+# DeleteNoteById
+############################
+
+
+@router.delete("/{id}/delete", response_model=bool)
+async def delete_note_by_id(request: Request, id: str, user=Depends(get_verified_user)):
+    if user.role != "admin" and not has_permission(
+        user.id, "features.notes", request.app.state.config.USER_PERMISSIONS
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    note = Notes.get_note_by_id(id)
+    if not note:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
+        )
+
+    if user.role != "admin" or (
+        user.id != note.user_id
+        and not has_access(user.id, type="write", access_control=note.access_control)
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.DEFAULT()
+        )
+
+    try:
+        note = Notes.delete_note_by_id(id)
+        return True
+    except Exception as e:
+        log.exception(e)
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.DEFAULT()
+        )

backend/open_webui/routers/ollama.py

@@ -9,6 +9,8 @@ import os
 import random
 import re
 import time
+from datetime import datetime
+
 from typing import Optional, Union
 from urllib.parse import urlparse
 import aiohttp
@@ -54,6 +56,7 @@ from open_webui.config import (
 from open_webui.env import (
     ENV,
     SRC_LOG_LEVELS,
+    AIOHTTP_CLIENT_SESSION_SSL,
     AIOHTTP_CLIENT_TIMEOUT,
     AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST,
     BYPASS_MODEL_ACCESS_CONTROL,
@@ -91,6 +94,7 @@ async def send_get_request(url, key=None, user: UserModel = None):
                         else {}
                     ),
                 },
+                ssl=AIOHTTP_CLIENT_SESSION_SSL,
             ) as response:
                 return await response.json()
     except Exception as e:
@@ -141,6 +145,7 @@ async def send_post_request(
                     else {}
                 ),
             },
+            ssl=AIOHTTP_CLIENT_SESSION_SSL,
         )
         r.raise_for_status()
 
@@ -216,7 +221,8 @@ async def verify_connection(
     key = form_data.key
 
     async with aiohttp.ClientSession(
-        timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST)
+        trust_env=True,
+        timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST),
     ) as session:
         try:
             async with session.get(
@@ -234,6 +240,7 @@ async def verify_connection(
                         else {}
                     ),
                 },
+                ssl=AIOHTTP_CLIENT_SESSION_SSL,
             ) as r:
                 if r.status != 200:
                     detail = f"HTTP Error: {r.status}"
@@ -295,6 +302,22 @@ async def update_config(
     }
 
 
+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=1)
 async def get_all_models(request: Request, user: UserModel = None):
     log.info("get_all_models()")
@@ -335,6 +358,8 @@ 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", [])
@@ -347,31 +372,18 @@ 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,
@@ -379,6 +391,22 @@ 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": []}
 
@@ -459,6 +487,68 @@ 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):
@@ -532,36 +622,74 @@ 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),
-                user=user,
-            )
-            for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS)
-        ]
-        responses = await asyncio.gather(*request_tasks)
-
-        return dict(zip(request.app.state.config.OLLAMA_BASE_URLS, responses))
-    else:
-        return {}
-
-
 class ModelNameForm(BaseModel):
     name: str
 
 
+@router.post("/api/unload")
+async def unload_model(
+    request: Request,
+    form_data: ModelNameForm,
+    user=Depends(get_admin_user),
+):
+    model_name = form_data.name
+    if not model_name:
+        raise HTTPException(
+            status_code=400, detail="Missing 'name' of 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,
+            )
+            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)})
+
+    if len(errors) > 0:
+        raise HTTPException(
+            status_code=500,
+            detail=f"Failed to unload model on {len(errors)} nodes: {errors}",
+        )
+
+    return {"status": True}
+
+
 @router.post("/api/pull")
 @router.post("/api/pull/{url_idx}")
 async def pull_model(
@@ -878,8 +1006,16 @@ async def embed(
             )
 
     url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
+    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
+        str(url_idx),
+        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
+    )
     key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)
 
+    prefix_id = api_config.get("prefix_id", None)
+    if prefix_id:
+        form_data.model = form_data.model.replace(f"{prefix_id}.", "")
+
     try:
         r = requests.request(
             method="POST",
@@ -957,8 +1093,16 @@ async def embeddings(
             )
 
     url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
+    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
+        str(url_idx),
+        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
+    )
     key = get_api_key(url_idx, url, request.app.state.config.OLLAMA_API_CONFIGS)
 
+    prefix_id = api_config.get("prefix_id", None)
+    if prefix_id:
+        form_data.model = form_data.model.replace(f"{prefix_id}.", "")
+
     try:
         r = requests.request(
             method="POST",
@@ -1006,7 +1150,7 @@ class GenerateCompletionForm(BaseModel):
     prompt: str
     suffix: Optional[str] = None
     images: Optional[list[str]] = None
-    format: Optional[str] = None
+    format: Optional[Union[dict, str]] = None
     options: Optional[dict] = None
     system: Optional[str] = None
     template: Optional[str] = None
@@ -1088,6 +1232,9 @@ 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):
@@ -1125,7 +1272,9 @@ 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"]
 
@@ -1139,13 +1288,10 @@ 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":
@@ -1178,7 +1324,7 @@ 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),
@@ -1327,8 +1473,10 @@ 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":
@@ -1482,7 +1630,9 @@ async def download_file_stream(
     timeout = aiohttp.ClientTimeout(total=600)  # Set the timeout
 
     async with aiohttp.ClientSession(timeout=timeout, trust_env=True) as session:
-        async with session.get(file_url, headers=headers) as response:
+        async with session.get(
+            file_url, headers=headers, ssl=AIOHTTP_CLIENT_SESSION_SSL
+        ) as response:
             total_size = int(response.headers.get("content-length", 0)) + current_size
 
             with open(file_path, "ab+") as file:
@@ -1497,7 +1647,8 @@ async def download_file_stream(
 
                 if done:
                     file.seek(0)
-                    hashed = calculate_sha256(file)
+                    chunk_size = 1024 * 1024 * 2
+                    hashed = calculate_sha256(file, chunk_size)
                     file.seek(0)
 
                     url = f"{ollama_url}/api/blobs/sha256:{hashed}"
@@ -1561,7 +1712,9 @@ 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 ---
@@ -1606,13 +1759,13 @@ 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
 
@@ -1629,7 +1782,7 @@ 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

@@ -21,6 +21,7 @@ from open_webui.config import (
     CACHE_DIR,
 )
 from open_webui.env import (
+    AIOHTTP_CLIENT_SESSION_SSL,
     AIOHTTP_CLIENT_TIMEOUT,
     AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST,
     ENABLE_FORWARD_USER_INFO_HEADERS,
@@ -74,6 +75,7 @@ async def send_get_request(url, key=None, user: UserModel = None):
                         else {}
                     ),
                 },
+                ssl=AIOHTTP_CLIENT_SESSION_SSL,
             ) as response:
                 return await response.json()
     except Exception as e:
@@ -92,20 +94,19 @@ async def cleanup_response(
         await session.close()
 
 
-def openai_o1_o3_handler(payload):
+def openai_o_series_handler(payload):
     """
-    Handle o1, o3 specific parameters
+    Handle "o" series specific parameters
     """
     if "max_tokens" in payload:
-        # Remove "max_tokens" from the payload
+        # Convert "max_tokens" to "max_completion_tokens" for all o-series models
         payload["max_completion_tokens"] = payload["max_tokens"]
         del payload["max_tokens"]
 
-    # Fix: o1 and o3 do not support the "system" role directly.
-    # For older models like "o1-mini" or "o1-preview", use role "user".
-    # For newer o1/o3 models, replace "system" with "developer".
+    # Handle system role conversion based on model type
     if payload["messages"][0]["role"] == "system":
         model_lower = payload["model"].lower()
+        # Legacy models use "user" role instead of "system"
         if model_lower.startswith("o1-mini") or model_lower.startswith("o1-preview"):
             payload["messages"][0]["role"] = "user"
         else:
@@ -352,21 +353,22 @@ 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
 
@@ -414,6 +416,7 @@ 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
@@ -460,58 +463,74 @@ 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(
-            timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST)
+            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 {}
-                        ),
-                    },
-                ) 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": 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 {}
+                    ),
+                }
 
-                    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)}")
@@ -533,6 +552,8 @@ class ConnectionVerificationForm(BaseModel):
     url: str
     key: str
 
+    config: Optional[dict] = None
+
 
 @router.post("/verify")
 async def verify_connection(
@@ -541,37 +562,64 @@ async def verify_connection(
     url = form_data.url
     key = form_data.key
 
+    api_config = form_data.config or {}
+
     async with aiohttp.ClientSession(
-        timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST)
+        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 {}
-                    ),
-                },
-            ) 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": 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 {}
+                ),
+            }
 
-                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
@@ -585,6 +633,63 @@ async def verify_connection(
             raise HTTPException(status_code=500, detail=error_detail)
 
 
+def convert_to_azure_payload(
+    url,
+    payload: dict,
+):
+    model = payload.get("model", "")
+
+    # Filter allowed parameters based on Azure OpenAI API
+    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",
+    }
+
+    # 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,
@@ -610,8 +715,12 @@ 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":
@@ -666,10 +775,10 @@ async def generate_chat_completion(
     url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
     key = request.app.state.config.OPENAI_API_KEYS[idx]
 
-    # Fix: o1,o3 does not support the "max_tokens" parameter, Modify "max_tokens" to "max_completion_tokens"
-    is_o1_o3 = payload["model"].lower().startswith(("o1", "o3-"))
-    if is_o1_o3:
-        payload = openai_o1_o3_handler(payload)
+    # Check if model is from "o" series
+    is_o_series = payload["model"].lower().startswith(("o1", "o3", "o4"))
+    if is_o_series:
+        payload = openai_o_series_handler(payload)
     elif "api.openai.com" not in url:
         # Remove "max_completion_tokens" from the payload for backward compatibility
         if "max_completion_tokens" in payload:
@@ -685,6 +794,38 @@ 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": 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 {}
+        ),
+    }
+
+    if api_config.get("azure", False):
+        request_url, payload = convert_to_azure_payload(url, payload)
+        api_version = api_config.get("api_version", "") or "2023-03-15-preview"
+        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
@@ -699,30 +840,10 @@ 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,
         )
 
         # Check if response is SSE
@@ -766,27 +887,37 @@ async def generate_chat_completion(
             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)):
+async def embeddings(request: Request, form_data: dict, user):
     """
-    Deprecated: proxy all requests to OpenAI API
+    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.
     """
-
-    body = await request.body()
-
     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=request.method,
-            url=f"{url}/{path}",
+            method="POST",
+            url=f"{url}/embeddings",
             data=body,
             headers={
                 "Authorization": f"Bearer {key}",
@@ -798,12 +929,107 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
                         "X-OpenWebUI-User-Email": user.email,
                         "X-OpenWebUI-User-Role": user.role,
                     }
-                    if ENABLE_FORWARD_USER_INFO_HEADERS
+                    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)):
+    """
+    Deprecated: proxy all requests to OpenAI API
+    """
+
+    body = await request.body()
+
+    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": 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 {}
+            ),
+        }
+
+        if api_config.get("azure", False):
+            headers["api-key"] = key
+            headers["api-version"] = (
+                api_config.get("api_version", "") or "2023-03-15-preview"
+            )
+
+            payload = json.loads(body)
+            url, payload = convert_to_azure_payload(url, payload)
+            body = json.dumps(payload).encode()
+
+            request_url = f"{url}/{path}?api-version={api_config.get('api_version', '2023-03-15-preview')}"
+        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=request_url,
+            data=body,
+            headers=headers,
+            ssl=AIOHTTP_CLIENT_SESSION_SSL,
+        )
+        r.raise_for_status()
 
         # Check if response is SSE
         if "text/event-stream" in r.headers.get("Content-Type", ""):

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
 
@@ -66,10 +66,13 @@ async def process_pipeline_inlet_filter(request, payload, user, models):
     if "pipeline" in model:
         sorted_filters.append(model)
 
-    async with aiohttp.ClientSession() as session:
+    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()
@@ -115,10 +119,13 @@ async def process_pipeline_outlet_filter(request, payload, user, models):
     if "pipeline" in model:
         sorted_filters = [model] + sorted_filters
 
-    async with aiohttp.ClientSession() as session:
+    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/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,
@@ -10,17 +13,18 @@ from open_webui.models.tools import (
     ToolUserResponse,
     Tools,
 )
-from open_webui.utils.plugin import load_tools_module_by_id, replace_imports
+from open_webui.utils.plugin import load_tool_module_by_id, replace_imports
 from open_webui.config import CACHE_DIR
 from open_webui.constants import ERROR_MESSAGES
 from fastapi import APIRouter, Depends, HTTPException, Request, status
-from open_webui.utils.tools import get_tools_specs
+from open_webui.utils.tools import get_tool_specs
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.access_control import has_access, has_permission
 from open_webui.env import SRC_LOG_LEVELS
 
 from open_webui.utils.tools import get_tool_servers_data
 
+
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
@@ -45,22 +49,22 @@ async def get_tools(request: Request, user=Depends(get_verified_user)):
         )
 
     tools = Tools.get_tools()
-    for idx, server in enumerate(request.app.state.TOOL_SERVERS):
+    for server in request.app.state.TOOL_SERVERS:
         tools.append(
             ToolUserResponse(
                 **{
                     "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", ""),
                     },
                     "access_control": request.app.state.config.TOOL_SERVER_CONNECTIONS[
-                        idx
+                        server["idx"]
                     ]
                     .get("config", {})
                     .get("access_control", None),
@@ -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() 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
 ############################
@@ -137,15 +216,15 @@ async def create_new_tools(
     if tools is None:
         try:
             form_data.content = replace_imports(form_data.content)
-            tools_module, frontmatter = load_tools_module_by_id(
+            tool_module, frontmatter = load_tool_module_by_id(
                 form_data.id, content=form_data.content
             )
             form_data.meta.manifest = frontmatter
 
             TOOLS = request.app.state.TOOLS
-            TOOLS[form_data.id] = tools_module
+            TOOLS[form_data.id] = tool_module
 
-            specs = get_tools_specs(TOOLS[form_data.id])
+            specs = get_tool_specs(TOOLS[form_data.id])
             tools = Tools.insert_new_tool(user.id, form_data, specs)
 
             tool_cache_dir = CACHE_DIR / "tools" / form_data.id
@@ -226,15 +305,13 @@ async def update_tools_by_id(
 
     try:
         form_data.content = replace_imports(form_data.content)
-        tools_module, frontmatter = load_tools_module_by_id(
-            id, content=form_data.content
-        )
+        tool_module, frontmatter = load_tool_module_by_id(id, content=form_data.content)
         form_data.meta.manifest = frontmatter
 
         TOOLS = request.app.state.TOOLS
-        TOOLS[id] = tools_module
+        TOOLS[id] = tool_module
 
-        specs = get_tools_specs(TOOLS[id])
+        specs = get_tool_specs(TOOLS[id])
 
         updated = {
             **form_data.model_dump(exclude={"id"}),
@@ -332,7 +409,7 @@ async def get_tools_valves_spec_by_id(
         if id in request.app.state.TOOLS:
             tools_module = request.app.state.TOOLS[id]
         else:
-            tools_module, _ = load_tools_module_by_id(id)
+            tools_module, _ = load_tool_module_by_id(id)
             request.app.state.TOOLS[id] = tools_module
 
         if hasattr(tools_module, "Valves"):
@@ -375,7 +452,7 @@ async def update_tools_valves_by_id(
     if id in request.app.state.TOOLS:
         tools_module = request.app.state.TOOLS[id]
     else:
-        tools_module, _ = load_tools_module_by_id(id)
+        tools_module, _ = load_tool_module_by_id(id)
         request.app.state.TOOLS[id] = tools_module
 
     if not hasattr(tools_module, "Valves"):
@@ -431,7 +508,7 @@ async def get_tools_user_valves_spec_by_id(
         if id in request.app.state.TOOLS:
             tools_module = request.app.state.TOOLS[id]
         else:
-            tools_module, _ = load_tools_module_by_id(id)
+            tools_module, _ = load_tool_module_by_id(id)
             request.app.state.TOOLS[id] = tools_module
 
         if hasattr(tools_module, "UserValves"):
@@ -455,7 +532,7 @@ async def update_tools_user_valves_by_id(
         if id in request.app.state.TOOLS:
             tools_module = request.app.state.TOOLS[id]
         else:
-            tools_module, _ = load_tools_module_by_id(id)
+            tools_module, _ = load_tool_module_by_id(id)
             request.app.state.TOOLS[id] = tools_module
 
         if hasattr(tools_module, "UserValves"):

backend/open_webui/routers/users.py

@@ -6,6 +6,7 @@ from open_webui.models.groups import Groups
 from open_webui.models.chats import Chats
 from open_webui.models.users import (
     UserModel,
+    UserListResponse,
     UserRoleUpdateForm,
     Users,
     UserSettings,
@@ -20,7 +21,7 @@ from fastapi import APIRouter, Depends, HTTPException, Request, status
 from pydantic import BaseModel
 
 from open_webui.utils.auth import get_admin_user, get_password_hash, get_verified_user
-from open_webui.utils.access_control import get_permissions
+from open_webui.utils.access_control import get_permissions, has_permission
 
 
 log = logging.getLogger(__name__)
@@ -33,13 +34,38 @@ router = APIRouter()
 ############################
 
 
-@router.get("/", response_model=list[UserModel])
+PAGE_ITEM_COUNT = 30
+
+
+@router.get("/", response_model=UserListResponse)
 async def get_users(
-    skip: Optional[int] = None,
-    limit: Optional[int] = None,
+    query: Optional[str] = None,
+    order_by: Optional[str] = None,
+    direction: Optional[str] = None,
+    page: Optional[int] = 1,
     user=Depends(get_admin_user),
 ):
-    return Users.get_users(skip, limit)
+    limit = PAGE_ITEM_COUNT
+
+    page = max(1, page)
+    skip = (page - 1) * limit
+
+    filter = {}
+    if query:
+        filter["query"] = query
+    if order_by:
+        filter["order_by"] = order_by
+    if direction:
+        filter["direction"] = direction
+
+    return Users.get_users(filter=filter, skip=skip, limit=limit)
+
+
+@router.get("/all", response_model=UserListResponse)
+async def get_all_users(
+    user=Depends(get_admin_user),
+):
+    return Users.get_users()
 
 
 ############################
@@ -88,6 +114,12 @@ class ChatPermissions(BaseModel):
     file_upload: bool = True
     delete: bool = True
     edit: bool = True
+    share: bool = True
+    export: bool = True
+    stt: bool = True
+    tts: bool = True
+    call: bool = True
+    multiple_models: bool = True
     temporary: bool = True
     temporary_enforced: bool = False
 
@@ -97,6 +129,7 @@ class FeaturesPermissions(BaseModel):
     web_search: bool = True
     image_generation: bool = True
     code_interpreter: bool = True
+    notes: bool = True
 
 
 class UserPermissions(BaseModel):
@@ -132,22 +165,6 @@ 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
 ############################
@@ -172,9 +189,22 @@ async def get_user_settings_by_session_user(user=Depends(get_verified_user)):
 
 @router.post("/user/settings/update", response_model=UserSettings)
 async def update_user_settings_by_session_user(
-    form_data: UserSettings, user=Depends(get_verified_user)
+    request: Request, form_data: UserSettings, user=Depends(get_verified_user)
 ):
-    user = Users.update_user_settings_by_id(user.id, form_data.model_dump())
+    updated_user_settings = form_data.model_dump()
+    if (
+        user.role != "admin"
+        and "toolServers" in updated_user_settings.get("ui").keys()
+        and not has_permission(
+            user.id,
+            "features.direct_tool_servers",
+            request.app.state.config.USER_PERMISSIONS,
+        )
+    ):
+        # If the user is not an admin and does not have permission to use tool servers, remove the key
+        updated_user_settings["ui"].pop("toolServers", None)
+
+    user = Users.update_user_settings_by_id(user.id, updated_user_settings)
     if user:
         return user.settings
     else:
@@ -284,6 +314,32 @@ async def update_user_by_id(
     form_data: UserUpdateForm,
     session_user=Depends(get_admin_user),
 ):
+    # Prevent modification of the primary admin user by other admins
+    try:
+        first_user = Users.get_first_user()
+        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(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Could not verify primary admin status.",
+        )
+
     user = Users.get_user_by_id(user_id)
 
     if user:
@@ -304,6 +360,7 @@ 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,
@@ -331,6 +388,21 @@ async def update_user_by_id(
 
 @router.delete("/{user_id}", response_model=bool)
 async def delete_user_by_id(user_id: str, user=Depends(get_admin_user)):
+    # Prevent deletion of the primary admin user
+    try:
+        first_user = Users.get_first_user()
+        if first_user and user_id == first_user.id:
+            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(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="Could not verify primary admin status.",
+        )
+
     if user.id != user_id:
         result = Auths.delete_auth_by_id(user_id)
 
@@ -342,6 +414,7 @@ async def delete_user_by_id(user_id: str, user=Depends(get_admin_user)):
             detail=ERROR_MESSAGES.DELETE_USER_ERROR,
         )
 
+    # Prevent self-deletion
     raise HTTPException(
         status_code=status.HTTP_403_FORBIDDEN,
         detail=ERROR_MESSAGES.ACTION_PROHIBITED,

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

@@ -9,9 +9,8 @@ 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.utils.redis import (
-    parse_redis_sentinel_url,
     get_sentinels_from_env,
-    AsyncRedisSentinelManager,
+    get_sentinel_url_from_env,
 )
 
 from open_webui.env import (
@@ -38,15 +37,10 @@ log.setLevel(SRC_LOG_LEVELS["SOCKET"])
 
 if WEBSOCKET_MANAGER == "redis":
     if WEBSOCKET_SENTINEL_HOSTS:
-        redis_config = parse_redis_sentinel_url(WEBSOCKET_REDIS_URL)
-        mgr = AsyncRedisSentinelManager(
-            WEBSOCKET_SENTINEL_HOSTS.split(","),
-            sentinel_port=int(WEBSOCKET_SENTINEL_PORT),
-            redis_port=redis_config["port"],
-            service=redis_config["service"],
-            db=redis_config["db"],
-            username=redis_config["username"],
-            password=redis_config["password"],
+        mgr = socketio.AsyncRedisManager(
+            get_sentinel_url_from_env(
+                WEBSOCKET_REDIS_URL, WEBSOCKET_SENTINEL_HOSTS, WEBSOCKET_SENTINEL_PORT
+            )
         )
     else:
         mgr = socketio.AsyncRedisManager(WEBSOCKET_REDIS_URL)
@@ -165,18 +159,19 @@ def get_models_in_use():
 
 @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},
-    }
+        # 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()})
+        # Broadcast the usage data to all clients
+        await sio.emit("usage", {"models": get_models_in_use()})
 
 
 @sio.event
@@ -284,7 +279,8 @@ 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())})
+    if sid in SESSION_POOL:
+        await sio.emit("user-list", {"user_ids": list(USER_POOL.keys())})
 
 
 @sio.event
@@ -320,8 +316,8 @@ def get_event_emitter(request_info, update_db=True):
             )
         )
 
-        for session_id in session_ids:
-            await sio.emit(
+        emit_tasks = [
+            sio.emit(
                 "chat-events",
                 {
                     "chat_id": request_info.get("chat_id", None),
@@ -330,6 +326,10 @@ def get_event_emitter(request_info, update_db=True):
                 },
                 to=session_id,
             )
+            for session_id in session_ids
+        ]
+
+        await asyncio.gather(*emit_tasks)
 
         if update_db:
             if "type" in event_data and event_data["type"] == "status":
@@ -345,16 +345,17 @@ def get_event_emitter(request_info, update_db=True):
                     request_info["message_id"],
                 )
 
-                content = message.get("content", "")
-                content += event_data.get("data", {}).get("content", "")
+                if message:
+                    content = message.get("content", "")
+                    content += event_data.get("data", {}).get("content", "")
 
-                Chats.upsert_message_to_chat_by_id_and_message_id(
-                    request_info["chat_id"],
-                    request_info["message_id"],
-                    {
-                        "content": content,
-                    },
-                )
+                    Chats.upsert_message_to_chat_by_id_and_message_id(
+                        request_info["chat_id"],
+                        request_info["message_id"],
+                        {
+                            "content": content,
+                        },
+                    )
 
             if "type" in event_data and event_data["type"] == "replace":
                 content = event_data.get("data", {}).get("content", "")

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

@@ -2,8 +2,9 @@ import os
 import shutil
 import json
 import logging
+import re
 from abc import ABC, abstractmethod
-from typing import BinaryIO, Tuple
+from typing import BinaryIO, Tuple, Dict
 
 import boto3
 from botocore.config import Config
@@ -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,
@@ -44,7 +46,9 @@ class StorageProvider(ABC):
         pass
 
     @abstractmethod
-    def upload_file(self, file: BinaryIO, filename: str) -> Tuple[bytes, str]:
+    def upload_file(
+        self, file: BinaryIO, filename: str, tags: Dict[str, str]
+    ) -> Tuple[bytes, str]:
         pass
 
     @abstractmethod
@@ -58,7 +62,9 @@ class StorageProvider(ABC):
 
 class LocalStorageProvider(StorageProvider):
     @staticmethod
-    def upload_file(file: BinaryIO, filename: str) -> Tuple[bytes, str]:
+    def upload_file(
+        file: BinaryIO, filename: str, tags: Dict[str, str]
+    ) -> Tuple[bytes, str]:
         contents = file.read()
         if not contents:
             raise ValueError(ERROR_MESSAGES.EMPTY_CONTENT)
@@ -131,15 +137,37 @@ class S3StorageProvider(StorageProvider):
         self.bucket_name = S3_BUCKET_NAME
         self.key_prefix = S3_KEY_PREFIX if S3_KEY_PREFIX else ""
 
-    def upload_file(self, file: BinaryIO, filename: str) -> Tuple[bytes, str]:
+    @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)
+        _, file_path = LocalStorageProvider.upload_file(file, filename, tags)
+        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)
+            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}")
@@ -207,9 +235,11 @@ class GCSStorageProvider(StorageProvider):
             self.gcs_client = storage.Client()
         self.bucket = self.gcs_client.bucket(GCS_BUCKET_NAME)
 
-    def upload_file(self, file: BinaryIO, filename: str) -> Tuple[bytes, str]:
+    def upload_file(
+        self, file: BinaryIO, filename: str, tags: Dict[str, str]
+    ) -> Tuple[bytes, str]:
         """Handles uploading of the file to GCS storage."""
-        contents, file_path = LocalStorageProvider.upload_file(file, filename)
+        contents, file_path = LocalStorageProvider.upload_file(file, filename, tags)
         try:
             blob = self.bucket.blob(filename)
             blob.upload_from_filename(file_path)
@@ -277,9 +307,11 @@ class AzureStorageProvider(StorageProvider):
             self.container_name
         )
 
-    def upload_file(self, file: BinaryIO, filename: str) -> Tuple[bytes, str]:
+    def upload_file(
+        self, file: BinaryIO, filename: str, tags: Dict[str, str]
+    ) -> Tuple[bytes, str]:
         """Handles uploading of the file to Azure Blob Storage."""
-        contents, file_path = LocalStorageProvider.upload_file(file, filename)
+        contents, file_path = LocalStorageProvider.upload_file(file, filename, tags)
         try:
             blob_client = self.container_client.get_blob_client(filename)
             blob_client.upload_blob(contents, overwrite=True)

backend/open_webui/tasks.py

@@ -2,19 +2,97 @@
 import asyncio
 from typing import Dict
 from uuid import uuid4
+import json
+from redis.asyncio import Redis
+from fastapi import Request
+from typing import Dict, List, Optional
 
 # A dictionary to keep track of active tasks
 tasks: Dict[str, asyncio.Task] = {}
+chat_tasks = {}
 
 
-def cleanup_task(task_id: str):
+REDIS_TASKS_KEY = "open-webui:tasks"
+REDIS_CHAT_TASKS_KEY = "open-webui:tasks:chat"
+REDIS_PUBSUB_CHANNEL = "open-webui:tasks:commands"
+
+
+def is_redis(request: Request) -> bool:
+    # Called everywhere a request is available to check Redis
+    return hasattr(request.app.state, "redis") and (request.app.state.redis is not None)
+
+
+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:
+            print(f"Error handling distributed task command: {e}")
+
+
+### ------------------------------
+### REDIS-ENABLED HANDLERS
+### ------------------------------
+
+
+async def redis_save_task(redis: Redis, task_id: str, chat_id: Optional[str]):
+    pipe = redis.pipeline()
+    pipe.hset(REDIS_TASKS_KEY, task_id, chat_id or "")
+    if chat_id:
+        pipe.sadd(f"{REDIS_CHAT_TASKS_KEY}:{chat_id}", task_id)
+    await pipe.execute()
+
+
+async def redis_cleanup_task(redis: Redis, task_id: str, chat_id: Optional[str]):
+    pipe = redis.pipeline()
+    pipe.hdel(REDIS_TASKS_KEY, task_id)
+    if chat_id:
+        pipe.srem(f"{REDIS_CHAT_TASKS_KEY}:{chat_id}", task_id)
+        if (await pipe.scard(f"{REDIS_CHAT_TASKS_KEY}:{chat_id}").execute())[-1] == 0:
+            pipe.delete(f"{REDIS_CHAT_TASKS_KEY}:{chat_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_chat_tasks(redis: Redis, chat_id: str) -> List[str]:
+    return list(await redis.smembers(f"{REDIS_CHAT_TASKS_KEY}:{chat_id}"))
+
+
+async def redis_send_command(redis: Redis, command: dict):
+    await redis.publish(REDIS_PUBSUB_CHANNEL, json.dumps(command))
+
+
+async def cleanup_task(request, task_id: str, id=None):
     """
     Remove a completed or canceled task from the global `tasks` dictionary.
     """
+    if is_redis(request):
+        await redis_cleanup_task(request.app.state.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)
 
-def create_task(coroutine):
+
+async def create_task(request, coroutine, id=None):
     """
     Create a new asyncio task and add it to the global task dictionary.
     """
@@ -22,30 +100,57 @@ def create_task(coroutine):
     task = asyncio.create_task(coroutine)  # Create the task
 
     # Add a done callback for cleanup
-    task.add_done_callback(lambda t: cleanup_task(task_id))
-
+    task.add_done_callback(
+        lambda t: asyncio.create_task(cleanup_task(request, 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)
+    else:
+        chat_tasks[id] = [task_id]
+
+    if is_redis(request):
+        await redis_save_task(request.app.state.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(request):
     """
     List all currently active task IDs.
     """
+    if is_redis(request):
+        return await redis_list_tasks(request.app.state.redis)
     return list(tasks.keys())
 
 
-async def stop_task(task_id: str):
+async def list_task_ids_by_chat_id(request, id):
+    """
+    List all tasks associated with a specific ID.
+    """
+    if is_redis(request):
+        return await redis_list_chat_tasks(request.app.state.redis, id)
+    return chat_tasks.get(id, [])
+
+
+async def stop_task(request, task_id: str):
     """
     Cancel a running task and remove it from the global task list.
     """
+    if is_redis(request):
+        # PUBSUB: All instances check if they have this task, and stop if so.
+        await redis_send_command(
+            request.app.state.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.get(task_id)
     if not task:
         raise ValueError(f"Task with ID {task_id} not found.")

backend/open_webui/utils/audit.py

@@ -37,7 +37,7 @@ if TYPE_CHECKING:
 class AuditLogEntry:
     # `Metadata` audit level properties
     id: str
-    user: dict[str, Any]
+    user: Optional[dict[str, Any]]
     audit_level: str
     verb: str
     request_uri: str
@@ -190,21 +190,40 @@ class AuditLoggingMiddleware:
         finally:
             await self._log_audit_entry(request, context)
 
-    async def _get_authenticated_user(self, request: Request) -> UserModel:
-
+    async def _get_authenticated_user(self, request: Request) -> Optional[UserModel]:
         auth_header = request.headers.get("Authorization")
-        assert auth_header
-        user = get_current_user(request, None, get_http_authorization_cred(auth_header))
 
-        return user
+        try:
+            user = get_current_user(
+                request, None, get_http_authorization_cred(auth_header)
+            )
+            return user
+        except Exception as e:
+            logger.debug(f"Failed to get authenticated user: {str(e)}")
+
+        return None
 
     def _should_skip_auditing(self, request: Request) -> bool:
         if (
             request.method not in {"POST", "PUT", "PATCH", "DELETE"}
             or AUDIT_LOG_LEVEL == "NONE"
-            or not request.headers.get("authorization")
         ):
             return True
+
+        ALWAYS_LOG_ENDPOINTS = {
+            "/api/v1/auths/signin",
+            "/api/v1/auths/signout",
+            "/api/v1/auths/signup",
+        }
+        path = request.url.path.lower()
+        for endpoint in ALWAYS_LOG_ENDPOINTS:
+            if path.startswith(endpoint):
+                return False  # Do NOT skip logging for auth endpoints
+
+        # Skip logging if the request is not authenticated
+        if not request.headers.get("authorization"):
+            return True
+
         # match either /api/<resource>/...(for the endpoint /api/chat case) or /api/v1/<resource>/...
         pattern = re.compile(
             r"^/api(?:/v1)?/(" + "|".join(self.excluded_paths) + r")\b"
@@ -231,17 +250,32 @@ class AuditLoggingMiddleware:
         try:
             user = await self._get_authenticated_user(request)
 
+            user = (
+                user.model_dump(include={"id", "name", "email", "role"}) if user else {}
+            )
+
+            request_body = context.request_body.decode("utf-8", errors="replace")
+            response_body = context.response_body.decode("utf-8", errors="replace")
+
+            # Redact sensitive information
+            if "password" in request_body:
+                request_body = re.sub(
+                    r'"password":\s*"(.*?)"',
+                    '"password": "********"',
+                    request_body,
+                )
+
             entry = AuditLogEntry(
                 id=str(uuid.uuid4()),
-                user=user.model_dump(include={"id", "name", "email", "role"}),
+                user=user,
                 audit_level=self.audit_level.value,
                 verb=request.method,
                 request_uri=str(request.url),
                 response_status_code=context.metadata.get("response_status_code", None),
                 source_ip=request.client.host if request.client else None,
                 user_agent=request.headers.get("user-agent"),
-                request_object=context.request_body.decode("utf-8", errors="replace"),
-                response_object=context.response_body.decode("utf-8", errors="replace"),
+                request_object=request_body,
+                response_object=response_body,
             )
 
             self.audit_logger.write(entry)

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
@@ -155,6 +158,7 @@ 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 +198,17 @@ 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 +227,27 @@ 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)
+                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 +269,14 @@ 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

@@ -44,13 +44,15 @@ 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 = ""
-        self.session = aiohttp.ClientSession(base_url=self.base_url)
+        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/middleware.py

@@ -32,15 +32,22 @@ from open_webui.socket.main import (
 from open_webui.routers.tasks import (
     generate_queries,
     generate_title,
+    generate_follow_ups,
     generate_image_prompt,
     generate_chat_tags,
 )
 from open_webui.routers.retrieval import process_web_search, SearchForm
-from open_webui.routers.images import image_generations, GenerateImageForm
+from open_webui.routers.images import (
+    load_b64_image_data,
+    image_generations,
+    GenerateImageForm,
+    upload_image,
+)
 from open_webui.routers.pipelines import (
     process_pipeline_inlet_filter,
     process_pipeline_outlet_filter,
 )
+from open_webui.routers.memories import query_memory, QueryMemoryForm
 
 from open_webui.utils.webhook import post_webhook
 
@@ -235,46 +242,35 @@ async def chat_completion_tools_handler(
                 if isinstance(tool_result, str):
                     tool = tools[tool_function_name]
                     tool_id = tool.get("tool_id", "")
+
+                    tool_name = (
+                        f"{tool_id}/{tool_function_name}"
+                        if tool_id
+                        else f"{tool_function_name}"
+                    )
                     if tool.get("metadata", {}).get("citation", False) or tool.get(
                         "direct", False
                     ):
-
+                        # Citation is enabled for this tool
                         sources.append(
                             {
                                 "source": {
-                                    "name": (
-                                        f"TOOL:" + f"{tool_id}/{tool_function_name}"
-                                        if tool_id
-                                        else f"{tool_function_name}"
-                                    ),
+                                    "name": (f"TOOL:{tool_name}"),
                                 },
-                                "document": [tool_result, *tool_result_files],
+                                "document": [tool_result],
                                 "metadata": [
                                     {
-                                        "source": (
-                                            f"TOOL:" + f"{tool_id}/{tool_function_name}"
-                                            if tool_id
-                                            else f"{tool_function_name}"
-                                        )
+                                        "source": (f"TOOL:{tool_name}"),
+                                        "parameters": tool_function_params,
                                     }
                                 ],
                             }
                         )
                     else:
-                        sources.append(
-                            {
-                                "source": {},
-                                "document": [tool_result, *tool_result_files],
-                                "metadata": [
-                                    {
-                                        "source": (
-                                            f"TOOL:" + f"{tool_id}/{tool_function_name}"
-                                            if tool_id
-                                            else f"{tool_function_name}"
-                                        )
-                                    }
-                                ],
-                            }
+                        # Citation is not enabled for this tool
+                        body["messages"] = add_or_update_user_message(
+                            f"\nTool `{tool_name}` Output: {tool_result}",
+                            body["messages"],
                         )
 
                     if (
@@ -306,6 +302,45 @@ async def chat_completion_tools_handler(
     return body, {"sources": sources}
 
 
+async def chat_memory_handler(
+    request: Request, form_data: dict, extra_params: dict, user
+):
+    try:
+        results = await query_memory(
+            request,
+            QueryMemoryForm(
+                **{
+                    "content": get_last_user_message(form_data["messages"]) or "",
+                    "k": 3,
+                }
+            ),
+            user,
+        )
+    except Exception as e:
+        log.debug(e)
+        results = None
+
+    user_context = ""
+    if results and hasattr(results, "documents"):
+        if results.documents and len(results.documents) > 0:
+            for doc_idx, doc in enumerate(results.documents[0]):
+                created_at_date = "Unknown Date"
+
+                if results.metadatas[0][doc_idx].get("created_at"):
+                    created_at_timestamp = results.metadatas[0][doc_idx]["created_at"]
+                    created_at_date = time.strftime(
+                        "%Y-%m-%d", time.localtime(created_at_timestamp)
+                    )
+
+                user_context += f"{doc_idx + 1}. [{created_at_date}] {doc}\n"
+
+    form_data["messages"] = add_or_update_system_message(
+        f"User Context:\n{user_context}\n", form_data["messages"], append=True
+    )
+
+    return form_data
+
+
 async def chat_web_search_handler(
     request: Request, form_data: dict, extra_params: dict, user
 ):
@@ -356,6 +391,11 @@ async def chat_web_search_handler(
         log.exception(e)
         queries = [user_message]
 
+    # Check if generated queries are empty
+    if len(queries) == 1 and queries[0].strip() == "":
+        queries = [user_message]
+
+    # Check if queries are not found
     if len(queries) == 0:
         await event_emitter(
             {
@@ -369,115 +409,88 @@ async def chat_web_search_handler(
         )
         return form_data
 
-    all_results = []
+    await event_emitter(
+        {
+            "type": "status",
+            "data": {
+                "action": "web_search",
+                "description": "Searching the web",
+                "done": False,
+            },
+        }
+    )
 
-    for searchQuery in queries:
-        await event_emitter(
-            {
-                "type": "status",
-                "data": {
-                    "action": "web_search",
-                    "description": 'Searching "{{searchQuery}}"',
-                    "query": searchQuery,
-                    "done": False,
-                },
-            }
+    try:
+        results = await process_web_search(
+            request,
+            SearchForm(queries=queries),
+            user=user,
         )
 
-        try:
-            results = await process_web_search(
-                request,
-                SearchForm(
-                    **{
-                        "query": searchQuery,
+        if results:
+            files = form_data.get("files", [])
+
+            if results.get("collection_names"):
+                for col_idx, collection_name in enumerate(
+                    results.get("collection_names")
+                ):
+                    files.append(
+                        {
+                            "collection_name": collection_name,
+                            "name": ", ".join(queries),
+                            "type": "web_search",
+                            "urls": results["filenames"],
+                            "queries": queries,
+                        }
+                    )
+            elif results.get("docs"):
+                # Invoked when bypass embedding and retrieval is set to True
+                docs = results["docs"]
+                files.append(
+                    {
+                        "docs": docs,
+                        "name": ", ".join(queries),
+                        "type": "web_search",
+                        "urls": results["filenames"],
+                        "queries": queries,
                     }
-                ),
-                user=user,
-            )
+                )
 
-            if results:
-                all_results.append(results)
-                files = form_data.get("files", [])
+            form_data["files"] = files
 
-                if results.get("collection_names"):
-                    for col_idx, collection_name in enumerate(
-                        results.get("collection_names")
-                    ):
-                        files.append(
-                            {
-                                "collection_name": collection_name,
-                                "name": searchQuery,
-                                "type": "web_search",
-                                "urls": [results["filenames"][col_idx]],
-                            }
-                        )
-                elif results.get("docs"):
-                    # Invoked when bypass embedding and retrieval is set to True
-                    docs = results["docs"]
-
-                    if len(docs) == len(results["filenames"]):
-                        # the number of docs and filenames (urls) should be the same
-                        for doc_idx, doc in enumerate(docs):
-                            files.append(
-                                {
-                                    "docs": [doc],
-                                    "name": searchQuery,
-                                    "type": "web_search",
-                                    "urls": [results["filenames"][doc_idx]],
-                                }
-                            )
-                    else:
-                        # edge case when the number of docs and filenames (urls) are not the same
-                        # this should not happen, but if it does, we will just append the docs
-                        files.append(
-                            {
-                                "docs": results.get("docs", []),
-                                "name": searchQuery,
-                                "type": "web_search",
-                                "urls": results["filenames"],
-                            }
-                        )
-
-                form_data["files"] = files
-        except Exception as e:
-            log.exception(e)
             await event_emitter(
                 {
                     "type": "status",
                     "data": {
                         "action": "web_search",
-                        "description": 'Error searching "{{searchQuery}}"',
-                        "query": searchQuery,
+                        "description": "Searched {{count}} sites",
+                        "urls": results["filenames"],
+                        "done": True,
+                    },
+                }
+            )
+        else:
+            await event_emitter(
+                {
+                    "type": "status",
+                    "data": {
+                        "action": "web_search",
+                        "description": "No search results found",
                         "done": True,
                         "error": True,
                     },
                 }
             )
 
-    if all_results:
-        urls = []
-        for results in all_results:
-            if "filenames" in results:
-                urls.extend(results["filenames"])
-
+    except Exception as e:
+        log.exception(e)
         await event_emitter(
             {
                 "type": "status",
                 "data": {
                     "action": "web_search",
-                    "description": "Searched {{count}} sites",
-                    "urls": urls,
-                    "done": True,
-                },
-            }
-        )
-    else:
-        await event_emitter(
-            {
-                "type": "status",
-                "data": {
-                    "action": "web_search",
-                    "description": "No search results found",
+                    "description": "An error occurred while searching the web",
+                    "queries": queries,
                     "done": True,
                     "error": True,
                 },
@@ -550,13 +563,20 @@ async def chat_image_generation_handler(
             }
         )
 
-        for image in images:
-            await __event_emitter__(
-                {
-                    "type": "message",
-                    "data": {"content": f"![Generated Image]({image['url']})\n"},
-                }
-            )
+        await __event_emitter__(
+            {
+                "type": "files",
+                "data": {
+                    "files": [
+                        {
+                            "type": "image",
+                            "url": image["url"],
+                        }
+                        for image in images
+                    ]
+                },
+            }
+        )
 
         system_message_content = "<context>User is shown the generated image, tell the user that the image has been generated</context>"
     except Exception as e:
@@ -636,6 +656,7 @@ async def chat_completion_files_handler(
                         reranking_function=request.app.state.rf,
                         k_reranker=request.app.state.config.TOP_K_RERANKER,
                         r=request.app.state.config.RELEVANCE_THRESHOLD,
+                        hybrid_bm25_weight=request.app.state.config.HYBRID_BM25_WEIGHT,
                         hybrid_search=request.app.state.config.ENABLE_RAG_HYBRID_SEARCH,
                         full_context=request.app.state.config.RAG_FULL_CONTEXT,
                     ),
@@ -650,35 +671,40 @@ async def chat_completion_files_handler(
 
 def apply_params_to_form_data(form_data, model):
     params = form_data.pop("params", {})
+    custom_params = params.pop("custom_params", {})
+
+    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]
+
+    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 custom_params are provided, merge them into params
+        params = deep_update(params, custom_params)
+
     if model.get("ollama"):
+        # Ollama specific parameters
         form_data["options"] = params
-
-        if "format" in params:
-            form_data["format"] = params["format"]
-
-        if "keep_alive" in params:
-            form_data["keep_alive"] = params["keep_alive"]
     else:
-        if "seed" in params and params["seed"] is not None:
-            form_data["seed"] = params["seed"]
-
-        if "stop" in params and params["stop"] is not None:
-            form_data["stop"] = params["stop"]
-
-        if "temperature" in params and params["temperature"] is not None:
-            form_data["temperature"] = params["temperature"]
-
-        if "max_tokens" in params and params["max_tokens"] is not None:
-            form_data["max_tokens"] = params["max_tokens"]
-
-        if "top_p" in params and params["top_p"] is not None:
-            form_data["top_p"] = params["top_p"]
-
-        if "frequency_penalty" in params and params["frequency_penalty"] is not None:
-            form_data["frequency_penalty"] = params["frequency_penalty"]
-
-        if "reasoning_effort" in params and params["reasoning_effort"] is not None:
-            form_data["reasoning_effort"] = params["reasoning_effort"]
+        if isinstance(params, dict):
+            for key, value in params.items():
+                if value is not None:
+                    form_data[key] = value
 
         if "logit_bias" in params and params["logit_bias"] is not None:
             try:
@@ -686,13 +712,12 @@ def apply_params_to_form_data(form_data, model):
                     convert_logit_bias_input_to_json(params["logit_bias"])
                 )
             except Exception as e:
-                print(f"Error parsing logit_bias: {e}")
+                log.exception(f"Error parsing logit_bias: {e}")
 
     return form_data
 
 
 async def process_chat_payload(request, form_data, user, metadata, model):
-
     form_data = apply_params_to_form_data(form_data, model)
     log.debug(f"form_data: {form_data}")
 
@@ -702,12 +727,7 @@ async def process_chat_payload(request, form_data, user, metadata, model):
     extra_params = {
         "__event_emitter__": event_emitter,
         "__event_call__": event_call,
-        "__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,
@@ -784,9 +804,12 @@ async def process_chat_payload(request, form_data, user, metadata, model):
         raise e
 
     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", [])
+            )
         ]
 
         form_data, flags = await process_filter_functions(
@@ -801,6 +824,11 @@ async def process_chat_payload(request, form_data, user, metadata, model):
 
     features = form_data.pop("features", None)
     if features:
+        if "memory" in features and features["memory"]:
+            form_data = await chat_memory_handler(
+                request, form_data, extra_params, user
+            )
+
         if "web_search" in features and features["web_search"]:
             form_data = await chat_web_search_handler(
                 request, form_data, extra_params, user
@@ -897,11 +925,24 @@ async def process_chat_payload(request, form_data, user, metadata, model):
     # If context is not empty, insert it into the messages
     if len(sources) > 0:
         context_string = ""
-        for source_idx, source in enumerate(sources):
+        citation_idx = {}
+        for source in sources:
             if "document" in source:
-                for doc_idx, doc_context in enumerate(source["document"]):
+                for doc_context, doc_meta in zip(
+                    source["document"], source["metadata"]
+                ):
+                    source_name = source.get("source", {}).get("name", None)
+                    citation_id = (
+                        doc_meta.get("source", None)
+                        or source.get("source", {}).get("id", None)
+                        or "N/A"
+                    )
+                    if citation_id not in citation_idx:
+                        citation_idx[citation_id] = len(citation_idx) + 1
                     context_string += (
-                        f'<source id="{source_idx + 1}">{doc_context}</source>\n'
+                        f'<source id="{citation_idx[citation_id]}"'
+                        + (f' name="{source_name}"' if source_name else "")
+                        + f">{doc_context}</source>\n"
                     )
 
         context_string = context_string.strip()
@@ -935,7 +976,12 @@ async def process_chat_payload(request, form_data, user, metadata, model):
             )
 
     # If there are citations, add them to the data_items
-    sources = [source for source in sources if source.get("source", {}).get("name", "")]
+    sources = [
+        source
+        for source in sources
+        if source.get("source", {}).get("name", "")
+        or source.get("source", {}).get("id", "")
+    ]
 
     if len(sources) > 0:
         events.append({"sources": sources})
@@ -964,9 +1010,93 @@ async def process_chat_response(
         message = message_map.get(metadata["message_id"]) if message_map else None
 
         if message:
-            messages = get_message_list(message_map, message.get("id"))
+            message_list = get_message_list(message_map, metadata["message_id"])
+
+            # Remove details tags and files from the messages.
+            # as get_message_list creates a new list, it does not affect
+            # the original messages outside of this handler
+
+            messages = []
+            for message in message_list:
+                content = message.get("content", "")
+                if isinstance(content, list):
+                    for item in content:
+                        if item.get("type") == "text":
+                            content = item["text"]
+                            break
+
+                if isinstance(content, str):
+                    content = re.sub(
+                        r"<details\b[^>]*>.*?<\/details>|!\[.*?\]\(.*?\)",
+                        "",
+                        content,
+                        flags=re.S | re.I,
+                    ).strip()
+
+                messages.append(
+                    {
+                        **message,
+                        "role": message.get(
+                            "role", "assistant"
+                        ),  # Safe fallback for missing role
+                        "content": content,
+                    }
+                )
 
             if tasks and messages:
+                if (
+                    TASKS.FOLLOW_UP_GENERATION in tasks
+                    and tasks[TASKS.FOLLOW_UP_GENERATION]
+                ):
+                    res = await generate_follow_ups(
+                        request,
+                        {
+                            "model": message["model"],
+                            "messages": messages,
+                            "message_id": metadata["message_id"],
+                            "chat_id": metadata["chat_id"],
+                        },
+                        user,
+                    )
+
+                    if res and isinstance(res, dict):
+                        if len(res.get("choices", [])) == 1:
+                            follow_ups_string = (
+                                res.get("choices", [])[0]
+                                .get("message", {})
+                                .get("content", "")
+                            )
+                        else:
+                            follow_ups_string = ""
+
+                        follow_ups_string = follow_ups_string[
+                            follow_ups_string.find("{") : follow_ups_string.rfind("}")
+                            + 1
+                        ]
+
+                        try:
+                            follow_ups = json.loads(follow_ups_string).get(
+                                "follow_ups", []
+                            )
+                            Chats.upsert_message_to_chat_by_id_and_message_id(
+                                metadata["chat_id"],
+                                metadata["message_id"],
+                                {
+                                    "followUps": follow_ups,
+                                },
+                            )
+
+                            await event_emitter(
+                                {
+                                    "type": "chat:message:follow_ups",
+                                    "data": {
+                                        "follow_ups": follow_ups,
+                                    },
+                                }
+                            )
+                        except Exception as e:
+                            pass
+
                 if TASKS.TITLE_GENERATION in tasks:
                     if tasks[TASKS.TITLE_GENERATION]:
                         res = await generate_title(
@@ -1129,12 +1259,13 @@ async def process_chat_response(
                         metadata["chat_id"],
                         metadata["message_id"],
                         {
+                            "role": "assistant",
                             "content": content,
                         },
                     )
 
                     # Send a webhook notification if the user is not active
-                    if get_active_status_by_user_id(user.id) is None:
+                    if not get_active_status_by_user_id(user.id):
                         webhook_url = Users.get_user_webhook_url_by_id(user.id)
                         if webhook_url:
                             post_webhook(
@@ -1151,8 +1282,34 @@ async def process_chat_response(
 
                     await background_tasks_handler()
 
+            if events and isinstance(events, list) and isinstance(response, dict):
+                extra_response = {}
+                for event in events:
+                    if isinstance(event, dict):
+                        extra_response.update(event)
+                    else:
+                        extra_response[event] = True
+
+                response = {
+                    **extra_response,
+                    **response,
+                }
+
             return response
         else:
+            if events and isinstance(events, list) and isinstance(response, dict):
+                extra_response = {}
+                for event in events:
+                    if isinstance(event, dict):
+                        extra_response.update(event)
+                    else:
+                        extra_response[event] = True
+
+                response = {
+                    **extra_response,
+                    **response,
+                }
+
             return response
 
     # Non standard response
@@ -1165,19 +1322,16 @@ async def process_chat_response(
     extra_params = {
         "__event_emitter__": event_emitter,
         "__event_call__": event_caller,
-        "__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,
     }
     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", [])
+        )
     ]
 
     # Streaming response
@@ -1422,6 +1576,9 @@ async def process_chat_response(
 
                             if after_tag:
                                 content_blocks[-1]["content"] = after_tag
+                                tag_content_handler(
+                                    content_type, tags, after_tag, content_blocks
+                                )
 
                             break
                 elif content_blocks[-1]["type"] == content_type:
@@ -1609,6 +1766,9 @@ async def process_chat_response(
                             )
 
                             if data:
+                                if "event" in data:
+                                    await event_emitter(data.get("event", {}))
+
                                 if "selected_model_id" in data:
                                     model_id = data["selected_model_id"]
                                     Chats.upsert_message_to_chat_by_id_and_message_id(
@@ -1653,14 +1813,36 @@ async def process_chat_response(
                                             )
 
                                             if tool_call_index is not None:
-                                                if (
-                                                    len(response_tool_calls)
-                                                    <= tool_call_index
-                                                ):
+                                                # Check if the tool call already exists
+                                                current_response_tool_call = None
+                                                for (
+                                                    response_tool_call
+                                                ) in response_tool_calls:
+                                                    if (
+                                                        response_tool_call.get("index")
+                                                        == tool_call_index
+                                                    ):
+                                                        current_response_tool_call = (
+                                                            response_tool_call
+                                                        )
+                                                        break
+
+                                                if current_response_tool_call is None:
+                                                    # Add the new tool call
+                                                    delta_tool_call.setdefault(
+                                                        "function", {}
+                                                    )
+                                                    delta_tool_call[
+                                                        "function"
+                                                    ].setdefault("name", "")
+                                                    delta_tool_call[
+                                                        "function"
+                                                    ].setdefault("arguments", "")
                                                     response_tool_calls.append(
                                                         delta_tool_call
                                                     )
                                                 else:
+                                                    # Update the existing tool call
                                                     delta_name = delta_tool_call.get(
                                                         "function", {}
                                                     ).get("name")
@@ -1671,24 +1853,24 @@ async def process_chat_response(
                                                     )
 
                                                     if delta_name:
-                                                        response_tool_calls[
-                                                            tool_call_index
-                                                        ]["function"][
-                                                            "name"
-                                                        ] += delta_name
+                                                        current_response_tool_call[
+                                                            "function"
+                                                        ]["name"] += delta_name
 
                                                     if delta_arguments:
-                                                        response_tool_calls[
-                                                            tool_call_index
-                                                        ]["function"][
+                                                        current_response_tool_call[
+                                                            "function"
+                                                        ][
                                                             "arguments"
                                                         ] += delta_arguments
 
                                     value = delta.get("content")
 
-                                    reasoning_content = delta.get(
-                                        "reasoning_content"
-                                    ) or delta.get("reasoning")
+                                    reasoning_content = (
+                                        delta.get("reasoning_content")
+                                        or delta.get("reasoning")
+                                        or delta.get("thinking")
+                                    )
                                     if reasoning_content:
                                         if (
                                             not content_blocks
@@ -2079,28 +2261,21 @@ async def process_chat_response(
                                         stdoutLines = stdout.split("\n")
                                         for idx, line in enumerate(stdoutLines):
                                             if "data:image/png;base64" in line:
-                                                id = str(uuid4())
-
-                                                # ensure the path exists
-                                                os.makedirs(
-                                                    os.path.join(CACHE_DIR, "images"),
-                                                    exist_ok=True,
+                                                image_url = ""
+                                                # Extract base64 image data from the line
+                                                image_data, content_type = (
+                                                    load_b64_image_data(line)
                                                 )
-
-                                                image_path = os.path.join(
-                                                    CACHE_DIR,
-                                                    f"images/{id}.png",
-                                                )
-
-                                                with open(image_path, "wb") as f:
-                                                    f.write(
-                                                        base64.b64decode(
-                                                            line.split(",")[1]
-                                                        )
+                                                if image_data is not None:
+                                                    image_url = upload_image(
+                                                        request,
+                                                        image_data,
+                                                        content_type,
+                                                        metadata,
+                                                        user,
                                                     )
-
                                                 stdoutLines[idx] = (
-                                                    f"![Output Image {idx}](/cache/images/{id}.png)"
+                                                    f"![Output Image]({image_url})"
                                                 )
 
                                         output["stdout"] = "\n".join(stdoutLines)
@@ -2111,30 +2286,22 @@ async def process_chat_response(
                                         resultLines = result.split("\n")
                                         for idx, line in enumerate(resultLines):
                                             if "data:image/png;base64" in line:
-                                                id = str(uuid4())
-
-                                                # ensure the path exists
-                                                os.makedirs(
-                                                    os.path.join(CACHE_DIR, "images"),
-                                                    exist_ok=True,
+                                                image_url = ""
+                                                # Extract base64 image data from the line
+                                                image_data, content_type = (
+                                                    load_b64_image_data(line)
                                                 )
-
-                                                image_path = os.path.join(
-                                                    CACHE_DIR,
-                                                    f"images/{id}.png",
-                                                )
-
-                                                with open(image_path, "wb") as f:
-                                                    f.write(
-                                                        base64.b64decode(
-                                                            line.split(",")[1]
-                                                        )
+                                                if image_data is not None:
+                                                    image_url = upload_image(
+                                                        request,
+                                                        image_data,
+                                                        content_type,
+                                                        metadata,
+                                                        user,
                                                     )
-
                                                 resultLines[idx] = (
-                                                    f"![Output Image {idx}](/cache/images/{id}.png)"
+                                                    f"![Output Image]({image_url})"
                                                 )
-
                                         output["result"] = "\n".join(resultLines)
                         except Exception as e:
                             output = str(e)
@@ -2202,7 +2369,7 @@ async def process_chat_response(
                     )
 
                 # Send a webhook notification if the user is not active
-                if get_active_status_by_user_id(user.id) is None:
+                if not get_active_status_by_user_id(user.id):
                     webhook_url = Users.get_user_webhook_url_by_id(user.id)
                     if webhook_url:
                         post_webhook(
@@ -2243,7 +2410,9 @@ async def process_chat_response(
                 await response.background()
 
         # background_tasks.add_task(post_response_handler, response, events)
-        task_id, _ = create_task(post_response_handler(response, events))
+        task_id, _ = await create_task(
+            request, post_response_handler(response, events), id=metadata["chat_id"]
+        )
         return {"status": True, "task_id": task_id}
 
     else:

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
 
 
@@ -30,34 +34,46 @@ 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):
@@ -110,25 +126,43 @@ 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:
             for model in models:
-                if (
-                    custom_model.id == model["id"]
-                    or custom_model.id == model["id"].split(":")[0]
+                if custom_model.id == model["id"] or (
+                    model.get("owned_by") == "ollama"
+                    and custom_model.id
+                    == model["id"].split(":")[
+                        0
+                    ]  # Ollama may return model ids in different formats (e.g., 'llama3' vs. 'llama3:7b')
                 ):
                     if custom_model.is_active:
                         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)
 
@@ -137,7 +171,9 @@ async def get_all_models(request, user: UserModel = None):
         ):
             owned_by = "openai"
             pipe = None
+
             action_ids = []
+            filter_ids = []
 
             for model in models:
                 if (
@@ -151,9 +187,13 @@ 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}",
@@ -165,6 +205,7 @@ 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,
                 }
             )
 
@@ -178,8 +219,11 @@ 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
@@ -190,16 +234,28 @@ 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:
         action_ids = [
@@ -207,6 +263,11 @@ 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:
@@ -218,6 +279,20 @@ 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

@@ -3,6 +3,7 @@ import logging
 import mimetypes
 import sys
 import uuid
+import json
 
 import aiohttp
 from authlib.integrations.starlette_client import OAuth
@@ -15,7 +16,7 @@ from starlette.responses import RedirectResponse
 
 from open_webui.models.auths import Auths
 from open_webui.models.users import Users
-from open_webui.models.groups import Groups, GroupModel, GroupUpdateForm
+from open_webui.models.groups import Groups, GroupModel, GroupUpdateForm, GroupForm
 from open_webui.config import (
     DEFAULT_USER_ROLE,
     ENABLE_OAUTH_SIGNUP,
@@ -23,6 +24,8 @@ from open_webui.config import (
     OAUTH_PROVIDERS,
     ENABLE_OAUTH_ROLE_MANAGEMENT,
     ENABLE_OAUTH_GROUP_MANAGEMENT,
+    ENABLE_OAUTH_GROUP_CREATION,
+    OAUTH_BLOCKED_GROUPS,
     OAUTH_ROLES_CLAIM,
     OAUTH_GROUPS_CLAIM,
     OAUTH_EMAIL_CLAIM,
@@ -31,12 +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,
@@ -57,6 +62,8 @@ auth_manager_config.ENABLE_OAUTH_SIGNUP = ENABLE_OAUTH_SIGNUP
 auth_manager_config.OAUTH_MERGE_ACCOUNTS_BY_EMAIL = OAUTH_MERGE_ACCOUNTS_BY_EMAIL
 auth_manager_config.ENABLE_OAUTH_ROLE_MANAGEMENT = ENABLE_OAUTH_ROLE_MANAGEMENT
 auth_manager_config.ENABLE_OAUTH_GROUP_MANAGEMENT = ENABLE_OAUTH_GROUP_MANAGEMENT
+auth_manager_config.ENABLE_OAUTH_GROUP_CREATION = ENABLE_OAUTH_GROUP_CREATION
+auth_manager_config.OAUTH_BLOCKED_GROUPS = OAUTH_BLOCKED_GROUPS
 auth_manager_config.OAUTH_ROLES_CLAIM = OAUTH_ROLES_CLAIM
 auth_manager_config.OAUTH_GROUPS_CLAIM = OAUTH_GROUPS_CLAIM
 auth_manager_config.OAUTH_EMAIL_CLAIM = OAUTH_EMAIL_CLAIM
@@ -67,6 +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:
@@ -140,6 +148,12 @@ class OAuthManager:
         log.debug("Running OAUTH Group management")
         oauth_claim = auth_manager_config.OAUTH_GROUPS_CLAIM
 
+        try:
+            blocked_groups = json.loads(auth_manager_config.OAUTH_BLOCKED_GROUPS)
+        except Exception as e:
+            log.exception(f"Error loading OAUTH_BLOCKED_GROUPS: {e}")
+            blocked_groups = []
+
         user_oauth_groups = []
         # Nested claim search for groups claim
         if oauth_claim:
@@ -147,11 +161,62 @@ class OAuthManager:
             nested_claims = oauth_claim.split(".")
             for nested_claim in nested_claims:
                 claim_data = claim_data.get(nested_claim, {})
-            user_oauth_groups = claim_data if isinstance(claim_data, list) else []
+
+            if isinstance(claim_data, list):
+                user_oauth_groups = claim_data
+            elif isinstance(claim_data, str):
+                user_oauth_groups = [claim_data]
+            else:
+                user_oauth_groups = []
 
         user_current_groups: list[GroupModel] = Groups.get_groups_by_member_id(user.id)
         all_available_groups: list[GroupModel] = Groups.get_groups()
 
+        # Create groups if they don't exist and creation is enabled
+        if auth_manager_config.ENABLE_OAUTH_GROUP_CREATION:
+            log.debug("Checking for missing groups to create...")
+            all_group_names = {g.name for g in all_available_groups}
+            groups_created = False
+            # Determine creator ID: Prefer admin, fallback to current user if no admin exists
+            admin_user = Users.get_super_admin_user()
+            creator_id = admin_user.id if admin_user else user.id
+            log.debug(f"Using creator ID {creator_id} for potential group creation.")
+
+            for group_name in user_oauth_groups:
+                if group_name not in all_group_names:
+                    log.info(
+                        f"Group '{group_name}' not found via OAuth claim. Creating group..."
+                    )
+                    try:
+                        new_group_form = GroupForm(
+                            name=group_name,
+                            description=f"Group '{group_name}' created automatically via OAuth.",
+                            permissions=default_permissions,  # Use default permissions from function args
+                            user_ids=[],  # Start with no users, user will be added later by subsequent logic
+                        )
+                        # Use determined creator ID (admin or fallback to current user)
+                        created_group = Groups.insert_new_group(
+                            creator_id, new_group_form
+                        )
+                        if created_group:
+                            log.info(
+                                f"Successfully created group '{group_name}' with ID {created_group.id} using creator ID {creator_id}"
+                            )
+                            groups_created = True
+                            # Add to local set to prevent duplicate creation attempts in this run
+                            all_group_names.add(group_name)
+                        else:
+                            log.error(
+                                f"Failed to create group '{group_name}' via OAuth."
+                            )
+                    except Exception as e:
+                        log.error(f"Error creating group '{group_name}' via OAuth: {e}")
+
+            # Refresh the list of all available groups if any were created
+            if groups_created:
+                all_available_groups = Groups.get_groups()
+                log.debug("Refreshed list of all available groups after creation.")
+
         log.debug(f"Oauth Groups claim: {oauth_claim}")
         log.debug(f"User oauth groups: {user_oauth_groups}")
         log.debug(f"User's current groups: {[g.name for g in user_current_groups]}")
@@ -161,7 +226,11 @@ class OAuthManager:
 
         # Remove groups that user is no longer a part of
         for group_model in user_current_groups:
-            if user_oauth_groups and group_model.name not in user_oauth_groups:
+            if (
+                user_oauth_groups
+                and group_model.name not in user_oauth_groups
+                and group_model.name not in blocked_groups
+            ):
                 # Remove group from user
                 log.debug(
                     f"Removing user from group {group_model.name} as it is no longer in their oauth groups"
@@ -191,6 +260,7 @@ class OAuthManager:
                 user_oauth_groups
                 and group_model.name in user_oauth_groups
                 and not any(gm.name == group_model.name for gm in user_current_groups)
+                and group_model.name not in blocked_groups
             ):
                 # Add user to group
                 log.debug(
@@ -215,6 +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)
@@ -257,9 +372,11 @@ class OAuthManager:
                 try:
                     access_token = token.get("access_token")
                     headers = {"Authorization": f"Bearer {access_token}"}
-                    async with aiohttp.ClientSession() as session:
+                    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()
@@ -315,6 +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()
 
@@ -330,40 +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() 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"
 
@@ -434,5 +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 = request.app.state.config.WEBUI_URL or request.base_url
+        if isinstance(redirect_base_url, str) and 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

@@ -68,7 +68,7 @@ def replace_imports(content):
     return content
 
 
-def load_tools_module_by_id(tool_id, content=None):
+def load_tool_module_by_id(tool_id, content=None):
 
     if content is None:
         tool = Tools.get_tool_by_id(tool_id)
@@ -115,7 +115,7 @@ def load_tools_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:
@@ -157,7 +157,8 @@ def load_function_module_by_id(function_id, content=None):
             raise Exception("No Function class found in the module")
     except Exception as e:
         log.error(f"Error loading module: {function_id}: {e}")
-        del sys.modules[module_name]  # Cleanup by removing the module in case of error
+        # Cleanup by removing the module in case of error
+        del sys.modules[module_name]
 
         Functions.update_function_by_id(function_id, {"is_active": False})
         raise e
@@ -165,6 +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:
@@ -182,3 +239,32 @@ def install_frontmatter_requirements(requirements: str):
 
     else:
         log.info("No requirements found in frontmatter.")
+
+
+def install_tool_and_function_dependencies():
+    """
+    Install all dependencies for all admin tools and active functions.
+
+    By first collecting all dependencies from the frontmatter of each tool and function,
+    and then installing them using pip. Duplicates or similar version specifications are
+    handled by pip as much as possible.
+    """
+    function_list = Functions.get_functions(active_only=True)
+    tool_list = Tools.get_tools()
+
+    all_dependencies = ""
+    try:
+        for function in function_list:
+            frontmatter = extract_frontmatter(replace_imports(function.content))
+            if dependencies := frontmatter.get("requirements"):
+                all_dependencies += f"{dependencies}, "
+        for tool in tool_list:
+            # Only install requirements for admin tools
+            if tool.user.role == "admin":
+                frontmatter = extract_frontmatter(replace_imports(tool.content))
+                if dependencies := frontmatter.get("requirements"):
+                    all_dependencies += f"{dependencies}, "
+
+        install_frontmatter_requirements(all_dependencies.strip(", "))
+    except Exception as e:
+        log.error(f"Error installing requirements: {e}")

backend/open_webui/utils/redis.py

@@ -1,10 +1,9 @@
 import socketio
-import redis
-from redis import asyncio as aioredis
 from urllib.parse import urlparse
+from typing import Optional
 
 
-def parse_redis_sentinel_url(redis_url):
+def parse_redis_service_url(redis_url):
     parsed_url = urlparse(redis_url)
     if parsed_url.scheme != "redis":
         raise ValueError("Invalid Redis URL scheme. Must be 'redis'.")
@@ -18,23 +17,46 @@ def parse_redis_sentinel_url(redis_url):
     }
 
 
-def get_redis_connection(redis_url, redis_sentinels, decode_responses=True):
-    if redis_sentinels:
-        redis_config = parse_redis_sentinel_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):
@@ -45,65 +67,14 @@ def get_sentinels_from_env(sentinel_hosts_env, sentinel_port_env):
     return []
 
 
-class AsyncRedisSentinelManager(socketio.AsyncRedisManager):
-    def __init__(
-        self,
-        sentinel_hosts,
-        sentinel_port=26379,
-        redis_port=6379,
-        service="mymaster",
-        db=0,
-        username=None,
-        password=None,
-        channel="socketio",
-        write_only=False,
-        logger=None,
-        redis_options=None,
-    ):
-        """
-        Initialize the Redis Sentinel Manager.
-        This implementation mostly replicates the __init__ of AsyncRedisManager and
-        overrides _redis_connect() with a version that uses Redis Sentinel
-
-        :param sentinel_hosts: List of Sentinel hosts
-        :param sentinel_port: Sentinel Port
-        :param redis_port: Redis Port (currently unsupported by aioredis!)
-        :param service: Master service name in Sentinel
-        :param db: Redis database to use
-        :param username: Redis username (if any) (currently unsupported by aioredis!)
-        :param password: Redis password (if any)
-        :param channel: The channel name on which the server sends and receives
-                        notifications. Must be the same in all the servers.
-        :param write_only: If set to ``True``, only initialize to emit events. The
-                           default of ``False`` initializes the class for emitting
-                           and receiving.
-        :param redis_options: additional keyword arguments to be passed to
-                              ``aioredis.from_url()``.
-        """
-        self._sentinels = [(host, sentinel_port) for host in sentinel_hosts]
-        self._redis_port = redis_port
-        self._service = service
-        self._db = db
-        self._username = username
-        self._password = password
-        self._channel = channel
-        self.redis_options = redis_options or {}
-
-        # connect and call grandparent constructor
-        self._redis_connect()
-        super(socketio.AsyncRedisManager, self).__init__(
-            channel=channel, write_only=write_only, logger=logger
-        )
-
-    def _redis_connect(self):
-        """Establish connections to Redis through Sentinel."""
-        sentinel = aioredis.sentinel.Sentinel(
-            self._sentinels,
-            port=self._redis_port,
-            db=self._db,
-            password=self._password,
-            **self.redis_options,
-        )
-
-        self.redis = sentinel.master_for(self._service)
-        self.pubsub = self.redis.pubsub(ignore_subscribe_messages=True)
+def get_sentinel_url_from_env(redis_url, sentinel_hosts_env, sentinel_port_env):
+    redis_config = parse_redis_service_url(redis_url)
+    username = redis_config["username"] or ""
+    password = redis_config["password"] or ""
+    auth_part = ""
+    if username or password:
+        auth_part = f"{username}:{password}@"
+    hosts_part = ",".join(
+        f"{host}:{sentinel_port_env}" for host in sentinel_hosts_env.split(",")
+    )
+    return f"redis+sentinel://{auth_part}{hosts_part}/{redis_config['db']}/{redis_config['service']}"

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:
@@ -152,6 +152,8 @@ def rag_template(template: str, context: str, query: str):
     if template.strip() == "":
         template = DEFAULT_RAG_TEMPLATE
 
+    template = prompt_template(template)
+
     if "[context]" not in template and "{{CONTEXT}}" not in template:
         log.debug(
             "WARNING: The RAG template does not contain the '[context]' or '{{CONTEXT}}' placeholder."
@@ -205,6 +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/tools.py

@@ -4,23 +4,48 @@ import re
 import inspect
 import aiohttp
 import asyncio
+import yaml
 
-from typing import Any, Awaitable, Callable, get_type_hints, Dict, List, Union, Optional
+from pydantic import BaseModel
+from pydantic.fields import FieldInfo
+from typing import (
+    Any,
+    Awaitable,
+    Callable,
+    get_type_hints,
+    get_args,
+    get_origin,
+    Dict,
+    List,
+    Tuple,
+    Union,
+    Optional,
+    Type,
+)
 from functools import update_wrapper, partial
 
 
 from fastapi import Request
 from pydantic import BaseModel, Field, create_model
-from langchain_core.utils.function_calling import convert_to_openai_function
+
+from langchain_core.utils.function_calling import (
+    convert_to_openai_function as convert_pydantic_model_to_openai_function_spec,
+)
 
 
 from open_webui.models.tools import Tools
 from open_webui.models.users import UserModel
-from open_webui.utils.plugin import load_tools_module_by_id
+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,
+)
 
 import copy
 
 log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["MODELS"])
 
 
 def get_async_tool_function_and_apply_extra_params(
@@ -55,7 +80,12 @@ def get_tools(
                 tool_server_connection = (
                     request.app.state.config.TOOL_SERVER_CONNECTIONS[server_idx]
                 )
-                tool_server_data = request.app.state.TOOL_SERVERS[server_idx]
+                tool_server_data = None
+                for server in request.app.state.TOOL_SERVERS:
+                    if server["idx"] == server_idx:
+                        tool_server_data = server
+                        break
+                assert tool_server_data is not None
                 specs = tool_server_data.get("specs", [])
 
                 for spec in specs:
@@ -112,7 +142,7 @@ def get_tools(
         else:
             module = request.app.state.TOOLS.get(tool_id, None)
             if module is None:
-                module, _ = load_tools_module_by_id(tool_id)
+                module, _ = load_tool_module_by_id(tool_id)
                 request.app.state.TOOLS[tool_id] = module
 
             extra_params["__id__"] = tool_id
@@ -130,7 +160,7 @@ 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__)
@@ -233,7 +263,7 @@ def parse_docstring(docstring):
     return param_descriptions
 
 
-def function_to_pydantic_model(func: Callable) -> type[BaseModel]:
+def convert_function_to_pydantic_model(func: Callable) -> type[BaseModel]:
     """
     Converts a Python function's type hints and docstring to a Pydantic model,
     including support for nested types, default values, and descriptions.
@@ -250,45 +280,59 @@ def function_to_pydantic_model(func: Callable) -> type[BaseModel]:
     parameters = signature.parameters
 
     docstring = func.__doc__
-    descriptions = parse_docstring(docstring)
 
-    tool_description = parse_description(docstring)
+    function_description = parse_description(docstring)
+    function_param_descriptions = parse_docstring(docstring)
 
     field_defs = {}
     for name, param in parameters.items():
+
         type_hint = type_hints.get(name, Any)
         default_value = param.default if param.default is not param.empty else ...
-        description = descriptions.get(name, None)
-        if not description:
+
+        param_description = function_param_descriptions.get(name, None)
+
+        if param_description:
+            field_defs[name] = type_hint, Field(
+                default_value, description=param_description
+            )
+        else:
             field_defs[name] = type_hint, default_value
-            continue
-        field_defs[name] = type_hint, Field(default_value, description=description)
 
     model = create_model(func.__name__, **field_defs)
-    model.__doc__ = tool_description
+    model.__doc__ = function_description
 
     return model
 
 
-def get_callable_attributes(tool: object) -> list[Callable]:
+def get_functions_from_tool(tool: object) -> list[Callable]:
     return [
         getattr(tool, func)
         for func in dir(tool)
-        if callable(getattr(tool, func))
-        and not func.startswith("__")
-        and not inspect.isclass(getattr(tool, func))
+        if callable(
+            getattr(tool, func)
+        )  # checks if the attribute is callable (a method or function).
+        and not func.startswith(
+            "__"
+        )  # filters out special (dunder) methods like init, str, etc. — these are usually built-in functions of an object that you might not need to use directly.
+        and not inspect.isclass(
+            getattr(tool, func)
+        )  # ensures that the callable is not a class itself, just a method or function.
     ]
 
 
-def get_tools_specs(tool_class: object) -> list[dict]:
-    function_model_list = map(
-        function_to_pydantic_model, get_callable_attributes(tool_class)
+def get_tool_specs(tool_module: object) -> list[dict]:
+    function_models = map(
+        convert_function_to_pydantic_model, get_functions_from_tool(tool_module)
     )
-    return [
-        convert_to_openai_function(function_model)
-        for function_model in function_model_list
+
+    specs = [
+        convert_pydantic_model_to_openai_function_spec(function_model)
+        for function_model in function_models
     ]
 
+    return specs
+
 
 def resolve_schema(schema, components):
     """
@@ -334,51 +378,64 @@ def convert_openapi_to_tool_payload(openapi_spec):
 
     for path, methods in openapi_spec.get("paths", {}).items():
         for method, operation in methods.items():
-            tool = {
-                "type": "function",
-                "name": operation.get("operationId"),
-                "description": operation.get(
-                    "description", operation.get("summary", "No description available.")
-                ),
-                "parameters": {"type": "object", "properties": {}, "required": []},
-            }
-
-            # Extract path and query parameters
-            for param in operation.get("parameters", []):
-                param_name = param["name"]
-                param_schema = param.get("schema", {})
-                tool["parameters"]["properties"][param_name] = {
-                    "type": param_schema.get("type"),
-                    "description": param_schema.get("description", ""),
+            if operation.get("operationId"):
+                tool = {
+                    "type": "function",
+                    "name": operation.get("operationId"),
+                    "description": operation.get(
+                        "description",
+                        operation.get("summary", "No description available."),
+                    ),
+                    "parameters": {"type": "object", "properties": {}, "required": []},
                 }
-                if param.get("required"):
-                    tool["parameters"]["required"].append(param_name)
 
-            # Extract and resolve requestBody if available
-            request_body = operation.get("requestBody")
-            if request_body:
-                content = request_body.get("content", {})
-                json_schema = content.get("application/json", {}).get("schema")
-                if json_schema:
-                    resolved_schema = resolve_schema(
-                        json_schema, openapi_spec.get("components", {})
-                    )
-
-                    if resolved_schema.get("properties"):
-                        tool["parameters"]["properties"].update(
-                            resolved_schema["properties"]
+                # Extract path and query parameters
+                for param in operation.get("parameters", []):
+                    param_name = param["name"]
+                    param_schema = param.get("schema", {})
+                    description = param_schema.get("description", "")
+                    if not description:
+                        description = param.get("description") or ""
+                    if param_schema.get("enum") and isinstance(
+                        param_schema.get("enum"), list
+                    ):
+                        description += (
+                            f". Possible values: {', '.join(param_schema.get('enum'))}"
                         )
-                        if "required" in resolved_schema:
-                            tool["parameters"]["required"] = list(
-                                set(
-                                    tool["parameters"]["required"]
-                                    + resolved_schema["required"]
-                                )
-                            )
-                    elif resolved_schema.get("type") == "array":
-                        tool["parameters"] = resolved_schema  # special case for array
+                    tool["parameters"]["properties"][param_name] = {
+                        "type": param_schema.get("type"),
+                        "description": description,
+                    }
+                    if param.get("required"):
+                        tool["parameters"]["required"].append(param_name)
 
-            tool_payload.append(tool)
+                # Extract and resolve requestBody if available
+                request_body = operation.get("requestBody")
+                if request_body:
+                    content = request_body.get("content", {})
+                    json_schema = content.get("application/json", {}).get("schema")
+                    if json_schema:
+                        resolved_schema = resolve_schema(
+                            json_schema, openapi_spec.get("components", {})
+                        )
+
+                        if resolved_schema.get("properties"):
+                            tool["parameters"]["properties"].update(
+                                resolved_schema["properties"]
+                            )
+                            if "required" in resolved_schema:
+                                tool["parameters"]["required"] = list(
+                                    set(
+                                        tool["parameters"]["required"]
+                                        + resolved_schema["required"]
+                                    )
+                                )
+                        elif resolved_schema.get("type") == "array":
+                            tool["parameters"] = (
+                                resolved_schema  # special case for array
+                            )
+
+                tool_payload.append(tool)
 
     return tool_payload
 
@@ -393,14 +450,23 @@ async def get_tool_server_data(token: str, url: str) -> Dict[str, Any]:
 
     error = None
     try:
-        async with aiohttp.ClientSession() as session:
-            async with session.get(url, headers=headers) as response:
+        timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA)
+        async with aiohttp.ClientSession(timeout=timeout, trust_env=True) as session:
+            async with session.get(
+                url, headers=headers, ssl=AIOHTTP_CLIENT_SESSION_TOOL_SERVER_SSL
+            ) as response:
                 if response.status != 200:
                     error_body = await response.json()
                     raise Exception(error_body)
-                res = await response.json()
+
+                # Check if URL ends with .yaml or .yml to determine format
+                if url.lower().endswith((".yaml", ".yml")):
+                    text_content = await response.text()
+                    res = yaml.safe_load(text_content)
+                else:
+                    res = await response.json()
     except Exception as err:
-        print("Error:", err)
+        log.exception(f"Could not fetch tool server spec from {url}")
         if isinstance(err, dict) and "detail" in err:
             error = err["detail"]
         else:
@@ -413,7 +479,7 @@ 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("Fetched data:", data)
     return data
 
 
@@ -424,8 +490,19 @@ 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
@@ -434,26 +511,37 @@ 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"),
             }
@@ -529,19 +617,26 @@ async def execute_tool_server(
         if token:
             headers["Authorization"] = f"Bearer {token}"
 
-        async with aiohttp.ClientSession() as session:
+        async with aiohttp.ClientSession(trust_env=True) as session:
             request_method = getattr(session, http_method.lower())
 
             if http_method in ["post", "put", "patch"]:
                 async with request_method(
-                    final_url, json=body_params, headers=headers
+                    final_url,
+                    json=body_params,
+                    headers=headers,
+                    ssl=AIOHTTP_CLIENT_SESSION_TOOL_SERVER_SSL,
                 ) as response:
                     if response.status >= 400:
                         text = await response.text()
                         raise Exception(f"HTTP error {response.status}: {text}")
                     return await response.json()
             else:
-                async with request_method(final_url, headers=headers) as response:
+                async with request_method(
+                    final_url,
+                    headers=headers,
+                    ssl=AIOHTTP_CLIENT_SESSION_TOOL_SERVER_SSL,
+                ) as response:
                     if response.status >= 400:
                         text = await response.text()
                         raise Exception(f"HTTP error {response.status}: {text}")
@@ -549,5 +644,5 @@ async def execute_tool_server(
 
     except Exception as err:
         error = str(err)
-        print("API Request Error:", error)
+        log.exception("API Request Error:", error)
         return {"error": error}

backend/requirements.txt

@@ -1,24 +1,25 @@
 fastapi==0.115.7
-uvicorn[standard]==0.34.0
+uvicorn[standard]==0.34.2
 pydantic==2.10.6
 python-multipart==0.0.20
 
-python-socketio==5.11.3
+python-socketio==5.13.0
 python-jose==3.4.0
 passlib[bcrypt]==1.7.4
 
-requests==2.32.3
+requests==2.32.4
 aiohttp==3.11.11
 async-timeout
 aiocache
 aiofiles
+starlette-compress==1.6.0
 
 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==0.3.5
+pgvector==0.4.0
 PyMySQL==1.1.1
 bcrypt==4.3.0
 
@@ -31,29 +32,30 @@ APScheduler==3.10.4
 
 RestrictedPython==8.0
 
-loguru==0.7.2
+loguru==0.7.3
 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==0.3.19
-langchain-community==0.3.18
+langchain==0.3.24
+langchain-community==0.3.23
 
 fake-useragent==2.1.0
-chromadb==0.6.2
+chromadb==0.6.3
 pymilvus==2.5.0
 qdrant-client~=1.12.0
 opensearch-py==2.8.0
 playwright==1.49.1 # Caution: version must match docker-compose.playwright.yaml
-elasticsearch==8.17.1
-
+elasticsearch==9.0.1
+pinecone==6.0.2
 
 transformers
-sentence-transformers==3.3.1
+sentence-transformers==4.1.0
 accelerate
 colbert-ai==0.2.21
 einops==0.8.1
@@ -73,15 +75,15 @@ pandas==2.2.3
 openpyxl==3.1.5
 pyxlsb==1.0.10
 xlrd==2.0.1
-validators==0.34.0
+validators==0.35.0
 psutil
 sentencepiece
 soundfile==0.13.1
-azure-ai-documentintelligence==1.0.0
+azure-ai-documentintelligence==1.0.2
 
-pillow==11.1.0
+pillow==11.2.1
 opencv-python-headless==4.11.0.86
-rapidocr-onnxruntime==1.3.24
+rapidocr-onnxruntime==1.4.4
 rank-bm25==0.2.2
 
 onnxruntime==1.20.1
@@ -93,12 +95,12 @@ authlib==1.4.1
 
 black==25.1.0
 langfuse==2.44.0
-youtube-transcript-api==0.6.3
+youtube-transcript-api==1.0.3
 pytube==15.0.0
 
 extract_msg
 pydub
-duckduckgo-search~=7.3.2
+duckduckgo-search==8.0.2
 
 ## Google Drive
 google-api-python-client
@@ -107,13 +109,13 @@ google-auth-oauthlib
 
 ## Tests
 docker~=7.1.0
-pytest~=8.3.2
+pytest~=8.3.5
 pytest-docker~=3.1.1
 
 googleapis-common-protos==1.63.2
 google-cloud-storage==2.19.0
 
-azure-identity==1.20.0
+azure-identity==1.21.0
 azure-storage-blob==12.24.1
 
 
@@ -123,15 +125,18 @@ ldap3==2.9.1
 ## Firecrawl
 firecrawl-py==1.12.0
 
+# Sougou API SDK(Tencentcloud SDK)
+tencentcloud-sdk-python==3.0.1336
+
 ## Trace
-opentelemetry-api==1.30.0
-opentelemetry-sdk==1.30.0
-opentelemetry-exporter-otlp==1.30.0
-opentelemetry-instrumentation==0.51b0
-opentelemetry-instrumentation-fastapi==0.51b0
-opentelemetry-instrumentation-sqlalchemy==0.51b0
-opentelemetry-instrumentation-redis==0.51b0
-opentelemetry-instrumentation-requests==0.51b0
-opentelemetry-instrumentation-logging==0.51b0
-opentelemetry-instrumentation-httpx==0.51b0
-opentelemetry-instrumentation-aiohttp-client==0.51b0
+opentelemetry-api==1.32.1
+opentelemetry-sdk==1.32.1
+opentelemetry-exporter-otlp==1.32.1
+opentelemetry-instrumentation==0.53b1
+opentelemetry-instrumentation-fastapi==0.53b1
+opentelemetry-instrumentation-sqlalchemy==0.53b1
+opentelemetry-instrumentation-redis==0.53b1
+opentelemetry-instrumentation-requests==0.53b1
+opentelemetry-instrumentation-logging==0.53b1
+opentelemetry-instrumentation-httpx==0.53b1
+opentelemetry-instrumentation-aiohttp-client==0.53b1