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

@@ -93,7 +93,8 @@ body:
           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:
+        - 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),
@@ -102,7 +103,6 @@ body:
             - 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:

.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]

.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 +5,83 @@ 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

README.md

@@ -73,7 +73,27 @@ Want to learn more about Open WebUI's features? Check out our [Open WebUI docume
       </a>
     </td>
     <td>
-      Does your interface have a backend yet?<br>Try <a href="https://n8n.io/">n8n</a>
+      <a href="https://n8n.io/">n8n</a> • Does your interface have a backend yet?<br>Try <a href="https://n8n.io/">n8n</a>
+    </td>
+  </tr>
+  <tr>
+    <td>
+      <a href="https://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>
@@ -171,6 +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).

backend/open_webui/config.py

@@ -347,6 +347,24 @@ MICROSOFT_CLIENT_TENANT_ID = PersistentConfig(
     os.environ.get("MICROSOFT_CLIENT_TENANT_ID", ""),
 )
 
+MICROSOFT_CLIENT_LOGIN_BASE_URL = PersistentConfig(
+    "MICROSOFT_CLIENT_LOGIN_BASE_URL",
+    "oauth.microsoft.login_base_url",
+    os.environ.get(
+        "MICROSOFT_CLIENT_LOGIN_BASE_URL", "https://login.microsoftonline.com"
+    ),
+)
+
+MICROSOFT_CLIENT_PICTURE_URL = PersistentConfig(
+    "MICROSOFT_CLIENT_PICTURE_URL",
+    "oauth.microsoft.picture_url",
+    os.environ.get(
+        "MICROSOFT_CLIENT_PICTURE_URL",
+        "https://graph.microsoft.com/v1.0/me/photo/$value",
+    ),
+)
+
+
 MICROSOFT_OAUTH_SCOPE = PersistentConfig(
     "MICROSOFT_OAUTH_SCOPE",
     "oauth.microsoft.scope",
@@ -542,7 +560,7 @@ def load_oauth_providers():
                 name="microsoft",
                 client_id=MICROSOFT_CLIENT_ID.value,
                 client_secret=MICROSOFT_CLIENT_SECRET.value,
-                server_metadata_url=f"https://login.microsoftonline.com/{MICROSOFT_CLIENT_TENANT_ID.value}/v2.0/.well-known/openid-configuration?appid={MICROSOFT_CLIENT_ID.value}",
+                server_metadata_url=f"{MICROSOFT_CLIENT_LOGIN_BASE_URL.value}/{MICROSOFT_CLIENT_TENANT_ID.value}/v2.0/.well-known/openid-configuration?appid={MICROSOFT_CLIENT_ID.value}",
                 client_kwargs={
                     "scope": MICROSOFT_OAUTH_SCOPE.value,
                 },
@@ -551,7 +569,7 @@ def load_oauth_providers():
 
         OAUTH_PROVIDERS["microsoft"] = {
             "redirect_uri": MICROSOFT_REDIRECT_URI.value,
-            "picture_url": "https://graph.microsoft.com/v1.0/me/photo/$value",
+            "picture_url": MICROSOFT_CLIENT_PICTURE_URL.value,
             "register": microsoft_oauth_register,
         }
 
@@ -901,9 +919,7 @@ TOOL_SERVER_CONNECTIONS = PersistentConfig(
 ####################################
 
 
-WEBUI_URL = PersistentConfig(
-    "WEBUI_URL", "webui.url", os.environ.get("WEBUI_URL", "http://localhost:3000")
-)
+WEBUI_URL = PersistentConfig("WEBUI_URL", "webui.url", os.environ.get("WEBUI_URL", ""))
 
 
 ENABLE_SIGNUP = PersistentConfig(
@@ -1247,12 +1263,6 @@ if THREAD_POOL_SIZE is not None and isinstance(THREAD_POOL_SIZE, str):
         THREAD_POOL_SIZE = None
 
 
-def validate_cors_origins(origins):
-    for origin in origins:
-        if origin != "*":
-            validate_cors_origin(origin)
-
-
 def validate_cors_origin(origin):
     parsed_url = urlparse(origin)
 
@@ -1272,16 +1282,17 @@ def validate_cors_origin(origin):
 # To test CORS_ALLOW_ORIGIN locally, you can set something like
 # CORS_ALLOW_ORIGIN=http://localhost:5173;http://localhost:8080
 # in your .env file depending on your frontend port, 5173 in this case.
-CORS_ALLOW_ORIGIN = os.environ.get(
-    "CORS_ALLOW_ORIGIN", "*;http://localhost:5173;http://localhost:8080"
-).split(";")
+CORS_ALLOW_ORIGIN = os.environ.get("CORS_ALLOW_ORIGIN", "*").split(";")
 
-if "*" in CORS_ALLOW_ORIGIN:
+if CORS_ALLOW_ORIGIN == ["*"]:
     log.warning(
         "\n\nWARNING: CORS_ALLOW_ORIGIN IS SET TO '*' - NOT RECOMMENDED FOR PRODUCTION DEPLOYMENTS.\n"
     )
-
-validate_cors_origins(CORS_ALLOW_ORIGIN)
+else:
+    # You have to pick between a single wildcard or a list of origins.
+    # Doing both will result in CORS errors in the browser.
+    for origin in CORS_ALLOW_ORIGIN:
+        validate_cors_origin(origin)
 
 
 class BannerModel(BaseModel):
@@ -1413,6 +1424,35 @@ Strictly return in JSON format:
 {{MESSAGES:END:6}}
 </chat_history>"""
 
+
+FOLLOW_UP_GENERATION_PROMPT_TEMPLATE = PersistentConfig(
+    "FOLLOW_UP_GENERATION_PROMPT_TEMPLATE",
+    "task.follow_up.prompt_template",
+    os.environ.get("FOLLOW_UP_GENERATION_PROMPT_TEMPLATE", ""),
+)
+
+DEFAULT_FOLLOW_UP_GENERATION_PROMPT_TEMPLATE = """### Task:
+Suggest 3-5 relevant follow-up questions or prompts that the user might naturally ask next in this conversation as a **user**, based on the chat history, to help continue or deepen the discussion.
+### Guidelines:
+- Write all follow-up questions from the user’s point of view, directed to the assistant.
+- Make questions concise, clear, and directly related to the discussed topic(s).
+- Only suggest follow-ups that make sense given the chat content and do not repeat what was already covered.
+- If the conversation is very short or not specific, suggest more general (but relevant) follow-ups the user might ask.
+- Use the conversation's primary language; default to English if multilingual.
+- Response must be a JSON array of strings, no extra text or formatting.
+### Output:
+JSON format: { "follow_ups": ["Question 1?", "Question 2?", "Question 3?"] }
+### Chat History:
+<chat_history>
+{{MESSAGES:END:6}}
+</chat_history>"""
+
+ENABLE_FOLLOW_UP_GENERATION = PersistentConfig(
+    "ENABLE_FOLLOW_UP_GENERATION",
+    "task.follow_up.enable",
+    os.environ.get("ENABLE_FOLLOW_UP_GENERATION", "True").lower() == "true",
+)
+
 ENABLE_TAGS_GENERATION = PersistentConfig(
     "ENABLE_TAGS_GENERATION",
     "task.tags.enable",
@@ -1785,6 +1825,13 @@ PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH = int(
     os.environ.get("PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH", "1536")
 )
 
+PGVECTOR_PGCRYPTO = os.getenv("PGVECTOR_PGCRYPTO", "false").lower() == "true"
+PGVECTOR_PGCRYPTO_KEY = os.getenv("PGVECTOR_PGCRYPTO_KEY", None)
+if PGVECTOR_PGCRYPTO and not PGVECTOR_PGCRYPTO_KEY:
+    raise ValueError(
+        "PGVECTOR_PGCRYPTO is enabled but PGVECTOR_PGCRYPTO_KEY is not set. Please provide a valid key."
+    )
+
 # Pinecone
 PINECONE_API_KEY = os.environ.get("PINECONE_API_KEY", None)
 PINECONE_ENVIRONMENT = os.environ.get("PINECONE_ENVIRONMENT", None)
@@ -1848,6 +1895,61 @@ CONTENT_EXTRACTION_ENGINE = PersistentConfig(
     os.environ.get("CONTENT_EXTRACTION_ENGINE", "").lower(),
 )
 
+DATALAB_MARKER_API_KEY = PersistentConfig(
+    "DATALAB_MARKER_API_KEY",
+    "rag.datalab_marker_api_key",
+    os.environ.get("DATALAB_MARKER_API_KEY", ""),
+)
+
+DATALAB_MARKER_LANGS = PersistentConfig(
+    "DATALAB_MARKER_LANGS",
+    "rag.datalab_marker_langs",
+    os.environ.get("DATALAB_MARKER_LANGS", ""),
+)
+
+DATALAB_MARKER_USE_LLM = PersistentConfig(
+    "DATALAB_MARKER_USE_LLM",
+    "rag.DATALAB_MARKER_USE_LLM",
+    os.environ.get("DATALAB_MARKER_USE_LLM", "false").lower() == "true",
+)
+
+DATALAB_MARKER_SKIP_CACHE = PersistentConfig(
+    "DATALAB_MARKER_SKIP_CACHE",
+    "rag.datalab_marker_skip_cache",
+    os.environ.get("DATALAB_MARKER_SKIP_CACHE", "false").lower() == "true",
+)
+
+DATALAB_MARKER_FORCE_OCR = PersistentConfig(
+    "DATALAB_MARKER_FORCE_OCR",
+    "rag.datalab_marker_force_ocr",
+    os.environ.get("DATALAB_MARKER_FORCE_OCR", "false").lower() == "true",
+)
+
+DATALAB_MARKER_PAGINATE = PersistentConfig(
+    "DATALAB_MARKER_PAGINATE",
+    "rag.datalab_marker_paginate",
+    os.environ.get("DATALAB_MARKER_PAGINATE", "false").lower() == "true",
+)
+
+DATALAB_MARKER_STRIP_EXISTING_OCR = PersistentConfig(
+    "DATALAB_MARKER_STRIP_EXISTING_OCR",
+    "rag.datalab_marker_strip_existing_ocr",
+    os.environ.get("DATALAB_MARKER_STRIP_EXISTING_OCR", "false").lower() == "true",
+)
+
+DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = PersistentConfig(
+    "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION",
+    "rag.datalab_marker_disable_image_extraction",
+    os.environ.get("DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION", "false").lower()
+    == "true",
+)
+
+DATALAB_MARKER_OUTPUT_FORMAT = PersistentConfig(
+    "DATALAB_MARKER_OUTPUT_FORMAT",
+    "rag.datalab_marker_output_format",
+    os.environ.get("DATALAB_MARKER_OUTPUT_FORMAT", "markdown"),
+)
+
 EXTERNAL_DOCUMENT_LOADER_URL = PersistentConfig(
     "EXTERNAL_DOCUMENT_LOADER_URL",
     "rag.external_document_loader_url",
@@ -1890,6 +1992,40 @@ DOCLING_DO_PICTURE_DESCRIPTION = PersistentConfig(
     os.getenv("DOCLING_DO_PICTURE_DESCRIPTION", "False").lower() == "true",
 )
 
+DOCLING_PICTURE_DESCRIPTION_MODE = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_MODE",
+    "rag.docling_picture_description_mode",
+    os.getenv("DOCLING_PICTURE_DESCRIPTION_MODE", ""),
+)
+
+
+docling_picture_description_local = os.getenv("DOCLING_PICTURE_DESCRIPTION_LOCAL", "")
+try:
+    docling_picture_description_local = json.loads(docling_picture_description_local)
+except json.JSONDecodeError:
+    docling_picture_description_local = {}
+
+
+DOCLING_PICTURE_DESCRIPTION_LOCAL = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_LOCAL",
+    "rag.docling_picture_description_local",
+    docling_picture_description_local,
+)
+
+docling_picture_description_api = os.getenv("DOCLING_PICTURE_DESCRIPTION_API", "")
+try:
+    docling_picture_description_api = json.loads(docling_picture_description_api)
+except json.JSONDecodeError:
+    docling_picture_description_api = {}
+
+
+DOCLING_PICTURE_DESCRIPTION_API = PersistentConfig(
+    "DOCLING_PICTURE_DESCRIPTION_API",
+    "rag.docling_picture_description_api",
+    docling_picture_description_api,
+)
+
+
 DOCUMENT_INTELLIGENCE_ENDPOINT = PersistentConfig(
     "DOCUMENT_INTELLIGENCE_ENDPOINT",
     "rag.document_intelligence_endpoint",
@@ -2129,6 +2265,22 @@ RAG_OPENAI_API_KEY = PersistentConfig(
     os.getenv("RAG_OPENAI_API_KEY", OPENAI_API_KEY),
 )
 
+RAG_AZURE_OPENAI_BASE_URL = PersistentConfig(
+    "RAG_AZURE_OPENAI_BASE_URL",
+    "rag.azure_openai.base_url",
+    os.getenv("RAG_AZURE_OPENAI_BASE_URL", ""),
+)
+RAG_AZURE_OPENAI_API_KEY = PersistentConfig(
+    "RAG_AZURE_OPENAI_API_KEY",
+    "rag.azure_openai.api_key",
+    os.getenv("RAG_AZURE_OPENAI_API_KEY", ""),
+)
+RAG_AZURE_OPENAI_API_VERSION = PersistentConfig(
+    "RAG_AZURE_OPENAI_API_VERSION",
+    "rag.azure_openai.api_version",
+    os.getenv("RAG_AZURE_OPENAI_API_VERSION", ""),
+)
+
 RAG_OLLAMA_BASE_URL = PersistentConfig(
     "RAG_OLLAMA_BASE_URL",
     "rag.ollama.url",
@@ -2373,6 +2525,18 @@ PERPLEXITY_API_KEY = PersistentConfig(
     os.getenv("PERPLEXITY_API_KEY", ""),
 )
 
+PERPLEXITY_MODEL = PersistentConfig(
+    "PERPLEXITY_MODEL",
+    "rag.web.search.perplexity_model",
+    os.getenv("PERPLEXITY_MODEL", "sonar"),
+)
+
+PERPLEXITY_SEARCH_CONTEXT_USAGE = PersistentConfig(
+    "PERPLEXITY_SEARCH_CONTEXT_USAGE",
+    "rag.web.search.perplexity_search_context_usage",
+    os.getenv("PERPLEXITY_SEARCH_CONTEXT_USAGE", "medium"),
+)
+
 SOUGOU_API_SID = PersistentConfig(
     "SOUGOU_API_SID",
     "rag.web.search.sougou_api_sid",

backend/open_webui/constants.py

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

backend/open_webui/env.py

@@ -5,6 +5,7 @@ import os
 import pkgutil
 import sys
 import shutil
+from uuid import uuid4
 from pathlib import Path
 
 import markdown
@@ -130,6 +131,7 @@ else:
         PACKAGE_DATA = {"version": "0.0.0"}
 
 VERSION = PACKAGE_DATA["version"]
+INSTANCE_ID = os.environ.get("INSTANCE_ID", str(uuid4()))
 
 
 # Function to parse each section

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
 
@@ -53,9 +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
-    function_module, _, _ = load_function_module_by_id(pipe_id)
-    request.app.state.FUNCTIONS[pipe_id] = function_module
+    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)
@@ -226,12 +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,
     }
@@ -252,8 +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/main.py

@@ -8,6 +8,8 @@ import shutil
 import sys
 import time
 import random
+from uuid import uuid4
+
 
 from contextlib import asynccontextmanager
 from urllib.parse import urlencode, parse_qs, urlparse
@@ -19,6 +21,7 @@ from aiocache import cached
 import aiohttp
 import anyio.to_thread
 import requests
+from redis import Redis
 
 
 from fastapi import (
@@ -37,7 +40,7 @@ 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
@@ -207,11 +210,23 @@ from open_webui.config import (
     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,
@@ -219,6 +234,9 @@ from open_webui.config import (
     DOCLING_OCR_ENGINE,
     DOCLING_OCR_LANG,
     DOCLING_DO_PICTURE_DESCRIPTION,
+    DOCLING_PICTURE_DESCRIPTION_MODE,
+    DOCLING_PICTURE_DESCRIPTION_LOCAL,
+    DOCLING_PICTURE_DESCRIPTION_API,
     DOCUMENT_INTELLIGENCE_ENDPOINT,
     DOCUMENT_INTELLIGENCE_KEY,
     MISTRAL_OCR_API_KEY,
@@ -256,6 +274,8 @@ from open_webui.config import (
     BRAVE_SEARCH_API_KEY,
     EXA_API_KEY,
     PERPLEXITY_API_KEY,
+    PERPLEXITY_MODEL,
+    PERPLEXITY_SEARCH_CONTEXT_USAGE,
     SOUGOU_API_SID,
     SOUGOU_API_SK,
     KAGI_SEARCH_API_KEY,
@@ -347,10 +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,
@@ -372,6 +394,7 @@ from open_webui.env import (
     SAFE_MODE,
     SRC_LOG_LEVELS,
     VERSION,
+    INSTANCE_ID,
     WEBUI_BUILD_HASH,
     WEBUI_SECRET_KEY,
     WEBUI_SESSION_COOKIE_SAME_SITE,
@@ -399,6 +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
 
@@ -412,8 +436,10 @@ from open_webui.utils.auth import (
 from open_webui.utils.plugin import install_tool_and_function_dependencies
 from open_webui.utils.oauth import OAuthManager
 from open_webui.utils.security_headers import SecurityHeadersMiddleware
+from open_webui.utils.redis import get_redis_connection
 
 from open_webui.tasks import (
+    redis_task_command_listener,
     list_task_ids_by_chat_id,
     stop_task,
     list_tasks,
@@ -465,7 +491,9 @@ https://github.com/open-webui/open-webui
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
+    app.state.instance_id = INSTANCE_ID
     start_logger()
+
     if RESET_CONFIG_ON_START:
         reset_config()
 
@@ -477,6 +505,19 @@ async def lifespan(app: FastAPI):
     log.info("Installing external dependencies of functions and tools...")
     install_tool_and_function_dependencies()
 
+    app.state.redis = get_redis_connection(
+        redis_url=REDIS_URL,
+        redis_sentinels=get_sentinels_from_env(
+            REDIS_SENTINEL_HOSTS, REDIS_SENTINEL_PORT
+        ),
+        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
@@ -485,6 +526,9 @@ async def lifespan(app: FastAPI):
 
     yield
 
+    if hasattr(app.state, "redis_task_command_listener"):
+        app.state.redis_task_command_listener.cancel()
+
 
 app = FastAPI(
     title="Open WebUI",
@@ -496,10 +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
@@ -637,8 +683,12 @@ app.state.WEBUI_AUTH_SIGNOUT_REDIRECT_URL = WEBUI_AUTH_SIGNOUT_REDIRECT_URL
 app.state.EXTERNAL_PWA_MANIFEST_URL = EXTERNAL_PWA_MANIFEST_URL
 
 app.state.USER_COUNT = None
+
 app.state.TOOLS = {}
+app.state.TOOL_CONTENTS = {}
+
 app.state.FUNCTIONS = {}
+app.state.FUNCTION_CONTENTS = {}
 
 ########################################
 #
@@ -662,6 +712,17 @@ app.state.config.ENABLE_RAG_HYBRID_SEARCH = ENABLE_RAG_HYBRID_SEARCH
 app.state.config.ENABLE_WEB_LOADER_SSL_VERIFICATION = ENABLE_WEB_LOADER_SSL_VERIFICATION
 
 app.state.config.CONTENT_EXTRACTION_ENGINE = CONTENT_EXTRACTION_ENGINE
+app.state.config.DATALAB_MARKER_API_KEY = DATALAB_MARKER_API_KEY
+app.state.config.DATALAB_MARKER_LANGS = DATALAB_MARKER_LANGS
+app.state.config.DATALAB_MARKER_SKIP_CACHE = DATALAB_MARKER_SKIP_CACHE
+app.state.config.DATALAB_MARKER_FORCE_OCR = DATALAB_MARKER_FORCE_OCR
+app.state.config.DATALAB_MARKER_PAGINATE = DATALAB_MARKER_PAGINATE
+app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR = DATALAB_MARKER_STRIP_EXISTING_OCR
+app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = (
+    DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
+)
+app.state.config.DATALAB_MARKER_USE_LLM = DATALAB_MARKER_USE_LLM
+app.state.config.DATALAB_MARKER_OUTPUT_FORMAT = DATALAB_MARKER_OUTPUT_FORMAT
 app.state.config.EXTERNAL_DOCUMENT_LOADER_URL = EXTERNAL_DOCUMENT_LOADER_URL
 app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY = EXTERNAL_DOCUMENT_LOADER_API_KEY
 app.state.config.TIKA_SERVER_URL = TIKA_SERVER_URL
@@ -669,6 +730,9 @@ app.state.config.DOCLING_SERVER_URL = DOCLING_SERVER_URL
 app.state.config.DOCLING_OCR_ENGINE = DOCLING_OCR_ENGINE
 app.state.config.DOCLING_OCR_LANG = DOCLING_OCR_LANG
 app.state.config.DOCLING_DO_PICTURE_DESCRIPTION = DOCLING_DO_PICTURE_DESCRIPTION
+app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE = DOCLING_PICTURE_DESCRIPTION_MODE
+app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL = DOCLING_PICTURE_DESCRIPTION_LOCAL
+app.state.config.DOCLING_PICTURE_DESCRIPTION_API = DOCLING_PICTURE_DESCRIPTION_API
 app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT = DOCUMENT_INTELLIGENCE_ENDPOINT
 app.state.config.DOCUMENT_INTELLIGENCE_KEY = DOCUMENT_INTELLIGENCE_KEY
 app.state.config.MISTRAL_OCR_API_KEY = MISTRAL_OCR_API_KEY
@@ -693,6 +757,10 @@ app.state.config.RAG_TEMPLATE = RAG_TEMPLATE
 app.state.config.RAG_OPENAI_API_BASE_URL = RAG_OPENAI_API_BASE_URL
 app.state.config.RAG_OPENAI_API_KEY = RAG_OPENAI_API_KEY
 
+app.state.config.RAG_AZURE_OPENAI_BASE_URL = RAG_AZURE_OPENAI_BASE_URL
+app.state.config.RAG_AZURE_OPENAI_API_KEY = RAG_AZURE_OPENAI_API_KEY
+app.state.config.RAG_AZURE_OPENAI_API_VERSION = RAG_AZURE_OPENAI_API_VERSION
+
 app.state.config.RAG_OLLAMA_BASE_URL = RAG_OLLAMA_BASE_URL
 app.state.config.RAG_OLLAMA_API_KEY = RAG_OLLAMA_API_KEY
 
@@ -740,6 +808,8 @@ app.state.config.BING_SEARCH_V7_ENDPOINT = BING_SEARCH_V7_ENDPOINT
 app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY = BING_SEARCH_V7_SUBSCRIPTION_KEY
 app.state.config.EXA_API_KEY = EXA_API_KEY
 app.state.config.PERPLEXITY_API_KEY = PERPLEXITY_API_KEY
+app.state.config.PERPLEXITY_MODEL = PERPLEXITY_MODEL
+app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE = PERPLEXITY_SEARCH_CONTEXT_USAGE
 app.state.config.SOUGOU_API_SID = SOUGOU_API_SID
 app.state.config.SOUGOU_API_SK = SOUGOU_API_SK
 app.state.config.EXTERNAL_WEB_SEARCH_URL = EXTERNAL_WEB_SEARCH_URL
@@ -787,14 +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
+    ),
 )
 
 ########################################
@@ -915,6 +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
@@ -922,6 +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
@@ -1153,6 +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,
@@ -1294,26 +1412,30 @@ async def chat_action(
 
 
 @app.post("/api/tasks/stop/{task_id}")
-async def stop_task_endpoint(task_id: str, user=Depends(get_verified_user)):
+async def stop_task_endpoint(
+    request: Request, task_id: str, user=Depends(get_verified_user)
+):
     try:
-        result = await stop_task(task_id)
+        result = await stop_task(request, task_id)
         return result
     except ValueError as e:
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(e))
 
 
 @app.get("/api/tasks")
-async def list_tasks_endpoint(user=Depends(get_verified_user)):
-    return {"tasks": list_tasks()}
+async def list_tasks_endpoint(request: Request, user=Depends(get_verified_user)):
+    return {"tasks": await list_tasks(request)}
 
 
 @app.get("/api/tasks/chat/{chat_id}")
-async def list_tasks_by_chat_id_endpoint(chat_id: str, user=Depends(get_verified_user)):
+async def list_tasks_by_chat_id_endpoint(
+    request: Request, chat_id: str, user=Depends(get_verified_user)
+):
     chat = Chats.get_chat_by_id(chat_id)
     if chat is None or chat.user_id != user.id:
         return {"task_ids": []}
 
-    task_ids = list_task_ids_by_chat_id(chat_id)
+    task_ids = await list_task_ids_by_chat_id(request, chat_id)
 
     print(f"Task IDs for chat {chat_id}: {task_ids}")
     return {"task_ids": task_ids}
@@ -1584,7 +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/models/users.py

@@ -95,6 +95,7 @@ class UserRoleUpdateForm(BaseModel):
 
 
 class UserUpdateForm(BaseModel):
+    role: str
     name: str
     email: str
     profile_image_url: str
@@ -369,7 +370,7 @@ class UsersTable:
         except Exception:
             return False
 
-    def update_user_api_key_by_id(self, id: str, api_key: str) -> str:
+    def update_user_api_key_by_id(self, id: str, api_key: str) -> bool:
         try:
             with get_db() as db:
                 result = db.query(User).filter_by(id=id).update({"api_key": api_key})

backend/open_webui/retrieval/loaders/datalab_marker.py

@@ -0,0 +1,251 @@
+import os
+import time
+import requests
+import logging
+import json
+from typing import List, Optional
+from langchain_core.documents import Document
+from fastapi import HTTPException, status
+
+log = logging.getLogger(__name__)
+
+
+class DatalabMarkerLoader:
+    def __init__(
+        self,
+        file_path: str,
+        api_key: str,
+        langs: Optional[str] = None,
+        use_llm: bool = False,
+        skip_cache: bool = False,
+        force_ocr: bool = False,
+        paginate: bool = False,
+        strip_existing_ocr: bool = False,
+        disable_image_extraction: bool = False,
+        output_format: str = None,
+    ):
+        self.file_path = file_path
+        self.api_key = api_key
+        self.langs = langs
+        self.use_llm = use_llm
+        self.skip_cache = skip_cache
+        self.force_ocr = force_ocr
+        self.paginate = paginate
+        self.strip_existing_ocr = strip_existing_ocr
+        self.disable_image_extraction = disable_image_extraction
+        self.output_format = output_format
+
+    def _get_mime_type(self, filename: str) -> str:
+        ext = filename.rsplit(".", 1)[-1].lower()
+        mime_map = {
+            "pdf": "application/pdf",
+            "xls": "application/vnd.ms-excel",
+            "xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+            "ods": "application/vnd.oasis.opendocument.spreadsheet",
+            "doc": "application/msword",
+            "docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+            "odt": "application/vnd.oasis.opendocument.text",
+            "ppt": "application/vnd.ms-powerpoint",
+            "pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+            "odp": "application/vnd.oasis.opendocument.presentation",
+            "html": "text/html",
+            "epub": "application/epub+zip",
+            "png": "image/png",
+            "jpeg": "image/jpeg",
+            "jpg": "image/jpeg",
+            "webp": "image/webp",
+            "gif": "image/gif",
+            "tiff": "image/tiff",
+        }
+        return mime_map.get(ext, "application/octet-stream")
+
+    def check_marker_request_status(self, request_id: str) -> dict:
+        url = f"https://www.datalab.to/api/v1/marker/{request_id}"
+        headers = {"X-Api-Key": self.api_key}
+        try:
+            response = requests.get(url, headers=headers)
+            response.raise_for_status()
+            result = response.json()
+            log.info(f"Marker API status check for request {request_id}: {result}")
+            return result
+        except requests.HTTPError as e:
+            log.error(f"Error checking Marker request status: {e}")
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY,
+                detail=f"Failed to check Marker request: {e}",
+            )
+        except ValueError as e:
+            log.error(f"Invalid JSON checking Marker request: {e}")
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY, detail=f"Invalid JSON: {e}"
+            )
+
+    def load(self) -> List[Document]:
+        url = "https://www.datalab.to/api/v1/marker"
+        filename = os.path.basename(self.file_path)
+        mime_type = self._get_mime_type(filename)
+        headers = {"X-Api-Key": self.api_key}
+
+        form_data = {
+            "langs": self.langs,
+            "use_llm": str(self.use_llm).lower(),
+            "skip_cache": str(self.skip_cache).lower(),
+            "force_ocr": str(self.force_ocr).lower(),
+            "paginate": str(self.paginate).lower(),
+            "strip_existing_ocr": str(self.strip_existing_ocr).lower(),
+            "disable_image_extraction": str(self.disable_image_extraction).lower(),
+            "output_format": self.output_format,
+        }
+
+        log.info(
+            f"Datalab Marker POST request parameters: {{'filename': '{filename}', 'mime_type': '{mime_type}', **{form_data}}}"
+        )
+
+        try:
+            with open(self.file_path, "rb") as f:
+                files = {"file": (filename, f, mime_type)}
+                response = requests.post(
+                    url, data=form_data, files=files, headers=headers
+                )
+                response.raise_for_status()
+                result = response.json()
+        except FileNotFoundError:
+            raise HTTPException(
+                status.HTTP_404_NOT_FOUND, detail=f"File not found: {self.file_path}"
+            )
+        except requests.HTTPError as e:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Datalab Marker request failed: {e}",
+            )
+        except ValueError as e:
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY, detail=f"Invalid JSON response: {e}"
+            )
+        except Exception as e:
+            raise HTTPException(status.HTTP_500_INTERNAL_SERVER_ERROR, detail=str(e))
+
+        if not result.get("success"):
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Datalab Marker request failed: {result.get('error', 'Unknown error')}",
+            )
+
+        check_url = result.get("request_check_url")
+        request_id = result.get("request_id")
+        if not check_url:
+            raise HTTPException(
+                status.HTTP_502_BAD_GATEWAY, detail="No request_check_url returned."
+            )
+
+        for _ in range(300):  # Up to 10 minutes
+            time.sleep(2)
+            try:
+                poll_response = requests.get(check_url, headers=headers)
+                poll_response.raise_for_status()
+                poll_result = poll_response.json()
+            except (requests.HTTPError, ValueError) as e:
+                raw_body = poll_response.text
+                log.error(f"Polling error: {e}, response body: {raw_body}")
+                raise HTTPException(
+                    status.HTTP_502_BAD_GATEWAY, detail=f"Polling failed: {e}"
+                )
+
+            status_val = poll_result.get("status")
+            success_val = poll_result.get("success")
+
+            if status_val == "complete":
+                summary = {
+                    k: poll_result.get(k)
+                    for k in (
+                        "status",
+                        "output_format",
+                        "success",
+                        "error",
+                        "page_count",
+                        "total_cost",
+                    )
+                }
+                log.info(
+                    f"Marker processing completed successfully: {json.dumps(summary, indent=2)}"
+                )
+                break
+
+            if status_val == "failed" or success_val is False:
+                log.error(
+                    f"Marker poll failed full response: {json.dumps(poll_result, indent=2)}"
+                )
+                error_msg = (
+                    poll_result.get("error")
+                    or "Marker returned failure without error message"
+                )
+                raise HTTPException(
+                    status.HTTP_400_BAD_REQUEST,
+                    detail=f"Marker processing failed: {error_msg}",
+                )
+        else:
+            raise HTTPException(
+                status.HTTP_504_GATEWAY_TIMEOUT, detail="Marker processing timed out"
+            )
+
+        if not poll_result.get("success", False):
+            error_msg = poll_result.get("error") or "Unknown processing error"
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Final processing failed: {error_msg}",
+            )
+
+        content_key = self.output_format.lower()
+        raw_content = poll_result.get(content_key)
+
+        if content_key == "json":
+            full_text = json.dumps(raw_content, indent=2)
+        elif content_key in {"markdown", "html"}:
+            full_text = str(raw_content).strip()
+        else:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=f"Unsupported output format: {self.output_format}",
+            )
+
+        if not full_text:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail="Datalab Marker returned empty content",
+            )
+
+        marker_output_dir = os.path.join("/app/backend/data/uploads", "marker_output")
+        os.makedirs(marker_output_dir, exist_ok=True)
+
+        file_ext_map = {"markdown": "md", "json": "json", "html": "html"}
+        file_ext = file_ext_map.get(content_key, "txt")
+        output_filename = f"{os.path.splitext(filename)[0]}.{file_ext}"
+        output_path = os.path.join(marker_output_dir, output_filename)
+
+        try:
+            with open(output_path, "w", encoding="utf-8") as f:
+                f.write(full_text)
+            log.info(f"Saved Marker output to: {output_path}")
+        except Exception as e:
+            log.warning(f"Failed to write marker output to disk: {e}")
+
+        metadata = {
+            "source": filename,
+            "output_format": poll_result.get("output_format", self.output_format),
+            "page_count": poll_result.get("page_count", 0),
+            "processed_with_llm": self.use_llm,
+            "request_id": request_id or "",
+        }
+
+        images = poll_result.get("images", {})
+        if images:
+            metadata["image_count"] = len(images)
+            metadata["images"] = json.dumps(list(images.keys()))
+
+        for k, v in metadata.items():
+            if isinstance(v, (dict, list)):
+                metadata[k] = json.dumps(v)
+            elif v is None:
+                metadata[k] = ""
+
+        return [Document(page_content=full_text, metadata=metadata)]

backend/open_webui/retrieval/loaders/main.py

@@ -2,6 +2,7 @@ import requests
 import logging
 import ftfy
 import sys
+import json
 
 from langchain_community.document_loaders import (
     AzureAIDocumentIntelligenceLoader,
@@ -21,9 +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
 
@@ -74,7 +77,6 @@ known_source_ext = [
     "swift",
     "vue",
     "svelte",
-    "msg",
     "ex",
     "exs",
     "erl",
@@ -145,17 +147,32 @@ class DoclingLoader:
                 )
             }
 
-            params = {
-                "image_export_mode": "placeholder",
-                "table_mode": "accurate",
-            }
+            params = {"image_export_mode": "placeholder", "table_mode": "accurate"}
 
             if self.params:
-                if self.params.get("do_picture_classification"):
-                    params["do_picture_classification"] = self.params.get(
-                        "do_picture_classification"
+                if self.params.get("do_picture_description"):
+                    params["do_picture_description"] = self.params.get(
+                        "do_picture_description"
                     )
 
+                    picture_description_mode = self.params.get(
+                        "picture_description_mode", ""
+                    ).lower()
+
+                    if picture_description_mode == "local" and self.params.get(
+                        "picture_description_local", {}
+                    ):
+                        params["picture_description_local"] = 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"] = [
@@ -236,21 +253,67 @@ class Loader:
                     mime_type=file_content_type,
                     extract_images=self.kwargs.get("PDF_EXTRACT_IMAGES"),
                 )
+        elif (
+            self.engine == "datalab_marker"
+            and self.kwargs.get("DATALAB_MARKER_API_KEY")
+            and file_ext
+            in [
+                "pdf",
+                "xls",
+                "xlsx",
+                "ods",
+                "doc",
+                "docx",
+                "odt",
+                "ppt",
+                "pptx",
+                "odp",
+                "html",
+                "epub",
+                "png",
+                "jpeg",
+                "jpg",
+                "webp",
+                "gif",
+                "tiff",
+            ]
+        ):
+            loader = DatalabMarkerLoader(
+                file_path=file_path,
+                api_key=self.kwargs["DATALAB_MARKER_API_KEY"],
+                langs=self.kwargs.get("DATALAB_MARKER_LANGS"),
+                use_llm=self.kwargs.get("DATALAB_MARKER_USE_LLM", False),
+                skip_cache=self.kwargs.get("DATALAB_MARKER_SKIP_CACHE", False),
+                force_ocr=self.kwargs.get("DATALAB_MARKER_FORCE_OCR", False),
+                paginate=self.kwargs.get("DATALAB_MARKER_PAGINATE", False),
+                strip_existing_ocr=self.kwargs.get(
+                    "DATALAB_MARKER_STRIP_EXISTING_OCR", False
+                ),
+                disable_image_extraction=self.kwargs.get(
+                    "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION", False
+                ),
+                output_format=self.kwargs.get(
+                    "DATALAB_MARKER_OUTPUT_FORMAT", "markdown"
+                ),
+            )
         elif self.engine == "docling" and self.kwargs.get("DOCLING_SERVER_URL"):
             if self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:
+                # Build params for DoclingLoader
+                params = self.kwargs.get("DOCLING_PARAMS", {})
+                if not isinstance(params, dict):
+                    try:
+                        params = json.loads(params)
+                    except json.JSONDecodeError:
+                        log.error("Invalid DOCLING_PARAMS format, expected JSON object")
+                        params = {}
+
                 loader = DoclingLoader(
                     url=self.kwargs.get("DOCLING_SERVER_URL"),
                     file_path=file_path,
                     mime_type=file_content_type,
-                    params={
-                        "ocr_engine": self.kwargs.get("DOCLING_OCR_ENGINE"),
-                        "ocr_lang": self.kwargs.get("DOCLING_OCR_LANG"),
-                        "do_picture_classification": self.kwargs.get(
-                            "DOCLING_DO_PICTURE_DESCRIPTION"
-                        ),
-                    },
+                    params=params,
                 )
         elif (
             self.engine == "document_intelligence"

backend/open_webui/retrieval/loaders/mistral.py

@@ -20,6 +20,14 @@ class MistralLoader:
     """
     Enhanced Mistral OCR loader with both sync and async support.
     Loads documents by processing them through the Mistral OCR API.
+
+    Performance Optimizations:
+    - Differentiated timeouts for different operations
+    - Intelligent retry logic with exponential backoff
+    - Memory-efficient file streaming for large files
+    - Connection pooling and keepalive optimization
+    - Semaphore-based concurrency control for batch processing
+    - Enhanced error handling with retryable error classification
     """
 
     BASE_API_URL = "https://api.mistral.ai/v1"
@@ -53,17 +61,40 @@ class MistralLoader:
         self.max_retries = max_retries
         self.debug = enable_debug_logging
 
-        # Pre-compute file info for performance
+        # PERFORMANCE OPTIMIZATION: Differentiated timeouts for different operations
+        # This prevents long-running OCR operations from affecting quick operations
+        # and improves user experience by failing fast on operations that should be quick
+        self.upload_timeout = min(
+            timeout, 120
+        )  # Cap upload at 2 minutes - prevents hanging on large files
+        self.url_timeout = (
+            30  # URL requests should be fast - fail quickly if API is slow
+        )
+        self.ocr_timeout = (
+            timeout  # OCR can take the full timeout - this is the heavy operation
+        )
+        self.cleanup_timeout = (
+            30  # Cleanup should be quick - don't hang on file deletion
+        )
+
+        # PERFORMANCE OPTIMIZATION: Pre-compute file info to avoid repeated filesystem calls
+        # This avoids multiple os.path.basename() and os.path.getsize() calls during processing
         self.file_name = os.path.basename(file_path)
         self.file_size = os.path.getsize(file_path)
 
+        # ENHANCEMENT: Added User-Agent for better API tracking and debugging
         self.headers = {
             "Authorization": f"Bearer {self.api_key}",
-            "User-Agent": "OpenWebUI-MistralLoader/2.0",
+            "User-Agent": "OpenWebUI-MistralLoader/2.0",  # Helps API provider track usage
         }
 
     def _debug_log(self, message: str, *args) -> None:
-        """Conditional debug logging for performance."""
+        """
+        PERFORMANCE OPTIMIZATION: Conditional debug logging for performance.
+
+        Only processes debug messages when debug mode is enabled, avoiding
+        string formatting overhead in production environments.
+        """
         if self.debug:
             log.debug(message, *args)
 
@@ -115,53 +146,118 @@ class MistralLoader:
             log.error(f"Unexpected error processing response: {e}")
             raise
 
+    def _is_retryable_error(self, error: Exception) -> bool:
+        """
+        ENHANCEMENT: Intelligent error classification for retry logic.
+
+        Determines if an error is retryable based on its type and status code.
+        This prevents wasting time retrying errors that will never succeed
+        (like authentication errors) while ensuring transient errors are retried.
+
+        Retryable errors:
+        - Network connection errors (temporary network issues)
+        - Timeouts (server might be temporarily overloaded)
+        - Server errors (5xx status codes - server-side issues)
+        - Rate limiting (429 status - temporary throttling)
+
+        Non-retryable errors:
+        - Authentication errors (401, 403 - won't fix with retry)
+        - Bad request errors (400 - malformed request)
+        - Not found errors (404 - resource doesn't exist)
+        """
+        if isinstance(error, requests.exceptions.ConnectionError):
+            return True  # Network issues are usually temporary
+        if isinstance(error, requests.exceptions.Timeout):
+            return True  # Timeouts might resolve on retry
+        if isinstance(error, requests.exceptions.HTTPError):
+            # Only retry on server errors (5xx) or rate limits (429)
+            if hasattr(error, "response") and error.response is not None:
+                status_code = error.response.status_code
+                return status_code >= 500 or status_code == 429
+            return False
+        if isinstance(
+            error, (aiohttp.ClientConnectionError, aiohttp.ServerTimeoutError)
+        ):
+            return True  # Async network/timeout errors are retryable
+        if isinstance(error, aiohttp.ClientResponseError):
+            return error.status >= 500 or error.status == 429
+        return False  # All other errors are non-retryable
+
     def _retry_request_sync(self, request_func, *args, **kwargs):
-        """Synchronous retry logic with exponential backoff."""
+        """
+        ENHANCEMENT: Synchronous retry logic with intelligent error classification.
+
+        Uses exponential backoff with jitter to avoid thundering herd problems.
+        The wait time increases exponentially but is capped at 30 seconds to
+        prevent excessive delays. Only retries errors that are likely to succeed
+        on subsequent attempts.
+        """
         for attempt in range(self.max_retries):
             try:
                 return request_func(*args, **kwargs)
-            except (requests.exceptions.RequestException, Exception) as e:
-                if attempt == self.max_retries - 1:
+            except Exception as e:
+                if attempt == self.max_retries - 1 or not self._is_retryable_error(e):
                     raise
 
-                wait_time = (2**attempt) + 0.5
+                # PERFORMANCE OPTIMIZATION: Exponential backoff with cap
+                # Prevents overwhelming the server while ensuring reasonable retry delays
+                wait_time = min((2**attempt) + 0.5, 30)  # Cap at 30 seconds
                 log.warning(
-                    f"Request failed (attempt {attempt + 1}/{self.max_retries}): {e}. Retrying in {wait_time}s..."
+                    f"Retryable error (attempt {attempt + 1}/{self.max_retries}): {e}. "
+                    f"Retrying in {wait_time}s..."
                 )
                 time.sleep(wait_time)
 
     async def _retry_request_async(self, request_func, *args, **kwargs):
-        """Async retry logic with exponential backoff."""
+        """
+        ENHANCEMENT: Async retry logic with intelligent error classification.
+
+        Async version of retry logic that doesn't block the event loop during
+        wait periods. Uses the same exponential backoff strategy as sync version.
+        """
         for attempt in range(self.max_retries):
             try:
                 return await request_func(*args, **kwargs)
-            except (aiohttp.ClientError, asyncio.TimeoutError) as e:
-                if attempt == self.max_retries - 1:
+            except Exception as e:
+                if attempt == self.max_retries - 1 or not self._is_retryable_error(e):
                     raise
 
-                wait_time = (2**attempt) + 0.5
+                # PERFORMANCE OPTIMIZATION: Non-blocking exponential backoff
+                wait_time = min((2**attempt) + 0.5, 30)  # Cap at 30 seconds
                 log.warning(
-                    f"Request failed (attempt {attempt + 1}/{self.max_retries}): {e}. Retrying in {wait_time}s..."
+                    f"Retryable error (attempt {attempt + 1}/{self.max_retries}): {e}. "
+                    f"Retrying in {wait_time}s..."
                 )
-                await asyncio.sleep(wait_time)
+                await asyncio.sleep(wait_time)  # Non-blocking wait
 
     def _upload_file(self) -> str:
-        """Uploads the file to Mistral for OCR processing (sync version)."""
+        """
+        PERFORMANCE OPTIMIZATION: Enhanced file upload with streaming consideration.
+
+        Uploads the file to Mistral for OCR processing (sync version).
+        Uses context manager for file handling to ensure proper resource cleanup.
+        Although streaming is not enabled for this endpoint, the file is opened
+        in a context manager to minimize memory usage duration.
+        """
         log.info("Uploading file to Mistral API")
         url = f"{self.BASE_API_URL}/files"
-        file_name = os.path.basename(self.file_path)
 
         def upload_request():
+            # MEMORY OPTIMIZATION: Use context manager to minimize file handle lifetime
+            # This ensures the file is closed immediately after reading, reducing memory usage
             with open(self.file_path, "rb") as f:
-                files = {"file": (file_name, f, "application/pdf")}
+                files = {"file": (self.file_name, f, "application/pdf")}
                 data = {"purpose": "ocr"}
 
+                # NOTE: stream=False is required for this endpoint
+                # The Mistral API doesn't support chunked uploads for this endpoint
                 response = requests.post(
                     url,
                     headers=self.headers,
                     files=files,
                     data=data,
-                    timeout=self.timeout,
+                    timeout=self.upload_timeout,  # Use specialized upload timeout
+                    stream=False,  # Keep as False for this endpoint
                 )
 
             return self._handle_response(response)
@@ -209,7 +305,7 @@ class MistralLoader:
                 url,
                 data=writer,
                 headers=self.headers,
-                timeout=aiohttp.ClientTimeout(total=self.timeout),
+                timeout=aiohttp.ClientTimeout(total=self.upload_timeout),
             ) as response:
                 return await self._handle_response_async(response)
 
@@ -231,7 +327,7 @@ class MistralLoader:
 
         def url_request():
             response = requests.get(
-                url, headers=signed_url_headers, params=params, timeout=self.timeout
+                url, headers=signed_url_headers, params=params, timeout=self.url_timeout
             )
             return self._handle_response(response)
 
@@ -261,7 +357,7 @@ class MistralLoader:
                 url,
                 headers=headers,
                 params=params,
-                timeout=aiohttp.ClientTimeout(total=self.timeout),
+                timeout=aiohttp.ClientTimeout(total=self.url_timeout),
             ) as response:
                 return await self._handle_response_async(response)
 
@@ -294,7 +390,7 @@ class MistralLoader:
 
         def ocr_request():
             response = requests.post(
-                url, headers=ocr_headers, json=payload, timeout=self.timeout
+                url, headers=ocr_headers, json=payload, timeout=self.ocr_timeout
             )
             return self._handle_response(response)
 
@@ -336,7 +432,7 @@ class MistralLoader:
                 url,
                 json=payload,
                 headers=headers,
-                timeout=aiohttp.ClientTimeout(total=self.timeout),
+                timeout=aiohttp.ClientTimeout(total=self.ocr_timeout),
             ) as response:
                 ocr_response = await self._handle_response_async(response)
 
@@ -353,7 +449,9 @@ class MistralLoader:
         url = f"{self.BASE_API_URL}/files/{file_id}"
 
         try:
-            response = requests.delete(url, headers=self.headers, timeout=30)
+            response = requests.delete(
+                url, headers=self.headers, timeout=self.cleanup_timeout
+            )
             delete_response = self._handle_response(response)
             log.info(f"File deleted successfully: {delete_response}")
         except Exception as e:
@@ -372,7 +470,7 @@ class MistralLoader:
                     url=f"{self.BASE_API_URL}/files/{file_id}",
                     headers=self.headers,
                     timeout=aiohttp.ClientTimeout(
-                        total=30
+                        total=self.cleanup_timeout
                     ),  # Shorter timeout for cleanup
                 ) as response:
                     return await self._handle_response_async(response)
@@ -388,29 +486,39 @@ class MistralLoader:
     async def _get_session(self):
         """Context manager for HTTP session with optimized settings."""
         connector = aiohttp.TCPConnector(
-            limit=10,  # Total connection limit
-            limit_per_host=5,  # Per-host connection limit
-            ttl_dns_cache=300,  # DNS cache TTL
+            limit=20,  # Increased total connection limit for better throughput
+            limit_per_host=10,  # Increased per-host limit for API endpoints
+            ttl_dns_cache=600,  # Longer DNS cache TTL (10 minutes)
             use_dns_cache=True,
-            keepalive_timeout=30,
+            keepalive_timeout=60,  # Increased keepalive for connection reuse
             enable_cleanup_closed=True,
+            force_close=False,  # Allow connection reuse
+            resolver=aiohttp.AsyncResolver(),  # Use async DNS resolver
+        )
+
+        timeout = aiohttp.ClientTimeout(
+            total=self.timeout,
+            connect=30,  # Connection timeout
+            sock_read=60,  # Socket read timeout
         )
 
         async with aiohttp.ClientSession(
             connector=connector,
-            timeout=aiohttp.ClientTimeout(total=self.timeout),
+            timeout=timeout,
             headers={"User-Agent": "OpenWebUI-MistralLoader/2.0"},
+            raise_for_status=False,  # We handle status codes manually
         ) as session:
             yield session
 
     def _process_results(self, ocr_response: Dict[str, Any]) -> List[Document]:
-        """Process OCR results into Document objects with enhanced metadata."""
+        """Process OCR results into Document objects with enhanced metadata and memory efficiency."""
         pages_data = ocr_response.get("pages")
         if not pages_data:
             log.warning("No pages found in OCR response.")
             return [
                 Document(
-                    page_content="No text content found", metadata={"error": "no_pages"}
+                    page_content="No text content found",
+                    metadata={"error": "no_pages", "file_name": self.file_name},
                 )
             ]
 
@@ -418,41 +526,44 @@ class MistralLoader:
         total_pages = len(pages_data)
         skipped_pages = 0
 
+        # Process pages in a memory-efficient way
         for page_data in pages_data:
             page_content = page_data.get("markdown")
             page_index = page_data.get("index")  # API uses 0-based index
 
-            if page_content is not None and page_index is not None:
-                # Clean up content efficiently
-                cleaned_content = (
-                    page_content.strip()
-                    if isinstance(page_content, str)
-                    else str(page_content)
-                )
-
-                if cleaned_content:  # Only add non-empty pages
-                    documents.append(
-                        Document(
-                            page_content=cleaned_content,
-                            metadata={
-                                "page": page_index,  # 0-based index from API
-                                "page_label": page_index
-                                + 1,  # 1-based label for convenience
-                                "total_pages": total_pages,
-                                "file_name": self.file_name,
-                                "file_size": self.file_size,
-                                "processing_engine": "mistral-ocr",
-                            },
-                        )
-                    )
-                else:
-                    skipped_pages += 1
-                    self._debug_log(f"Skipping empty page {page_index}")
-            else:
+            if page_content is None or page_index is None:
                 skipped_pages += 1
                 self._debug_log(
-                    f"Skipping page due to missing 'markdown' or 'index'. Data: {page_data}"
+                    f"Skipping page due to missing 'markdown' or 'index'. Data keys: {list(page_data.keys())}"
                 )
+                continue
+
+            # Clean up content efficiently with early exit for empty content
+            if isinstance(page_content, str):
+                cleaned_content = page_content.strip()
+            else:
+                cleaned_content = str(page_content).strip()
+
+            if not cleaned_content:
+                skipped_pages += 1
+                self._debug_log(f"Skipping empty page {page_index}")
+                continue
+
+            # Create document with optimized metadata
+            documents.append(
+                Document(
+                    page_content=cleaned_content,
+                    metadata={
+                        "page": page_index,  # 0-based index from API
+                        "page_label": page_index + 1,  # 1-based label for convenience
+                        "total_pages": total_pages,
+                        "file_name": self.file_name,
+                        "file_size": self.file_size,
+                        "processing_engine": "mistral-ocr",
+                        "content_length": len(cleaned_content),
+                    },
+                )
+            )
 
         if skipped_pages > 0:
             log.info(
@@ -467,7 +578,11 @@ class MistralLoader:
             return [
                 Document(
                     page_content="No valid text content found in document",
-                    metadata={"error": "no_valid_pages", "total_pages": total_pages},
+                    metadata={
+                        "error": "no_valid_pages",
+                        "total_pages": total_pages,
+                        "file_name": self.file_name,
+                    },
                 )
             ]
 
@@ -585,12 +700,14 @@ class MistralLoader:
     @staticmethod
     async def load_multiple_async(
         loaders: List["MistralLoader"],
+        max_concurrent: int = 5,  # Limit concurrent requests
     ) -> List[List[Document]]:
         """
-        Process multiple files concurrently for maximum performance.
+        Process multiple files concurrently with controlled concurrency.
 
         Args:
             loaders: List of MistralLoader instances
+            max_concurrent: Maximum number of concurrent requests
 
         Returns:
             List of document lists, one for each loader
@@ -598,11 +715,20 @@ class MistralLoader:
         if not loaders:
             return []
 
-        log.info(f"Starting concurrent processing of {len(loaders)} files")
+        log.info(
+            f"Starting concurrent processing of {len(loaders)} files with max {max_concurrent} concurrent"
+        )
         start_time = time.time()
 
-        # Process all files concurrently
-        tasks = [loader.load_async() for loader in loaders]
+        # Use semaphore to control concurrency
+        semaphore = asyncio.Semaphore(max_concurrent)
+
+        async def process_with_semaphore(loader: "MistralLoader") -> List[Document]:
+            async with semaphore:
+                return await loader.load_async()
+
+        # Process all files with controlled concurrency
+        tasks = [process_with_semaphore(loader) for loader in loaders]
         results = await asyncio.gather(*tasks, return_exceptions=True)
 
         # Handle any exceptions in results
@@ -624,10 +750,18 @@ class MistralLoader:
             else:
                 processed_results.append(result)
 
+        # MONITORING: Log comprehensive batch processing statistics
         total_time = time.time() - start_time
         total_docs = sum(len(docs) for docs in processed_results)
+        success_count = sum(
+            1 for result in results if not isinstance(result, Exception)
+        )
+        failure_count = len(results) - success_count
+
         log.info(
-            f"Batch processing completed in {total_time:.2f}s, produced {total_docs} total documents"
+            f"Batch processing completed in {total_time:.2f}s: "
+            f"{success_count} files succeeded, {failure_count} files failed, "
+            f"produced {total_docs} total documents"
         )
 
         return processed_results

backend/open_webui/retrieval/loaders/youtube.py

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

backend/open_webui/retrieval/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
@@ -400,12 +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,
@@ -414,6 +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):
@@ -697,6 +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],
@@ -761,38 +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
 
 

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,
@@ -28,7 +32,12 @@ from open_webui.retrieval.vector.main import (
     SearchResult,
     GetResult,
 )
-from open_webui.config import PGVECTOR_DB_URL, PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH
+from open_webui.config import (
+    PGVECTOR_DB_URL,
+    PGVECTOR_INITIALIZE_MAX_VECTOR_LENGTH,
+    PGVECTOR_PGCRYPTO,
+    PGVECTOR_PGCRYPTO_KEY,
+)
 
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -39,14 +48,27 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
 
 
+def pgcrypto_encrypt(val, key):
+    return func.pgp_sym_encrypt(val, literal(key))
+
+
+def pgcrypto_decrypt(col, key, outtype="text"):
+    return func.cast(func.pgp_sym_decrypt(col, literal(key)), outtype)
+
+
 class DocumentChunk(Base):
     __tablename__ = "document_chunk"
 
     id = Column(Text, primary_key=True)
     vector = Column(Vector(dim=VECTOR_LENGTH), nullable=True)
     collection_name = Column(Text, nullable=False)
-    text = Column(Text, nullable=True)
-    vmetadata = Column(MutableDict.as_mutable(JSONB), nullable=True)
+
+    if PGVECTOR_PGCRYPTO:
+        text = Column(LargeBinary, nullable=True)
+        vmetadata = Column(LargeBinary, nullable=True)
+    else:
+        text = Column(Text, nullable=True)
+        vmetadata = Column(MutableDict.as_mutable(JSONB), nullable=True)
 
 
 class PgvectorClient(VectorDBBase):
@@ -70,6 +92,15 @@ class PgvectorClient(VectorDBBase):
             # Ensure the pgvector extension is available
             self.session.execute(text("CREATE EXTENSION IF NOT EXISTS vector;"))
 
+            if PGVECTOR_PGCRYPTO:
+                # Ensure the pgcrypto extension is available for encryption
+                self.session.execute(text("CREATE EXTENSION IF NOT EXISTS pgcrypto;"))
+
+                if not PGVECTOR_PGCRYPTO_KEY:
+                    raise ValueError(
+                        "PGVECTOR_PGCRYPTO_KEY must be set when PGVECTOR_PGCRYPTO is enabled."
+                    )
+
             # Check vector length consistency
             self.check_vector_length()
 
@@ -147,44 +178,39 @@ class PgvectorClient(VectorDBBase):
 
     def insert(self, collection_name: str, items: List[VectorItem]) -> None:
         try:
-            new_items = []
-            for item in items:
-                vector = self.adjust_vector_length(item["vector"])
-                new_chunk = DocumentChunk(
-                    id=item["id"],
-                    vector=vector,
-                    collection_name=collection_name,
-                    text=item["text"],
-                    vmetadata=item["metadata"],
-                )
-                new_items.append(new_chunk)
-            self.session.bulk_save_objects(new_items)
-            self.session.commit()
-            log.info(
-                f"Inserted {len(new_items)} items into collection '{collection_name}'."
-            )
-        except Exception as e:
-            self.session.rollback()
-            log.exception(f"Error during insert: {e}")
-            raise
-
-    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
-        try:
-            for item in items:
-                vector = self.adjust_vector_length(item["vector"])
-                existing = (
-                    self.session.query(DocumentChunk)
-                    .filter(DocumentChunk.id == item["id"])
-                    .first()
-                )
-                if existing:
-                    existing.vector = vector
-                    existing.text = item["text"]
-                    existing.vmetadata = item["metadata"]
-                    existing.collection_name = (
-                        collection_name  # Update collection_name if necessary
+            if PGVECTOR_PGCRYPTO:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    # Use raw SQL for BYTEA/pgcrypto
+                    self.session.execute(
+                        text(
+                            """
+                            INSERT INTO document_chunk
+                            (id, vector, collection_name, text, vmetadata)
+                            VALUES (
+                                :id, :vector, :collection_name,
+                                pgp_sym_encrypt(:text, :key),
+                                pgp_sym_encrypt(:metadata::text, :key)
+                            )
+                            ON CONFLICT (id) DO NOTHING
+                        """
+                        ),
+                        {
+                            "id": item["id"],
+                            "vector": vector,
+                            "collection_name": collection_name,
+                            "text": item["text"],
+                            "metadata": json.dumps(item["metadata"]),
+                            "key": PGVECTOR_PGCRYPTO_KEY,
+                        },
                     )
-                else:
+                self.session.commit()
+                log.info(f"Encrypted & inserted {len(items)} into '{collection_name}'")
+
+            else:
+                new_items = []
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
                     new_chunk = DocumentChunk(
                         id=item["id"],
                         vector=vector,
@@ -192,11 +218,78 @@ class PgvectorClient(VectorDBBase):
                         text=item["text"],
                         vmetadata=item["metadata"],
                     )
-                    self.session.add(new_chunk)
-            self.session.commit()
-            log.info(
-                f"Upserted {len(items)} items into collection '{collection_name}'."
-            )
+                    new_items.append(new_chunk)
+                self.session.bulk_save_objects(new_items)
+                self.session.commit()
+                log.info(
+                    f"Inserted {len(new_items)} items into collection '{collection_name}'."
+                )
+        except Exception as e:
+            self.session.rollback()
+            log.exception(f"Error during insert: {e}")
+            raise
+
+    def upsert(self, collection_name: str, items: List[VectorItem]) -> None:
+        try:
+            if PGVECTOR_PGCRYPTO:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    self.session.execute(
+                        text(
+                            """
+                            INSERT INTO document_chunk
+                            (id, vector, collection_name, text, vmetadata)
+                            VALUES (
+                                :id, :vector, :collection_name,
+                                pgp_sym_encrypt(:text, :key),
+                                pgp_sym_encrypt(:metadata::text, :key)
+                            )
+                            ON CONFLICT (id) DO UPDATE SET
+                              vector = EXCLUDED.vector,
+                              collection_name = EXCLUDED.collection_name,
+                              text = EXCLUDED.text,
+                              vmetadata = EXCLUDED.vmetadata
+                        """
+                        ),
+                        {
+                            "id": item["id"],
+                            "vector": vector,
+                            "collection_name": collection_name,
+                            "text": item["text"],
+                            "metadata": json.dumps(item["metadata"]),
+                            "key": PGVECTOR_PGCRYPTO_KEY,
+                        },
+                    )
+                self.session.commit()
+                log.info(f"Encrypted & upserted {len(items)} into '{collection_name}'")
+            else:
+                for item in items:
+                    vector = self.adjust_vector_length(item["vector"])
+                    existing = (
+                        self.session.query(DocumentChunk)
+                        .filter(DocumentChunk.id == item["id"])
+                        .first()
+                    )
+                    if existing:
+                        existing.vector = vector
+                        existing.text = item["text"]
+                        existing.vmetadata = item["metadata"]
+                        existing.collection_name = (
+                            collection_name  # Update collection_name if necessary
+                        )
+                    else:
+                        new_chunk = DocumentChunk(
+                            id=item["id"],
+                            vector=vector,
+                            collection_name=collection_name,
+                            text=item["text"],
+                            vmetadata=item["metadata"],
+                        )
+                        self.session.add(new_chunk)
+                self.session.commit()
+                log.info(
+                    f"Upserted {len(items)} items into collection '{collection_name}'."
+                )
         except Exception as e:
             self.session.rollback()
             log.exception(f"Error during upsert: {e}")
@@ -230,16 +323,32 @@ class PgvectorClient(VectorDBBase):
                 .alias("query_vectors")
             )
 
+            result_fields = [
+                DocumentChunk.id,
+            ]
+            if PGVECTOR_PGCRYPTO:
+                result_fields.append(
+                    pgcrypto_decrypt(
+                        DocumentChunk.text, PGVECTOR_PGCRYPTO_KEY, Text
+                    ).label("text")
+                )
+                result_fields.append(
+                    pgcrypto_decrypt(
+                        DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                    ).label("vmetadata")
+                )
+            else:
+                result_fields.append(DocumentChunk.text)
+                result_fields.append(DocumentChunk.vmetadata)
+            result_fields.append(
+                (DocumentChunk.vector.cosine_distance(query_vectors.c.q_vector)).label(
+                    "distance"
+                )
+            )
+
             # Build the lateral subquery for each query vector
             subq = (
-                select(
-                    DocumentChunk.id,
-                    DocumentChunk.text,
-                    DocumentChunk.vmetadata,
-                    (
-                        DocumentChunk.vector.cosine_distance(query_vectors.c.q_vector)
-                    ).label("distance"),
-                )
+                select(*result_fields)
                 .where(DocumentChunk.collection_name == collection_name)
                 .order_by(
                     (DocumentChunk.vector.cosine_distance(query_vectors.c.q_vector))
@@ -299,17 +408,43 @@ class PgvectorClient(VectorDBBase):
         self, collection_name: str, filter: Dict[str, Any], limit: Optional[int] = None
     ) -> Optional[GetResult]:
         try:
-            query = self.session.query(DocumentChunk).filter(
-                DocumentChunk.collection_name == collection_name
-            )
+            if PGVECTOR_PGCRYPTO:
+                # Build where clause for vmetadata filter
+                where_clauses = [DocumentChunk.collection_name == collection_name]
+                for key, value in filter.items():
+                    # decrypt then check key: JSON filter after decryption
+                    where_clauses.append(
+                        pgcrypto_decrypt(
+                            DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                        )[key].astext
+                        == str(value)
+                    )
+                stmt = select(
+                    DocumentChunk.id,
+                    pgcrypto_decrypt(
+                        DocumentChunk.text, PGVECTOR_PGCRYPTO_KEY, Text
+                    ).label("text"),
+                    pgcrypto_decrypt(
+                        DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                    ).label("vmetadata"),
+                ).where(*where_clauses)
+                if limit is not None:
+                    stmt = stmt.limit(limit)
+                results = self.session.execute(stmt).all()
+            else:
+                query = self.session.query(DocumentChunk).filter(
+                    DocumentChunk.collection_name == collection_name
+                )
 
-            for key, value in filter.items():
-                query = query.filter(DocumentChunk.vmetadata[key].astext == str(value))
+                for key, value in filter.items():
+                    query = query.filter(
+                        DocumentChunk.vmetadata[key].astext == str(value)
+                    )
 
-            if limit is not None:
-                query = query.limit(limit)
+                if limit is not None:
+                    query = query.limit(limit)
 
-            results = query.all()
+                results = query.all()
 
             if not results:
                 return None
@@ -331,20 +466,38 @@ class PgvectorClient(VectorDBBase):
         self, collection_name: str, limit: Optional[int] = None
     ) -> Optional[GetResult]:
         try:
-            query = self.session.query(DocumentChunk).filter(
-                DocumentChunk.collection_name == collection_name
-            )
-            if limit is not None:
-                query = query.limit(limit)
+            if PGVECTOR_PGCRYPTO:
+                stmt = select(
+                    DocumentChunk.id,
+                    pgcrypto_decrypt(
+                        DocumentChunk.text, PGVECTOR_PGCRYPTO_KEY, Text
+                    ).label("text"),
+                    pgcrypto_decrypt(
+                        DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                    ).label("vmetadata"),
+                ).where(DocumentChunk.collection_name == collection_name)
+                if limit is not None:
+                    stmt = stmt.limit(limit)
+                results = self.session.execute(stmt).all()
+                ids = [[row.id for row in results]]
+                documents = [[row.text for row in results]]
+                metadatas = [[row.vmetadata for row in results]]
+            else:
 
-            results = query.all()
+                query = self.session.query(DocumentChunk).filter(
+                    DocumentChunk.collection_name == collection_name
+                )
+                if limit is not None:
+                    query = query.limit(limit)
 
-            if not results:
-                return None
+                results = query.all()
 
-            ids = [[result.id for result in results]]
-            documents = [[result.text for result in results]]
-            metadatas = [[result.vmetadata for result in results]]
+                if not results:
+                    return None
+
+                ids = [[result.id for result in results]]
+                documents = [[result.text for result in results]]
+                metadatas = [[result.vmetadata for result in results]]
 
             return GetResult(ids=ids, documents=documents, metadatas=metadatas)
         except Exception as e:
@@ -358,17 +511,33 @@ class PgvectorClient(VectorDBBase):
         filter: Optional[Dict[str, Any]] = None,
     ) -> None:
         try:
-            query = self.session.query(DocumentChunk).filter(
-                DocumentChunk.collection_name == collection_name
-            )
-            if ids:
-                query = query.filter(DocumentChunk.id.in_(ids))
-            if filter:
-                for key, value in filter.items():
-                    query = query.filter(
-                        DocumentChunk.vmetadata[key].astext == str(value)
-                    )
-            deleted = query.delete(synchronize_session=False)
+            if PGVECTOR_PGCRYPTO:
+                wheres = [DocumentChunk.collection_name == collection_name]
+                if ids:
+                    wheres.append(DocumentChunk.id.in_(ids))
+                if filter:
+                    for key, value in filter.items():
+                        wheres.append(
+                            pgcrypto_decrypt(
+                                DocumentChunk.vmetadata, PGVECTOR_PGCRYPTO_KEY, JSONB
+                            )[key].astext
+                            == str(value)
+                        )
+                stmt = DocumentChunk.__table__.delete().where(*wheres)
+                result = self.session.execute(stmt)
+                deleted = result.rowcount
+            else:
+                query = self.session.query(DocumentChunk).filter(
+                    DocumentChunk.collection_name == collection_name
+                )
+                if ids:
+                    query = query.filter(DocumentChunk.id.in_(ids))
+                if filter:
+                    for key, value in filter.items():
+                        query = query.filter(
+                            DocumentChunk.vmetadata[key].astext == str(value)
+                        )
+                deleted = query.delete(synchronize_session=False)
             self.session.commit()
             log.info(f"Deleted {deleted} items from collection '{collection_name}'.")
         except Exception as e:

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

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

backend/open_webui/retrieval/web/perplexity.py

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

backend/open_webui/routers/audio.py

@@ -96,12 +96,9 @@ def is_audio_conversion_required(file_path):
             # File is AAC/mp4a audio, recommend mp3 conversion
             return True
 
-        # If the codec name or file extension is in the supported formats
-        if (
-            codec_name in SUPPORTED_FORMATS
-            or os.path.splitext(file_path)[1][1:].lower() in SUPPORTED_FORMATS
-        ):
-            return False  # Already supported
+        # If the codec name is in the supported formats
+        if codec_name in SUPPORTED_FORMATS:
+            return False
 
         return True
     except Exception as e:

backend/open_webui/routers/functions.py

@@ -12,7 +12,11 @@ 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
@@ -358,8 +362,9 @@ async def get_function_valves_spec_by_id(
 ):
     function = Functions.get_function_by_id(id)
     if function:
-        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
@@ -383,8 +388,9 @@ async def update_function_valves_by_id(
 ):
     function = Functions.get_function_by_id(id)
     if function:
-        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
@@ -443,8 +449,9 @@ async def get_function_user_valves_spec_by_id(
 ):
     function = Functions.get_function_by_id(id)
     if function:
-        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
@@ -464,8 +471,9 @@ async def update_function_user_valves_by_id(
     function = Functions.get_function_by_id(id)
 
     if function:
-        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

@@ -420,7 +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"
@@ -428,7 +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):
@@ -561,7 +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
@@ -612,9 +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})
@@ -665,9 +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/notes.py

@@ -124,9 +124,8 @@ async def get_note_by_id(request: Request, id: str, user=Depends(get_verified_us
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if (
-        user.role != "admin"
-        and user.id != note.user_id
+    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(
@@ -159,9 +158,8 @@ async def update_note_by_id(
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if (
-        user.role != "admin"
-        and user.id != note.user_id
+    if user.role != "admin" or (
+        user.id != note.user_id
         and not has_access(user.id, type="write", access_control=note.access_control)
     ):
         raise HTTPException(
@@ -199,9 +197,8 @@ async def delete_note_by_id(request: Request, id: str, user=Depends(get_verified
             status_code=status.HTTP_404_NOT_FOUND, detail=ERROR_MESSAGES.NOT_FOUND
         )
 
-    if (
-        user.role != "admin"
-        and user.id != note.user_id
+    if user.role != "admin" or (
+        user.id != note.user_id
         and not has_access(user.id, type="write", access_control=note.access_control)
     ):
         raise HTTPException(

backend/open_webui/routers/ollama.py

@@ -1232,6 +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):
@@ -1269,7 +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"]
 
@@ -1283,13 +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":
@@ -1322,7 +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),
@@ -1471,8 +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":

backend/open_webui/routers/openai.py

@@ -715,8 +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":
@@ -883,6 +887,88 @@ async def generate_chat_completion(
             await session.close()
 
 
+async def embeddings(request: Request, form_data: dict, user):
+    """
+    Calls the embeddings endpoint for OpenAI-compatible providers.
+
+    Args:
+        request (Request): The FastAPI request context.
+        form_data (dict): OpenAI-compatible embeddings payload.
+        user (UserModel): The authenticated user.
+
+    Returns:
+        dict: OpenAI-compatible embeddings response.
+    """
+    idx = 0
+    # Prepare payload/body
+    body = json.dumps(form_data)
+    # Find correct backend url/key based on model
+    await get_all_models(request, user=user)
+    model_id = form_data.get("model")
+    models = request.app.state.OPENAI_MODELS
+    if model_id in models:
+        idx = models[model_id]["urlIdx"]
+    url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
+    key = request.app.state.config.OPENAI_API_KEYS[idx]
+    r = None
+    session = None
+    streaming = False
+    try:
+        session = aiohttp.ClientSession(trust_env=True)
+        r = await session.request(
+            method="POST",
+            url=f"{url}/embeddings",
+            data=body,
+            headers={
+                "Authorization": f"Bearer {key}",
+                "Content-Type": "application/json",
+                **(
+                    {
+                        "X-OpenWebUI-User-Name": 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 {}
+                ),
+            },
+        )
+        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)):
     """

backend/open_webui/routers/retrieval.py

@@ -239,6 +239,11 @@ async def get_embedding_config(request: Request, user=Depends(get_admin_user)):
             "url": request.app.state.config.RAG_OLLAMA_BASE_URL,
             "key": request.app.state.config.RAG_OLLAMA_API_KEY,
         },
+        "azure_openai_config": {
+            "url": request.app.state.config.RAG_AZURE_OPENAI_BASE_URL,
+            "key": request.app.state.config.RAG_AZURE_OPENAI_API_KEY,
+            "version": request.app.state.config.RAG_AZURE_OPENAI_API_VERSION,
+        },
     }
 
 
@@ -252,9 +257,16 @@ class OllamaConfigForm(BaseModel):
     key: str
 
 
+class AzureOpenAIConfigForm(BaseModel):
+    url: str
+    key: str
+    version: str
+
+
 class EmbeddingModelUpdateForm(BaseModel):
     openai_config: Optional[OpenAIConfigForm] = None
     ollama_config: Optional[OllamaConfigForm] = None
+    azure_openai_config: Optional[AzureOpenAIConfigForm] = None
     embedding_engine: str
     embedding_model: str
     embedding_batch_size: Optional[int] = 1
@@ -271,7 +283,11 @@ async def update_embedding_config(
         request.app.state.config.RAG_EMBEDDING_ENGINE = form_data.embedding_engine
         request.app.state.config.RAG_EMBEDDING_MODEL = form_data.embedding_model
 
-        if request.app.state.config.RAG_EMBEDDING_ENGINE in ["ollama", "openai"]:
+        if request.app.state.config.RAG_EMBEDDING_ENGINE in [
+            "ollama",
+            "openai",
+            "azure_openai",
+        ]:
             if form_data.openai_config is not None:
                 request.app.state.config.RAG_OPENAI_API_BASE_URL = (
                     form_data.openai_config.url
@@ -288,6 +304,17 @@ async def update_embedding_config(
                     form_data.ollama_config.key
                 )
 
+            if form_data.azure_openai_config is not None:
+                request.app.state.config.RAG_AZURE_OPENAI_BASE_URL = (
+                    form_data.azure_openai_config.url
+                )
+                request.app.state.config.RAG_AZURE_OPENAI_API_KEY = (
+                    form_data.azure_openai_config.key
+                )
+                request.app.state.config.RAG_AZURE_OPENAI_API_VERSION = (
+                    form_data.azure_openai_config.version
+                )
+
             request.app.state.config.RAG_EMBEDDING_BATCH_SIZE = (
                 form_data.embedding_batch_size
             )
@@ -304,14 +331,27 @@ async def update_embedding_config(
             (
                 request.app.state.config.RAG_OPENAI_API_BASE_URL
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_BASE_URL
+                else (
+                    request.app.state.config.RAG_OLLAMA_BASE_URL
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_BASE_URL
+                )
             ),
             (
                 request.app.state.config.RAG_OPENAI_API_KEY
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_API_KEY
+                else (
+                    request.app.state.config.RAG_OLLAMA_API_KEY
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_API_KEY
+                )
             ),
             request.app.state.config.RAG_EMBEDDING_BATCH_SIZE,
+            azure_api_version=(
+                request.app.state.config.RAG_AZURE_OPENAI_API_VERSION
+                if request.app.state.config.RAG_EMBEDDING_ENGINE == "azure_openai"
+                else None
+            ),
         )
 
         return {
@@ -327,6 +367,11 @@ async def update_embedding_config(
                 "url": request.app.state.config.RAG_OLLAMA_BASE_URL,
                 "key": request.app.state.config.RAG_OLLAMA_API_KEY,
             },
+            "azure_openai_config": {
+                "url": request.app.state.config.RAG_AZURE_OPENAI_BASE_URL,
+                "key": request.app.state.config.RAG_AZURE_OPENAI_API_KEY,
+                "version": request.app.state.config.RAG_AZURE_OPENAI_API_VERSION,
+            },
         }
     except Exception as e:
         log.exception(f"Problem updating embedding model: {e}")
@@ -353,6 +398,15 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         # Content extraction settings
         "CONTENT_EXTRACTION_ENGINE": request.app.state.config.CONTENT_EXTRACTION_ENGINE,
         "PDF_EXTRACT_IMAGES": request.app.state.config.PDF_EXTRACT_IMAGES,
+        "DATALAB_MARKER_API_KEY": request.app.state.config.DATALAB_MARKER_API_KEY,
+        "DATALAB_MARKER_LANGS": request.app.state.config.DATALAB_MARKER_LANGS,
+        "DATALAB_MARKER_SKIP_CACHE": request.app.state.config.DATALAB_MARKER_SKIP_CACHE,
+        "DATALAB_MARKER_FORCE_OCR": request.app.state.config.DATALAB_MARKER_FORCE_OCR,
+        "DATALAB_MARKER_PAGINATE": request.app.state.config.DATALAB_MARKER_PAGINATE,
+        "DATALAB_MARKER_STRIP_EXISTING_OCR": request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR,
+        "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION": request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+        "DATALAB_MARKER_USE_LLM": request.app.state.config.DATALAB_MARKER_USE_LLM,
+        "DATALAB_MARKER_OUTPUT_FORMAT": request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT,
         "EXTERNAL_DOCUMENT_LOADER_URL": request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL,
         "EXTERNAL_DOCUMENT_LOADER_API_KEY": request.app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY,
         "TIKA_SERVER_URL": request.app.state.config.TIKA_SERVER_URL,
@@ -360,6 +414,9 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
         "DOCLING_OCR_ENGINE": request.app.state.config.DOCLING_OCR_ENGINE,
         "DOCLING_OCR_LANG": request.app.state.config.DOCLING_OCR_LANG,
         "DOCLING_DO_PICTURE_DESCRIPTION": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+        "DOCLING_PICTURE_DESCRIPTION_MODE": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL,
+        "DOCLING_PICTURE_DESCRIPTION_API": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API,
         "DOCUMENT_INTELLIGENCE_ENDPOINT": request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
         "DOCUMENT_INTELLIGENCE_KEY": request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
         "MISTRAL_OCR_API_KEY": request.app.state.config.MISTRAL_OCR_API_KEY,
@@ -413,6 +470,8 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
             "BING_SEARCH_V7_SUBSCRIPTION_KEY": request.app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY,
             "EXA_API_KEY": request.app.state.config.EXA_API_KEY,
             "PERPLEXITY_API_KEY": request.app.state.config.PERPLEXITY_API_KEY,
+            "PERPLEXITY_MODEL": request.app.state.config.PERPLEXITY_MODEL,
+            "PERPLEXITY_SEARCH_CONTEXT_USAGE": request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE,
             "SOUGOU_API_SID": request.app.state.config.SOUGOU_API_SID,
             "SOUGOU_API_SK": request.app.state.config.SOUGOU_API_SK,
             "WEB_LOADER_ENGINE": request.app.state.config.WEB_LOADER_ENGINE,
@@ -466,6 +525,8 @@ class WebConfig(BaseModel):
     BING_SEARCH_V7_SUBSCRIPTION_KEY: Optional[str] = None
     EXA_API_KEY: Optional[str] = None
     PERPLEXITY_API_KEY: Optional[str] = None
+    PERPLEXITY_MODEL: Optional[str] = None
+    PERPLEXITY_SEARCH_CONTEXT_USAGE: Optional[str] = None
     SOUGOU_API_SID: Optional[str] = None
     SOUGOU_API_SK: Optional[str] = None
     WEB_LOADER_ENGINE: Optional[str] = None
@@ -500,6 +561,15 @@ class ConfigForm(BaseModel):
     # Content extraction settings
     CONTENT_EXTRACTION_ENGINE: Optional[str] = None
     PDF_EXTRACT_IMAGES: Optional[bool] = None
+    DATALAB_MARKER_API_KEY: Optional[str] = None
+    DATALAB_MARKER_LANGS: Optional[str] = None
+    DATALAB_MARKER_SKIP_CACHE: Optional[bool] = None
+    DATALAB_MARKER_FORCE_OCR: Optional[bool] = None
+    DATALAB_MARKER_PAGINATE: Optional[bool] = None
+    DATALAB_MARKER_STRIP_EXISTING_OCR: Optional[bool] = None
+    DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION: Optional[bool] = None
+    DATALAB_MARKER_USE_LLM: Optional[bool] = None
+    DATALAB_MARKER_OUTPUT_FORMAT: Optional[str] = None
     EXTERNAL_DOCUMENT_LOADER_URL: Optional[str] = None
     EXTERNAL_DOCUMENT_LOADER_API_KEY: Optional[str] = None
 
@@ -508,6 +578,9 @@ class ConfigForm(BaseModel):
     DOCLING_OCR_ENGINE: Optional[str] = None
     DOCLING_OCR_LANG: Optional[str] = None
     DOCLING_DO_PICTURE_DESCRIPTION: Optional[bool] = None
+    DOCLING_PICTURE_DESCRIPTION_MODE: Optional[str] = None
+    DOCLING_PICTURE_DESCRIPTION_LOCAL: Optional[dict] = None
+    DOCLING_PICTURE_DESCRIPTION_API: Optional[dict] = None
     DOCUMENT_INTELLIGENCE_ENDPOINT: Optional[str] = None
     DOCUMENT_INTELLIGENCE_KEY: Optional[str] = None
     MISTRAL_OCR_API_KEY: Optional[str] = None
@@ -599,6 +672,51 @@ async def update_rag_config(
         if form_data.PDF_EXTRACT_IMAGES is not None
         else request.app.state.config.PDF_EXTRACT_IMAGES
     )
+    request.app.state.config.DATALAB_MARKER_API_KEY = (
+        form_data.DATALAB_MARKER_API_KEY
+        if form_data.DATALAB_MARKER_API_KEY is not None
+        else request.app.state.config.DATALAB_MARKER_API_KEY
+    )
+    request.app.state.config.DATALAB_MARKER_LANGS = (
+        form_data.DATALAB_MARKER_LANGS
+        if form_data.DATALAB_MARKER_LANGS is not None
+        else request.app.state.config.DATALAB_MARKER_LANGS
+    )
+    request.app.state.config.DATALAB_MARKER_SKIP_CACHE = (
+        form_data.DATALAB_MARKER_SKIP_CACHE
+        if form_data.DATALAB_MARKER_SKIP_CACHE is not None
+        else request.app.state.config.DATALAB_MARKER_SKIP_CACHE
+    )
+    request.app.state.config.DATALAB_MARKER_FORCE_OCR = (
+        form_data.DATALAB_MARKER_FORCE_OCR
+        if form_data.DATALAB_MARKER_FORCE_OCR is not None
+        else request.app.state.config.DATALAB_MARKER_FORCE_OCR
+    )
+    request.app.state.config.DATALAB_MARKER_PAGINATE = (
+        form_data.DATALAB_MARKER_PAGINATE
+        if form_data.DATALAB_MARKER_PAGINATE is not None
+        else request.app.state.config.DATALAB_MARKER_PAGINATE
+    )
+    request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR = (
+        form_data.DATALAB_MARKER_STRIP_EXISTING_OCR
+        if form_data.DATALAB_MARKER_STRIP_EXISTING_OCR is not None
+        else request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR
+    )
+    request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION = (
+        form_data.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
+        if form_data.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION is not None
+        else request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION
+    )
+    request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT = (
+        form_data.DATALAB_MARKER_OUTPUT_FORMAT
+        if form_data.DATALAB_MARKER_OUTPUT_FORMAT is not None
+        else request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT
+    )
+    request.app.state.config.DATALAB_MARKER_USE_LLM = (
+        form_data.DATALAB_MARKER_USE_LLM
+        if form_data.DATALAB_MARKER_USE_LLM is not None
+        else request.app.state.config.DATALAB_MARKER_USE_LLM
+    )
     request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL = (
         form_data.EXTERNAL_DOCUMENT_LOADER_URL
         if form_data.EXTERNAL_DOCUMENT_LOADER_URL is not None
@@ -636,6 +754,22 @@ async def update_rag_config(
         else request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION
     )
 
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_MODE
+        if form_data.DOCLING_PICTURE_DESCRIPTION_MODE is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL
+        if form_data.DOCLING_PICTURE_DESCRIPTION_LOCAL is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL
+    )
+    request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API = (
+        form_data.DOCLING_PICTURE_DESCRIPTION_API
+        if form_data.DOCLING_PICTURE_DESCRIPTION_API is not None
+        else request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API
+    )
+
     request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT = (
         form_data.DOCUMENT_INTELLIGENCE_ENDPOINT
         if form_data.DOCUMENT_INTELLIGENCE_ENDPOINT is not None
@@ -799,6 +933,10 @@ async def update_rag_config(
         )
         request.app.state.config.EXA_API_KEY = form_data.web.EXA_API_KEY
         request.app.state.config.PERPLEXITY_API_KEY = form_data.web.PERPLEXITY_API_KEY
+        request.app.state.config.PERPLEXITY_MODEL = form_data.web.PERPLEXITY_MODEL
+        request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE = (
+            form_data.web.PERPLEXITY_SEARCH_CONTEXT_USAGE
+        )
         request.app.state.config.SOUGOU_API_SID = form_data.web.SOUGOU_API_SID
         request.app.state.config.SOUGOU_API_SK = form_data.web.SOUGOU_API_SK
 
@@ -853,6 +991,15 @@ async def update_rag_config(
         # Content extraction settings
         "CONTENT_EXTRACTION_ENGINE": request.app.state.config.CONTENT_EXTRACTION_ENGINE,
         "PDF_EXTRACT_IMAGES": request.app.state.config.PDF_EXTRACT_IMAGES,
+        "DATALAB_MARKER_API_KEY": request.app.state.config.DATALAB_MARKER_API_KEY,
+        "DATALAB_MARKER_LANGS": request.app.state.config.DATALAB_MARKER_LANGS,
+        "DATALAB_MARKER_SKIP_CACHE": request.app.state.config.DATALAB_MARKER_SKIP_CACHE,
+        "DATALAB_MARKER_FORCE_OCR": request.app.state.config.DATALAB_MARKER_FORCE_OCR,
+        "DATALAB_MARKER_PAGINATE": request.app.state.config.DATALAB_MARKER_PAGINATE,
+        "DATALAB_MARKER_STRIP_EXISTING_OCR": request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR,
+        "DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION": request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+        "DATALAB_MARKER_USE_LLM": request.app.state.config.DATALAB_MARKER_USE_LLM,
+        "DATALAB_MARKER_OUTPUT_FORMAT": request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT,
         "EXTERNAL_DOCUMENT_LOADER_URL": request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL,
         "EXTERNAL_DOCUMENT_LOADER_API_KEY": request.app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY,
         "TIKA_SERVER_URL": request.app.state.config.TIKA_SERVER_URL,
@@ -860,6 +1007,9 @@ async def update_rag_config(
         "DOCLING_OCR_ENGINE": request.app.state.config.DOCLING_OCR_ENGINE,
         "DOCLING_OCR_LANG": request.app.state.config.DOCLING_OCR_LANG,
         "DOCLING_DO_PICTURE_DESCRIPTION": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+        "DOCLING_PICTURE_DESCRIPTION_MODE": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+        "DOCLING_PICTURE_DESCRIPTION_LOCAL": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL,
+        "DOCLING_PICTURE_DESCRIPTION_API": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API,
         "DOCUMENT_INTELLIGENCE_ENDPOINT": request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
         "DOCUMENT_INTELLIGENCE_KEY": request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
         "MISTRAL_OCR_API_KEY": request.app.state.config.MISTRAL_OCR_API_KEY,
@@ -913,6 +1063,8 @@ async def update_rag_config(
             "BING_SEARCH_V7_SUBSCRIPTION_KEY": request.app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY,
             "EXA_API_KEY": request.app.state.config.EXA_API_KEY,
             "PERPLEXITY_API_KEY": request.app.state.config.PERPLEXITY_API_KEY,
+            "PERPLEXITY_MODEL": request.app.state.config.PERPLEXITY_MODEL,
+            "PERPLEXITY_SEARCH_CONTEXT_USAGE": request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE,
             "SOUGOU_API_SID": request.app.state.config.SOUGOU_API_SID,
             "SOUGOU_API_SK": request.app.state.config.SOUGOU_API_SK,
             "WEB_LOADER_ENGINE": request.app.state.config.WEB_LOADER_ENGINE,
@@ -1057,14 +1209,27 @@ def save_docs_to_vector_db(
             (
                 request.app.state.config.RAG_OPENAI_API_BASE_URL
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_BASE_URL
+                else (
+                    request.app.state.config.RAG_OLLAMA_BASE_URL
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_BASE_URL
+                )
             ),
             (
                 request.app.state.config.RAG_OPENAI_API_KEY
                 if request.app.state.config.RAG_EMBEDDING_ENGINE == "openai"
-                else request.app.state.config.RAG_OLLAMA_API_KEY
+                else (
+                    request.app.state.config.RAG_OLLAMA_API_KEY
+                    if request.app.state.config.RAG_EMBEDDING_ENGINE == "ollama"
+                    else request.app.state.config.RAG_AZURE_OPENAI_API_KEY
+                )
             ),
             request.app.state.config.RAG_EMBEDDING_BATCH_SIZE,
+            azure_api_version=(
+                request.app.state.config.RAG_AZURE_OPENAI_API_VERSION
+                if request.app.state.config.RAG_EMBEDDING_ENGINE == "azure_openai"
+                else None
+            ),
         )
 
         embeddings = embedding_function(
@@ -1178,13 +1343,27 @@ def process_file(
                 file_path = Storage.get_file(file_path)
                 loader = Loader(
                     engine=request.app.state.config.CONTENT_EXTRACTION_ENGINE,
+                    DATALAB_MARKER_API_KEY=request.app.state.config.DATALAB_MARKER_API_KEY,
+                    DATALAB_MARKER_LANGS=request.app.state.config.DATALAB_MARKER_LANGS,
+                    DATALAB_MARKER_SKIP_CACHE=request.app.state.config.DATALAB_MARKER_SKIP_CACHE,
+                    DATALAB_MARKER_FORCE_OCR=request.app.state.config.DATALAB_MARKER_FORCE_OCR,
+                    DATALAB_MARKER_PAGINATE=request.app.state.config.DATALAB_MARKER_PAGINATE,
+                    DATALAB_MARKER_STRIP_EXISTING_OCR=request.app.state.config.DATALAB_MARKER_STRIP_EXISTING_OCR,
+                    DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION=request.app.state.config.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION,
+                    DATALAB_MARKER_USE_LLM=request.app.state.config.DATALAB_MARKER_USE_LLM,
+                    DATALAB_MARKER_OUTPUT_FORMAT=request.app.state.config.DATALAB_MARKER_OUTPUT_FORMAT,
                     EXTERNAL_DOCUMENT_LOADER_URL=request.app.state.config.EXTERNAL_DOCUMENT_LOADER_URL,
                     EXTERNAL_DOCUMENT_LOADER_API_KEY=request.app.state.config.EXTERNAL_DOCUMENT_LOADER_API_KEY,
                     TIKA_SERVER_URL=request.app.state.config.TIKA_SERVER_URL,
                     DOCLING_SERVER_URL=request.app.state.config.DOCLING_SERVER_URL,
-                    DOCLING_OCR_ENGINE=request.app.state.config.DOCLING_OCR_ENGINE,
-                    DOCLING_OCR_LANG=request.app.state.config.DOCLING_OCR_LANG,
-                    DOCLING_DO_PICTURE_DESCRIPTION=request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+                    DOCLING_PARAMS={
+                        "ocr_engine": request.app.state.config.DOCLING_OCR_ENGINE,
+                        "ocr_lang": request.app.state.config.DOCLING_OCR_LANG,
+                        "do_picture_description": request.app.state.config.DOCLING_DO_PICTURE_DESCRIPTION,
+                        "picture_description_mode": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_MODE,
+                        "picture_description_local": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_LOCAL,
+                        "picture_description_api": request.app.state.config.DOCLING_PICTURE_DESCRIPTION_API,
+                    },
                     PDF_EXTRACT_IMAGES=request.app.state.config.PDF_EXTRACT_IMAGES,
                     DOCUMENT_INTELLIGENCE_ENDPOINT=request.app.state.config.DOCUMENT_INTELLIGENCE_ENDPOINT,
                     DOCUMENT_INTELLIGENCE_KEY=request.app.state.config.DOCUMENT_INTELLIGENCE_KEY,
@@ -1601,19 +1780,14 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
             request.app.state.config.WEB_SEARCH_RESULT_COUNT,
             request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
         )
-    elif engine == "exa":
-        return search_exa(
-            request.app.state.config.EXA_API_KEY,
-            query,
-            request.app.state.config.WEB_SEARCH_RESULT_COUNT,
-            request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
-        )
     elif engine == "perplexity":
         return search_perplexity(
             request.app.state.config.PERPLEXITY_API_KEY,
             query,
             request.app.state.config.WEB_SEARCH_RESULT_COUNT,
             request.app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST,
+            model=request.app.state.config.PERPLEXITY_MODEL,
+            search_context_usage=request.app.state.config.PERPLEXITY_SEARCH_CONTEXT_USAGE,
         )
     elif engine == "sougou":
         if (
@@ -1835,6 +2009,7 @@ class QueryCollectionsForm(BaseModel):
     k_reranker: Optional[int] = None
     r: Optional[float] = None
     hybrid: Optional[bool] = None
+    hybrid_bm25_weight: Optional[float] = None
 
 
 @router.post("/query/collection")

backend/open_webui/routers/tasks.py

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

backend/open_webui/routers/tools.py

@@ -2,6 +2,9 @@ import logging
 from pathlib import Path
 from typing import Optional
 import time
+import re
+import aiohttp
+from pydantic import BaseModel, HttpUrl
 
 from open_webui.models.tools import (
     ToolForm,
@@ -21,6 +24,7 @@ from open_webui.env import SRC_LOG_LEVELS
 
 from open_webui.utils.tools import get_tool_servers_data
 
+
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
@@ -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
 ############################

backend/open_webui/routers/users.py

@@ -165,22 +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
 ############################
@@ -333,11 +317,22 @@ async def update_user_by_id(
     # Prevent modification of the primary admin user by other admins
     try:
         first_user = Users.get_first_user()
-        if first_user and user_id == first_user.id and session_user.id != user_id:
-            raise HTTPException(
-                status_code=status.HTTP_403_FORBIDDEN,
-                detail=ERROR_MESSAGES.ACTION_PROHIBITED,
-            )
+        if first_user:
+            if user_id == first_user.id:
+                if session_user.id != user_id:
+                    # If the user trying to update is the primary admin, and they are not the primary admin themselves
+                    raise HTTPException(
+                        status_code=status.HTTP_403_FORBIDDEN,
+                        detail=ERROR_MESSAGES.ACTION_PROHIBITED,
+                    )
+
+                if form_data.role != "admin":
+                    # If the primary admin is trying to change their own role, prevent it
+                    raise HTTPException(
+                        status_code=status.HTTP_403_FORBIDDEN,
+                        detail=ERROR_MESSAGES.ACTION_PROHIBITED,
+                    )
+
     except Exception as e:
         log.error(f"Error checking primary admin status: {e}")
         raise HTTPException(
@@ -365,6 +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,

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/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/tasks.py

@@ -2,16 +2,87 @@
 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, id=None):
+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
@@ -21,7 +92,7 @@ def cleanup_task(task_id: str, id=None):
             chat_tasks.pop(id, None)
 
 
-def create_task(coroutine, id=None):
+async def create_task(request, coroutine, id=None):
     """
     Create a new asyncio task and add it to the global task dictionary.
     """
@@ -29,7 +100,9 @@ def create_task(coroutine, id=None):
     task = asyncio.create_task(coroutine)  # Create the task
 
     # Add a done callback for cleanup
-    task.add_done_callback(lambda t: cleanup_task(task_id, id))
+    task.add_done_callback(
+        lambda t: asyncio.create_task(cleanup_task(request, task_id, id))
+    )
     tasks[task_id] = task
 
     # If an ID is provided, associate the task with that ID
@@ -38,34 +111,46 @@ def create_task(coroutine, id=None):
     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())
 
 
-def list_task_ids_by_chat_id(id):
+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(task_id: str):
+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/auth.py

@@ -23,6 +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
@@ -157,6 +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),
 ):
@@ -225,6 +227,19 @@ 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:

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 (
@@ -317,12 +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,
@@ -392,8 +390,7 @@ async def chat_action(request: Request, action_id: str, form_data: dict, user: A
         }
     )
 
-    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/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
 
@@ -9,14 +12,13 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
 
-def get_function_module(request, function_id):
+def get_function_module(request, function_id, load_from_db=True):
     """
     Get the function module by its ID.
     """
-
-    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, load_from_db
+    )
     return function_module
 
 
@@ -66,7 +68,9 @@ async def process_filter_functions(
         if not filter:
             continue
 
-        function_module = get_function_module(request, filter_id)
+        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

@@ -32,11 +32,17 @@ 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,
@@ -299,13 +305,20 @@ async def chat_completion_tools_handler(
 async def chat_memory_handler(
     request: Request, form_data: dict, extra_params: dict, user
 ):
-    results = await query_memory(
-        request,
-        QueryMemoryForm(
-            **{"content": get_last_user_message(form_data["messages"]), "k": 3}
-        ),
-        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"):
@@ -658,38 +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 "presence_penalty" in params and params["presence_penalty"] is not None:
-            form_data["presence_penalty"] = params["presence_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:
@@ -703,7 +718,6 @@ def apply_params_to_form_data(form_data, model):
 
 
 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}")
 
@@ -713,12 +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,
@@ -1035,6 +1044,59 @@ async def process_chat_response(
                 )
 
             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(
@@ -1260,12 +1322,7 @@ 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,
@@ -1809,9 +1866,11 @@ async def process_chat_response(
 
                                     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
@@ -2202,28 +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)
@@ -2234,30 +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)
@@ -2366,8 +2410,8 @@ 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), id=metadata["chat_id"]
+        task_id, _ = await create_task(
+            request, post_response_handler(response, events), id=metadata["chat_id"]
         )
         return {"status": True, "task_id": task_id}
 

backend/open_webui/utils/misc.py

@@ -74,12 +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
 
 
@@ -208,6 +208,7 @@ def openai_chat_message_template(model: str):
 def openai_chat_chunk_message_template(
     model: str,
     content: Optional[str] = None,
+    reasoning_content: Optional[str] = None,
     tool_calls: Optional[list[dict]] = None,
     usage: Optional[dict] = None,
 ) -> dict:
@@ -220,6 +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
 
@@ -234,6 +238,7 @@ def openai_chat_chunk_message_template(
 def openai_chat_completion_message_template(
     model: str,
     message: Optional[str] = None,
+    reasoning_content: Optional[str] = None,
     tool_calls: Optional[list[dict]] = None,
     usage: Optional[dict] = None,
 ) -> dict:
@@ -241,8 +246,9 @@ def openai_chat_completion_message_template(
     template["object"] = "chat.completion"
     if message is not None:
         template["choices"][0]["message"] = {
-            "content": message,
             "role": "assistant",
+            "content": message,
+            **({"reasoning_content": reasoning_content} if reasoning_content else {}),
             **({"tool_calls": tool_calls} if tool_calls else {}),
         }
 

backend/open_webui/utils/models.py

@@ -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
 
 
@@ -30,35 +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,
-                "connection_type": model.get("connection_type", "local"),
-                "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):
@@ -239,8 +254,7 @@ async def get_all_models(request, user: UserModel = None):
         ]
 
     def get_function_module_by_id(function_id):
-        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:

backend/open_webui/utils/oauth.py

@@ -538,7 +538,7 @@ class OAuthManager:
         # Redirect back to the frontend with the JWT token
 
         redirect_base_url = request.app.state.config.WEBUI_URL or request.base_url
-        if redirect_base_url.endswith("/"):
+        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}"
 

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
 
@@ -45,15 +45,60 @@ 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,
@@ -71,6 +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",
@@ -113,16 +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]:
@@ -217,36 +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:
@@ -267,3 +357,32 @@ def convert_payload_openai_to_ollama(openai_payload: dict) -> dict:
             ollama_payload["format"] = format
 
     return ollama_payload
+
+
+def convert_embedding_payload_openai_to_ollama(openai_payload: dict) -> dict:
+    """
+    Convert an embeddings request payload from OpenAI format to Ollama format.
+
+    Args:
+        openai_payload (dict): The original payload designed for OpenAI API usage.
+
+    Returns:
+        dict: A payload compatible with the Ollama API embeddings endpoint.
+    """
+    ollama_payload = {"model": openai_payload.get("model")}
+    input_value = openai_payload.get("input")
+
+    # Ollama expects 'input' as a list, and 'prompt' as a single string.
+    if isinstance(input_value, list):
+        ollama_payload["input"] = input_value
+        ollama_payload["prompt"] = "\n".join(str(x) for x in input_value)
+    else:
+        ollama_payload["input"] = [input_value]
+        ollama_payload["prompt"] = str(input_value)
+
+    # Optionally forward other fields if present
+    for optional_key in ("options", "truncate", "keep_alive"):
+        if optional_key in openai_payload:
+            ollama_payload[optional_key] = openai_payload[optional_key]
+
+    return ollama_payload

backend/open_webui/utils/plugin.py

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

backend/open_webui/utils/redis.py

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

backend/open_webui/utils/response.py

@@ -83,6 +83,7 @@ def convert_ollama_usage_to_openai(data: dict) -> dict:
 def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
     model = ollama_response.get("model", "ollama")
     message_content = ollama_response.get("message", {}).get("content", "")
+    reasoning_content = ollama_response.get("message", {}).get("thinking", None)
     tool_calls = ollama_response.get("message", {}).get("tool_calls", None)
     openai_tool_calls = None
 
@@ -94,7 +95,7 @@ def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
     usage = convert_ollama_usage_to_openai(data)
 
     response = openai_chat_completion_message_template(
-        model, message_content, openai_tool_calls, usage
+        model, message_content, reasoning_content, openai_tool_calls, usage
     )
     return response
 
@@ -105,6 +106,7 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)
 
         model = data.get("model", "ollama")
         message_content = data.get("message", {}).get("content", None)
+        reasoning_content = data.get("message", {}).get("thinking", None)
         tool_calls = data.get("message", {}).get("tool_calls", None)
         openai_tool_calls = None
 
@@ -118,10 +120,71 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)
             usage = convert_ollama_usage_to_openai(data)
 
         data = openai_chat_chunk_message_template(
-            model, message_content, openai_tool_calls, usage
+            model, message_content, reasoning_content, openai_tool_calls, usage
         )
 
         line = f"data: {json.dumps(data)}\n\n"
         yield line
 
     yield "data: [DONE]\n\n"
+
+
+def convert_embedding_response_ollama_to_openai(response) -> dict:
+    """
+    Convert the response from Ollama embeddings endpoint to the OpenAI-compatible format.
+
+    Args:
+        response (dict): The response from the Ollama API,
+            e.g. {"embedding": [...], "model": "..."}
+            or {"embeddings": [{"embedding": [...], "index": 0}, ...], "model": "..."}
+
+    Returns:
+        dict: Response adapted to OpenAI's embeddings API format.
+            e.g. {
+                "object": "list",
+                "data": [
+                    {"object": "embedding", "embedding": [...], "index": 0},
+                    ...
+                ],
+                "model": "...",
+            }
+    """
+    # Ollama batch-style output
+    if isinstance(response, dict) and "embeddings" in response:
+        openai_data = []
+        for i, emb in enumerate(response["embeddings"]):
+            openai_data.append(
+                {
+                    "object": "embedding",
+                    "embedding": emb.get("embedding"),
+                    "index": emb.get("index", i),
+                }
+            )
+        return {
+            "object": "list",
+            "data": openai_data,
+            "model": response.get("model"),
+        }
+    # Ollama single output
+    elif isinstance(response, dict) and "embedding" in response:
+        return {
+            "object": "list",
+            "data": [
+                {
+                    "object": "embedding",
+                    "embedding": response["embedding"],
+                    "index": 0,
+                }
+            ],
+            "model": response.get("model"),
+        }
+    # Already OpenAI-compatible?
+    elif (
+        isinstance(response, dict)
+        and "data" in response
+        and isinstance(response["data"], list)
+    ):
+        return response
+
+    # Fallback: return as is if unrecognized
+    return response

backend/open_webui/utils/task.py

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

backend/requirements.txt

@@ -1,5 +1,5 @@
 fastapi==0.115.7
-uvicorn[standard]==0.34.0
+uvicorn[standard]==0.34.2
 pydantic==2.10.6
 python-multipart==0.0.20
 
@@ -7,14 +7,13 @@ 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==3.18.1
@@ -76,13 +75,13 @@ pandas==2.2.3
 openpyxl==3.1.5
 pyxlsb==1.0.10
 xlrd==2.0.1
-validators==0.34.0
+validators==0.35.0
 psutil
 sentencepiece
 soundfile==0.13.1
-azure-ai-documentintelligence==1.0.0
+azure-ai-documentintelligence==1.0.2
 
-pillow==11.1.0
+pillow==11.2.1
 opencv-python-headless==4.11.0.86
 rapidocr-onnxruntime==1.4.4
 rank-bm25==0.2.2

backend/start.sh

@@ -14,7 +14,11 @@ if [[ "${WEB_LOADER_ENGINE,,}" == "playwright" ]]; then
     python -c "import nltk; nltk.download('punkt_tab')"
 fi
 
-KEY_FILE=.webui_secret_key
+if [ -n "${WEBUI_SECRET_KEY_FILE}" ]; then
+    KEY_FILE="${WEBUI_SECRET_KEY_FILE}"
+else
+    KEY_FILE=".webui_secret_key"
+fi
 
 PORT="${PORT:-8080}"
 HOST="${HOST:-0.0.0.0}"

backend/start_windows.bat

@@ -18,6 +18,10 @@ IF /I "%WEB_LOADER_ENGINE%" == "playwright" (
 )
 
 SET "KEY_FILE=.webui_secret_key"
+IF NOT "%WEBUI_SECRET_KEY_FILE%" == "" (
+    SET "KEY_FILE=%WEBUI_SECRET_KEY_FILE%"
+)
+
 IF "%PORT%"=="" SET PORT=8080
 IF "%HOST%"=="" SET HOST=0.0.0.0
 SET "WEBUI_SECRET_KEY=%WEBUI_SECRET_KEY%"

package-lock.json

@@ -1,12 +1,12 @@
 {
 	"name": "open-webui",
-	"version": "0.6.11",
+	"version": "0.6.14",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "open-webui",
-			"version": "0.6.11",
+			"version": "0.6.14",
 			"dependencies": {
 				"@azure/msal-browser": "^4.5.0",
 				"@codemirror/lang-javascript": "^6.2.2",
@@ -31,7 +31,7 @@
 				"@tiptap/starter-kit": "^2.10.0",
 				"@xyflow/svelte": "^0.1.19",
 				"async": "^3.2.5",
-				"bits-ui": "^0.19.7",
+				"bits-ui": "^0.21.15",
 				"codemirror": "^6.0.1",
 				"codemirror-lang-elixir": "^4.0.0",
 				"codemirror-lang-hcl": "^0.1.0",
@@ -51,7 +51,7 @@
 				"idb": "^7.1.1",
 				"js-sha256": "^0.10.1",
 				"jspdf": "^3.0.0",
-				"katex": "^0.16.21",
+				"katex": "^0.16.22",
 				"kokoro-js": "^1.1.1",
 				"marked": "^9.1.0",
 				"mermaid": "^11.6.0",
@@ -1201,26 +1201,29 @@
 			}
 		},
 		"node_modules/@floating-ui/core": {
-			"version": "1.6.0",
-			"resolved": "https://registry.npmjs.org/@floating-ui/core/-/core-1.6.0.tgz",
-			"integrity": "sha512-PcF++MykgmTj3CIyOQbKA/hDzOAiqI3mhuoN44WRCopIs1sgoDoU4oty4Jtqaj/y3oDU6fnVSm4QG0a3t5i0+g==",
+			"version": "1.7.1",
+			"resolved": "https://registry.npmjs.org/@floating-ui/core/-/core-1.7.1.tgz",
+			"integrity": "sha512-azI0DrjMMfIug/ExbBaeDVJXcY0a7EPvPjb2xAJPa4HeimBX+Z18HK8QQR3jb6356SnDDdxx+hinMLcJEDdOjw==",
+			"license": "MIT",
 			"dependencies": {
-				"@floating-ui/utils": "^0.2.1"
+				"@floating-ui/utils": "^0.2.9"
 			}
 		},
 		"node_modules/@floating-ui/dom": {
-			"version": "1.6.3",
-			"resolved": "https://registry.npmjs.org/@floating-ui/dom/-/dom-1.6.3.tgz",
-			"integrity": "sha512-RnDthu3mzPlQ31Ss/BTwQ1zjzIhr3lk1gZB1OC56h/1vEtaXkESrOqL5fQVMfXpwGtRwX+YsZBdyHtJMQnkArw==",
+			"version": "1.7.1",
+			"resolved": "https://registry.npmjs.org/@floating-ui/dom/-/dom-1.7.1.tgz",
+			"integrity": "sha512-cwsmW/zyw5ltYTUeeYJ60CnQuPqmGwuGVhG9w0PRaRKkAyi38BT5CKrpIbb+jtahSwUl04cWzSx9ZOIxeS6RsQ==",
+			"license": "MIT",
 			"dependencies": {
-				"@floating-ui/core": "^1.0.0",
-				"@floating-ui/utils": "^0.2.0"
+				"@floating-ui/core": "^1.7.1",
+				"@floating-ui/utils": "^0.2.9"
 			}
 		},
 		"node_modules/@floating-ui/utils": {
-			"version": "0.2.1",
-			"resolved": "https://registry.npmjs.org/@floating-ui/utils/-/utils-0.2.1.tgz",
-			"integrity": "sha512-9TANp6GPoMtYzQdt54kfAyMmz1+osLlXdg2ENroU7zzrtflTLrrC/lgrIfaSe+Wu0b89GKccT7vxXA0MoAIO+Q=="
+			"version": "0.2.9",
+			"resolved": "https://registry.npmjs.org/@floating-ui/utils/-/utils-0.2.9.tgz",
+			"integrity": "sha512-MDWhGtE+eHw5JW7lq4qhc5yRLS11ERl1c7Z6Xd0a58DozHES6EnNNwUWbMiG4J9Cgj053Bhk8zvlhFYKVhULwg==",
+			"license": "MIT"
 		},
 		"node_modules/@gulpjs/to-absolute-glob": {
 			"version": "4.0.0",
@@ -1750,9 +1753,10 @@
 			}
 		},
 		"node_modules/@internationalized/date": {
-			"version": "3.5.2",
-			"resolved": "https://registry.npmjs.org/@internationalized/date/-/date-3.5.2.tgz",
-			"integrity": "sha512-vo1yOMUt2hzp63IutEaTUxROdvQg1qlMRsbCvbay2AK2Gai7wIgCyK5weEX3nHkiLgo4qCXHijFNC/ILhlRpOQ==",
+			"version": "3.8.2",
+			"resolved": "https://registry.npmjs.org/@internationalized/date/-/date-3.8.2.tgz",
+			"integrity": "sha512-/wENk7CbvLbkUvX1tu0mwq49CVkkWpkXubGel6birjRPyo6uQ4nQpnq5xZu823zRCwwn82zgHrvgF1vZyvmVgA==",
+			"license": "Apache-2.0",
 			"dependencies": {
 				"@swc/helpers": "^0.5.0"
 			}
@@ -2032,9 +2036,10 @@
 			"integrity": "sha512-CZWV/q6TTe8ta61cZXjfnnHsfWIdFhms03M9T7Cnd5y2mdpylJM0rF1qRq+wsQVRMLz1OYPVEBU9ph2Bx8cxrg=="
 		},
 		"node_modules/@melt-ui/svelte": {
-			"version": "0.76.0",
-			"resolved": "https://registry.npmjs.org/@melt-ui/svelte/-/svelte-0.76.0.tgz",
-			"integrity": "sha512-X1ktxKujjLjOBt8LBvfckHGDMrkHWceRt1jdsUTf0EH76ikNPP1ofSoiV0IhlduDoCBV+2YchJ8kXCDfDXfC9Q==",
+			"version": "0.76.2",
+			"resolved": "https://registry.npmjs.org/@melt-ui/svelte/-/svelte-0.76.2.tgz",
+			"integrity": "sha512-7SbOa11tXUS95T3fReL+dwDs5FyJtCEqrqG3inRziDws346SYLsxOQ6HmX+4BkIsQh1R8U3XNa+EMmdMt38lMA==",
+			"license": "MIT",
 			"dependencies": {
 				"@floating-ui/core": "^1.3.1",
 				"@floating-ui/dom": "^1.4.5",
@@ -2610,11 +2615,12 @@
 			}
 		},
 		"node_modules/@swc/helpers": {
-			"version": "0.5.7",
-			"resolved": "https://registry.npmjs.org/@swc/helpers/-/helpers-0.5.7.tgz",
-			"integrity": "sha512-BVvNZhx362+l2tSwSuyEUV4h7+jk9raNdoTSdLfwTshXJSaGmYKluGRJznziCI3KX02Z19DdsQrdfrpXAU3Hfg==",
+			"version": "0.5.17",
+			"resolved": "https://registry.npmjs.org/@swc/helpers/-/helpers-0.5.17.tgz",
+			"integrity": "sha512-5IKx/Y13RsYd+sauPb2x+U/xZikHjolzfuDgTAl/Tdf3Q8rslRvC19NKDLgAJQ6wsqADk10ntlv08nPFw/gO/A==",
+			"license": "Apache-2.0",
 			"dependencies": {
-				"tslib": "^2.4.0"
+				"tslib": "^2.8.0"
 			}
 		},
 		"node_modules/@tailwindcss/container-queries": {
@@ -4381,16 +4387,20 @@
 			}
 		},
 		"node_modules/bits-ui": {
-			"version": "0.19.7",
-			"resolved": "https://registry.npmjs.org/bits-ui/-/bits-ui-0.19.7.tgz",
-			"integrity": "sha512-GHUpKvN7QyazhnZNkUy0lxg6W1M6KJHWSZ4a/UGCjPE6nQgk6vKbGysY67PkDtQMknZTZAzVoMj1Eic4IKeCRQ==",
+			"version": "0.21.15",
+			"resolved": "https://registry.npmjs.org/bits-ui/-/bits-ui-0.21.15.tgz",
+			"integrity": "sha512-+m5WSpJnFdCcNdXSTIVC1WYBozipO03qRh03GFWgrdxoHiolCfwW71EYG4LPCWYPG6KcTZV0Cj6iHSiZ7cdKdg==",
+			"license": "MIT",
 			"dependencies": {
 				"@internationalized/date": "^3.5.1",
-				"@melt-ui/svelte": "0.76.0",
+				"@melt-ui/svelte": "0.76.2",
 				"nanoid": "^5.0.5"
 			},
+			"funding": {
+				"url": "https://github.com/sponsors/huntabyte"
+			},
 			"peerDependencies": {
-				"svelte": "^4.0.0"
+				"svelte": "^4.0.0 || ^5.0.0-next.118"
 			}
 		},
 		"node_modules/bl": {
@@ -7930,9 +7940,9 @@
 			}
 		},
 		"node_modules/katex": {
-			"version": "0.16.21",
-			"resolved": "https://registry.npmjs.org/katex/-/katex-0.16.21.tgz",
-			"integrity": "sha512-XvqR7FgOHtWupfMiigNzmh+MgUVmDGU2kXZm899ZkPfcuoPuFxyHmXsgATDpFZDAXCI8tvinaVcDo8PIIJSo4A==",
+			"version": "0.16.22",
+			"resolved": "https://registry.npmjs.org/katex/-/katex-0.16.22.tgz",
+			"integrity": "sha512-XCHRdUw4lf3SKBaJe4EvgqIuWwkPSo9XoeO8GjQW94Bp7TWv9hNhzZjZ+OH9yf1UmLygb7DIT5GSFQiyt16zYg==",
 			"funding": [
 				"https://opencollective.com/katex",
 				"https://github.com/sponsors/katex"
@@ -11842,9 +11852,10 @@
 			}
 		},
 		"node_modules/tslib": {
-			"version": "2.6.2",
-			"resolved": "https://registry.npmjs.org/tslib/-/tslib-2.6.2.tgz",
-			"integrity": "sha512-AEYxH93jGFPn/a2iVAwW87VuUIkR1FVUKB77NwMF7nBTDkDrrT/Hpt/IrCJ0QXhW27jTBDcf5ZY7w6RiqTMw2Q=="
+			"version": "2.8.1",
+			"resolved": "https://registry.npmjs.org/tslib/-/tslib-2.8.1.tgz",
+			"integrity": "sha512-oJFu94HQb+KVduSUQL7wnpmqnfmLsOA/nAh6b6EH0wCEoK0/mPeXU6c3wKDV83MkOuHPRHtSXKKU99IBazS/2w==",
+			"license": "0BSD"
 		},
 		"node_modules/tunnel-agent": {
 			"version": "0.6.0",

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "open-webui",
-	"version": "0.6.11",
+	"version": "0.6.14",
 	"private": true,
 	"scripts": {
 		"dev": "npm run pyodide:fetch && vite dev --host",
@@ -75,7 +75,7 @@
 		"@tiptap/starter-kit": "^2.10.0",
 		"@xyflow/svelte": "^0.1.19",
 		"async": "^3.2.5",
-		"bits-ui": "^0.19.7",
+		"bits-ui": "^0.21.15",
 		"codemirror": "^6.0.1",
 		"codemirror-lang-elixir": "^4.0.0",
 		"codemirror-lang-hcl": "^0.1.0",
@@ -95,7 +95,7 @@
 		"idb": "^7.1.1",
 		"js-sha256": "^0.10.1",
 		"jspdf": "^3.0.0",
-		"katex": "^0.16.21",
+		"katex": "^0.16.22",
 		"kokoro-js": "^1.1.1",
 		"marked": "^9.1.0",
 		"mermaid": "^11.6.0",

pyproject.toml

@@ -7,7 +7,7 @@ authors = [
 license = { file = "LICENSE" }
 dependencies = [
     "fastapi==0.115.7",
-    "uvicorn[standard]==0.34.0",
+    "uvicorn[standard]==0.34.2",
     "pydantic==2.10.6",
     "python-multipart==0.0.20",
 
@@ -15,12 +15,11 @@ dependencies = [
     "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",
@@ -83,13 +82,13 @@ dependencies = [
     "openpyxl==3.1.5",
     "pyxlsb==1.0.10",
     "xlrd==2.0.1",
-    "validators==0.34.0",
+    "validators==0.35.0",
     "psutil",
     "sentencepiece",
     "soundfile==0.13.1",
-    "azure-ai-documentintelligence==1.0.0",
+    "azure-ai-documentintelligence==1.0.2",
 
-    "pillow==11.1.0",
+    "pillow==11.2.1",
     "opencv-python-headless==4.11.0.86",
     "rapidocr-onnxruntime==1.4.4",
     "rank-bm25==0.2.2",

scripts/prepare-pyodide.js

@@ -12,7 +12,8 @@ const packages = [
 	'sympy',
 	'tiktoken',
 	'seaborn',
-	'pytz'
+	'pytz',
+	'black'
 ];
 
 import { loadPyodide } from 'pyodide';

src/app.css

@@ -44,6 +44,10 @@ code {
 	font-family: 'InstrumentSerif', sans-serif;
 }
 
+.marked a {
+	@apply underline;
+}
+
 math {
 	margin-top: 1rem;
 }

src/lib/apis/auths/index.ts

@@ -336,7 +336,7 @@ export const userSignOut = async () => {
 	})
 		.then(async (res) => {
 			if (!res.ok) throw await res.json();
-			return res;
+			return res.json();
 		})
 		.catch((err) => {
 			console.error(err);

src/lib/apis/index.ts

@@ -612,6 +612,78 @@ export const generateTitle = async (
 	}
 };
 
+export const generateFollowUps = async (
+	token: string = '',
+	model: string,
+	messages: string,
+	chat_id?: string
+) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_BASE_URL}/api/v1/tasks/follow_ups/completions`, {
+		method: 'POST',
+		headers: {
+			Accept: 'application/json',
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			model: model,
+			messages: messages,
+			...(chat_id && { chat_id: chat_id })
+		})
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.error(err);
+			if ('detail' in err) {
+				error = err.detail;
+			}
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	try {
+		// Step 1: Safely extract the response string
+		const response = res?.choices[0]?.message?.content ?? '';
+
+		// Step 2: Attempt to fix common JSON format issues like single quotes
+		const sanitizedResponse = response.replace(/['‘’`]/g, '"'); // Convert single quotes to double quotes for valid JSON
+
+		// Step 3: Find the relevant JSON block within the response
+		const jsonStartIndex = sanitizedResponse.indexOf('{');
+		const jsonEndIndex = sanitizedResponse.lastIndexOf('}');
+
+		// Step 4: Check if we found a valid JSON block (with both `{` and `}`)
+		if (jsonStartIndex !== -1 && jsonEndIndex !== -1) {
+			const jsonResponse = sanitizedResponse.substring(jsonStartIndex, jsonEndIndex + 1);
+
+			// Step 5: Parse the JSON block
+			const parsed = JSON.parse(jsonResponse);
+
+			// Step 6: If there's a "follow_ups" key, return the follow_ups array; otherwise, return an empty array
+			if (parsed && parsed.follow_ups) {
+				return Array.isArray(parsed.follow_ups) ? parsed.follow_ups : [];
+			} else {
+				return [];
+			}
+		}
+
+		// If no valid JSON block found, return an empty array
+		return [];
+	} catch (e) {
+		// Catch and safely return empty array on any parsing errors
+		console.error('Failed to parse response: ', e);
+		return [];
+	}
+};
+
 export const generateTags = async (
 	token: string = '',
 	model: string,

src/lib/apis/retrieval/index.ts

@@ -184,8 +184,15 @@ type OpenAIConfigForm = {
 	url: string;
 };
 
+type AzureOpenAIConfigForm = {
+	key: string;
+	url: string;
+	version: string;
+};
+
 type EmbeddingModelUpdateForm = {
 	openai_config?: OpenAIConfigForm;
+	azure_openai_config?: AzureOpenAIConfigForm;
 	embedding_engine: string;
 	embedding_model: string;
 	embedding_batch_size?: number;

src/lib/apis/tools/index.ts

@@ -31,6 +31,40 @@ export const createNewTool = async (token: string, tool: object) => {
 	return res;
 };
 
+export const loadToolByUrl = async (token: string = '', url: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/tools/load/url`, {
+		method: 'POST',
+		headers: {
+			Accept: 'application/json',
+			'Content-Type': 'application/json',
+			authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			url
+		})
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.then((json) => {
+			return json;
+		})
+		.catch((err) => {
+			error = err.detail;
+			console.error(err);
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
 export const getTools = async (token: string = '') => {
 	let error = null;
 

src/lib/apis/users/index.ts

@@ -393,6 +393,7 @@ export const updateUserById = async (token: string, userId: string, user: UserUp
 		},
 		body: JSON.stringify({
 			profile_image_url: user.profile_image_url,
+			role: user.role,
 			email: user.email,
 			name: user.name,
 			password: user.password !== '' ? user.password : undefined

src/lib/components/AddConnectionModal.svelte

@@ -49,6 +49,9 @@
 	let loading = false;
 
 	const verifyOllamaHandler = async () => {
+		// remove trailing slash from url
+		url = url.replace(/\/$/, '');
+
 		const res = await verifyOllamaConnection(localStorage.token, {
 			url,
 			key
@@ -62,6 +65,9 @@
 	};
 
 	const verifyOpenAIHandler = async () => {
+		// remove trailing slash from url
+		url = url.replace(/\/$/, '');
+
 		const res = await verifyOpenAIConnection(
 			localStorage.token,
 			{

src/lib/components/ImportModal.svelte

@@ -4,7 +4,6 @@
 	const i18n = getContext('i18n');
 
 	import Modal from '$lib/components/common/Modal.svelte';
-	import { loadFunctionByUrl } from '$lib/apis/functions';
 	import { extractFrontmatter } from '$lib/utils';
 
 	export let show = false;
@@ -12,6 +11,9 @@
 	export let onImport = (e) => {};
 	export let onClose = () => {};
 
+	export let loadUrlHandler: Function = () => {};
+	export let successMessage: string = '';
+
 	let loading = false;
 	let url = '';
 
@@ -24,14 +26,19 @@
 			return;
 		}
 
-		const res = await loadFunctionByUrl(localStorage.token, url).catch((err) => {
+		const res = await loadUrlHandler(url).catch((err) => {
 			toast.error(`${err}`);
-
+			loading = false;
 			return null;
 		});
 
 		if (res) {
-			toast.success($i18n.t('Function loaded successfully'));
+			if (!successMessage) {
+				successMessage = $i18n.t('Function imported successfully');
+			}
+
+			toast.success(successMessage);
+
 			let func = res;
 			func.id = func.id || func.name.replace(/\s+/g, '_').toLowerCase();
 
@@ -92,9 +99,11 @@
 									class="w-full text-sm bg-transparent disabled:text-gray-500 dark:disabled:text-gray-500 outline-hidden"
 									type="url"
 									bind:value={url}
-									placeholder={$i18n.t('Enter the URL of the function to import')}
+									placeholder={$i18n.t('Enter the URL to import')}
 									required
 								/>
+
+								<!-- $i18n.t('Enter the URL of the function to import') -->
 							</div>
 						</div>
 					</div>

src/lib/components/admin/Evaluations.svelte

@@ -1,6 +1,8 @@
 <script>
 	import { getContext, tick, onMount } from 'svelte';
-	import { toast } from 'svelte-sonner';
+	import { goto } from '$app/navigation';
+	import { page } from '$app/stores';
+
 	import Leaderboard from './Evaluations/Leaderboard.svelte';
 	import Feedbacks from './Evaluations/Feedbacks.svelte';
 
@@ -8,7 +10,24 @@
 
 	const i18n = getContext('i18n');
 
-	let selectedTab = 'leaderboard';
+	let selectedTab;
+	$: {
+		const pathParts = $page.url.pathname.split('/');
+		const tabFromPath = pathParts[pathParts.length - 1];
+		selectedTab = ['leaderboard', 'feedbacks'].includes(tabFromPath) ? tabFromPath : 'leaderboard';
+	}
+
+	$: if (selectedTab) {
+		// scroll to selectedTab
+		scrollToTab(selectedTab);
+	}
+
+	const scrollToTab = (tabId) => {
+		const tabElement = document.getElementById(tabId);
+		if (tabElement) {
+			tabElement.scrollIntoView({ behavior: 'smooth', block: 'nearest', inline: 'start' });
+		}
+	};
 
 	let loaded = false;
 	let feedbacks = [];
@@ -27,6 +46,9 @@
 				}
 			});
 		}
+
+		// Scroll to the selected tab on mount
+		scrollToTab(selectedTab);
 	});
 </script>
 
@@ -37,12 +59,13 @@
 			class="tabs flex flex-row overflow-x-auto gap-2.5 max-w-full lg:gap-1 lg:flex-col lg:flex-none lg:w-40 dark:text-gray-200 text-sm font-medium text-left scrollbar-none"
 		>
 			<button
+				id="leaderboard"
 				class="px-0.5 py-1 min-w-fit rounded-lg lg:flex-none flex text-right transition {selectedTab ===
 				'leaderboard'
 					? ''
 					: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 				on:click={() => {
-					selectedTab = 'leaderboard';
+					goto('/admin/evaluations/leaderboard');
 				}}
 			>
 				<div class=" self-center mr-2">
@@ -63,12 +86,13 @@
 			</button>
 
 			<button
+				id="feedbacks"
 				class="px-0.5 py-1 min-w-fit rounded-lg lg:flex-none flex text-right transition {selectedTab ===
 				'feedbacks'
 					? ''
 					: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 				on:click={() => {
-					selectedTab = 'feedbacks';
+					goto('/admin/evaluations/feedbacks');
 				}}
 			>
 				<div class=" self-center mr-2">

src/lib/components/admin/Functions.svelte

@@ -13,6 +13,7 @@
 		exportFunctions,
 		getFunctionById,
 		getFunctions,
+		loadFunctionByUrl,
 		toggleFunctionById,
 		toggleGlobalById
 	} from '$lib/apis/functions';
@@ -34,7 +35,7 @@
 	import ChevronRight from '../icons/ChevronRight.svelte';
 	import XMark from '../icons/XMark.svelte';
 	import AddFunctionMenu from './Functions/AddFunctionMenu.svelte';
-	import ImportModal from './Functions/ImportModal.svelte';
+	import ImportModal from '../ImportModal.svelte';
 
 	const i18n = getContext('i18n');
 
@@ -203,6 +204,9 @@
 
 <ImportModal
 	bind:show={showImportModal}
+	loadUrlHandler={async (url) => {
+		return await loadFunctionByUrl(localStorage.token, url);
+	}}
 	onImport={(func) => {
 		sessionStorage.function = JSON.stringify({
 			...func

src/lib/components/admin/Settings.svelte

@@ -1,5 +1,7 @@
 <script>
 	import { getContext, tick, onMount } from 'svelte';
+	import { page } from '$app/stores';
+	import { goto } from '$app/navigation';
 	import { toast } from 'svelte-sonner';
 
 	import { config } from '$lib/stores';
@@ -26,6 +28,41 @@
 
 	let selectedTab = 'general';
 
+	// Get current tab from URL pathname, default to 'general'
+	$: {
+		const pathParts = $page.url.pathname.split('/');
+		const tabFromPath = pathParts[pathParts.length - 1];
+		selectedTab = [
+			'general',
+			'connections',
+			'models',
+			'evaluations',
+			'tools',
+			'documents',
+			'web',
+			'code-execution',
+			'interface',
+			'audio',
+			'images',
+			'pipelines',
+			'db'
+		].includes(tabFromPath)
+			? tabFromPath
+			: 'general';
+	}
+
+	$: if (selectedTab) {
+		// scroll to selectedTab
+		scrollToTab(selectedTab);
+	}
+
+	const scrollToTab = (tabId) => {
+		const tabElement = document.getElementById(tabId);
+		if (tabElement) {
+			tabElement.scrollIntoView({ behavior: 'smooth', block: 'nearest', inline: 'start' });
+		}
+	};
+
 	onMount(() => {
 		const containerElement = document.getElementById('admin-settings-tabs-container');
 
@@ -37,6 +74,9 @@
 				}
 			});
 		}
+
+		// Scroll to the selected tab on mount
+		scrollToTab(selectedTab);
 	});
 </script>
 
@@ -46,12 +86,13 @@
 		class="tabs flex flex-row overflow-x-auto gap-2.5 max-w-full lg:gap-1 lg:flex-col lg:flex-none lg:w-40 dark:text-gray-200 text-sm font-medium text-left scrollbar-none"
 	>
 		<button
+			id="general"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 lg:flex-none flex text-right transition {selectedTab ===
 			'general'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'general';
+				goto('/admin/settings/general');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -72,12 +113,13 @@
 		</button>
 
 		<button
+			id="connections"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'connections'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'connections';
+				goto('/admin/settings/connections');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -96,12 +138,13 @@
 		</button>
 
 		<button
+			id="models"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'models'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'models';
+				goto('/admin/settings/models');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -122,12 +165,13 @@
 		</button>
 
 		<button
+			id="evaluations"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'evaluations'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'evaluations';
+				goto('/admin/settings/evaluations');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -137,12 +181,13 @@
 		</button>
 
 		<button
+			id="tools"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'tools'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'tools';
+				goto('/admin/settings/tools');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -163,12 +208,13 @@
 		</button>
 
 		<button
+			id="documents"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'documents'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'documents';
+				goto('/admin/settings/documents');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -193,12 +239,13 @@
 		</button>
 
 		<button
+			id="web"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'web'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'web';
+				goto('/admin/settings/web');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -217,12 +264,13 @@
 		</button>
 
 		<button
+			id="code-execution"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'code-execution'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'code-execution';
+				goto('/admin/settings/code-execution');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -243,12 +291,13 @@
 		</button>
 
 		<button
+			id="interface"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'interface'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'interface';
+				goto('/admin/settings/interface');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -269,12 +318,13 @@
 		</button>
 
 		<button
+			id="audio"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'audio'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'audio';
+				goto('/admin/settings/audio');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -296,12 +346,13 @@
 		</button>
 
 		<button
+			id="images"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'images'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'images';
+				goto('/admin/settings/images');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -322,12 +373,13 @@
 		</button>
 
 		<button
+			id="pipelines"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'pipelines'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'pipelines';
+				goto('/admin/settings/pipelines');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -352,12 +404,13 @@
 		</button>
 
 		<button
+			id="db"
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
 			'db'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'db';
+				goto('/admin/settings/db');
 			}}
 		>
 			<div class=" self-center mr-2">

src/lib/components/admin/Settings/Documents.svelte

@@ -46,6 +46,10 @@
 	let OpenAIUrl = '';
 	let OpenAIKey = '';
 
+	let AzureOpenAIUrl = '';
+	let AzureOpenAIKey = '';
+	let AzureOpenAIVersion = '';
+
 	let OllamaUrl = '';
 	let OllamaKey = '';
 
@@ -86,7 +90,14 @@
 			return;
 		}
 
-		if ((embeddingEngine === 'openai' && OpenAIKey === '') || OpenAIUrl === '') {
+		if (embeddingEngine === 'openai' && (OpenAIKey === '' || OpenAIUrl === '')) {
+			toast.error($i18n.t('OpenAI URL/Key required.'));
+			return;
+		}
+		if (
+			embeddingEngine === 'azure_openai' &&
+			(AzureOpenAIKey === '' || AzureOpenAIUrl === '' || AzureOpenAIVersion === '')
+		) {
 			toast.error($i18n.t('OpenAI URL/Key required.'));
 			return;
 		}
@@ -105,6 +116,11 @@
 			openai_config: {
 				key: OpenAIKey,
 				url: OpenAIUrl
+			},
+			azure_openai_config: {
+				key: AzureOpenAIKey,
+				url: AzureOpenAIUrl,
+				version: AzureOpenAIVersion
 			}
 		}).catch(async (error) => {
 			toast.error(`${error}`);
@@ -150,6 +166,14 @@
 			return;
 		}
 
+		if (
+			RAGConfig.CONTENT_EXTRACTION_ENGINE === 'datalab_marker' &&
+			!RAGConfig.DATALAB_MARKER_API_KEY
+		) {
+			toast.error($i18n.t('Datalab Marker API Key required.'));
+			return;
+		}
+
 		if (
 			RAGConfig.CONTENT_EXTRACTION_ENGINE === 'document_intelligence' &&
 			(RAGConfig.DOCUMENT_INTELLIGENCE_ENDPOINT === '' ||
@@ -170,12 +194,20 @@
 			await embeddingModelUpdateHandler();
 		}
 
-		RAGConfig.ALLOWED_FILE_EXTENSIONS = (RAGConfig?.ALLOWED_FILE_EXTENSIONS ?? '')
-			.split(',')
-			.map((ext) => ext.trim())
-			.filter((ext) => ext !== '');
-
-		const res = await updateRAGConfig(localStorage.token, RAGConfig);
+		const res = await updateRAGConfig(localStorage.token, {
+			...RAGConfig,
+			ALLOWED_FILE_EXTENSIONS: RAGConfig.ALLOWED_FILE_EXTENSIONS.split(',')
+				.map((ext) => ext.trim())
+				.filter((ext) => ext !== ''),
+			DATALAB_MARKER_LANGS: RAGConfig.DATALAB_MARKER_LANGS.split(',')
+				.map((code) => code.trim())
+				.filter((code) => code !== '')
+				.join(', '),
+			DOCLING_PICTURE_DESCRIPTION_LOCAL: JSON.parse(
+				RAGConfig.DOCLING_PICTURE_DESCRIPTION_LOCAL || '{}'
+			),
+			DOCLING_PICTURE_DESCRIPTION_API: JSON.parse(RAGConfig.DOCLING_PICTURE_DESCRIPTION_API || '{}')
+		});
 		dispatch('save');
 	};
 
@@ -192,6 +224,10 @@
 
 			OllamaKey = embeddingConfig.ollama_config.key;
 			OllamaUrl = embeddingConfig.ollama_config.url;
+
+			AzureOpenAIKey = embeddingConfig.azure_openai_config.key;
+			AzureOpenAIUrl = embeddingConfig.azure_openai_config.url;
+			AzureOpenAIVersion = embeddingConfig.azure_openai_config.version;
 		}
 	};
 	onMount(async () => {
@@ -200,6 +236,17 @@
 		const config = await getRAGConfig(localStorage.token);
 		config.ALLOWED_FILE_EXTENSIONS = (config?.ALLOWED_FILE_EXTENSIONS ?? []).join(', ');
 
+		config.DOCLING_PICTURE_DESCRIPTION_LOCAL = JSON.stringify(
+			config.DOCLING_PICTURE_DESCRIPTION_LOCAL ?? {},
+			null,
+			2
+		);
+		config.DOCLING_PICTURE_DESCRIPTION_API = JSON.stringify(
+			config.DOCLING_PICTURE_DESCRIPTION_API ?? {},
+			null,
+			2
+		);
+
 		RAGConfig = config;
 	});
 </script>
@@ -274,6 +321,7 @@
 									<option value="external">{$i18n.t('External')}</option>
 									<option value="tika">{$i18n.t('Tika')}</option>
 									<option value="docling">{$i18n.t('Docling')}</option>
+									<option value="datalab_marker">{$i18n.t('Datalab Marker API')}</option>
 									<option value="document_intelligence">{$i18n.t('Document Intelligence')}</option>
 									<option value="mistral_ocr">{$i18n.t('Mistral OCR')}</option>
 								</select>
@@ -291,6 +339,138 @@
 									</div>
 								</div>
 							</div>
+						{:else if RAGConfig.CONTENT_EXTRACTION_ENGINE === 'datalab_marker'}
+							<div class="my-0.5 flex gap-2 pr-2">
+								<SensitiveInput
+									placeholder={$i18n.t('Enter Datalab Marker API Key')}
+									required={false}
+									bind:value={RAGConfig.DATALAB_MARKER_API_KEY}
+								/>
+							</div>
+
+							<div class="flex justify-between w-full mt-2">
+								<div class="text-xs font-medium">
+									{$i18n.t('Languages')}
+								</div>
+
+								<input
+									class="text-sm bg-transparent outline-hidden"
+									type="text"
+									bind:value={RAGConfig.DATALAB_MARKER_LANGS}
+									placeholder={$i18n.t('e.g.) en,fr,de')}
+								/>
+							</div>
+
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t(
+											'Significantly improves accuracy by using an LLM to enhance tables, forms, inline math, and layout detection. Will increase latency. Defaults to True.'
+										)}
+										placement="top-start"
+									>
+										{$i18n.t('Use LLM')}
+									</Tooltip>
+								</div>
+								<div class="flex items-center">
+									<Switch bind:state={RAGConfig.DATALAB_MARKER_USE_LLM} />
+								</div>
+							</div>
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t('Skip the cache and re-run the inference. Defaults to False.')}
+										placement="top-start"
+									>
+										{$i18n.t('Skip Cache')}
+									</Tooltip>
+								</div>
+								<div class="flex items-center">
+									<Switch bind:state={RAGConfig.DATALAB_MARKER_SKIP_CACHE} />
+								</div>
+							</div>
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t(
+											'Force OCR on all pages of the PDF. This can lead to worse results if you have good text in your PDFs. Defaults to False.'
+										)}
+										placement="top-start"
+									>
+										{$i18n.t('Force OCR')}
+									</Tooltip>
+								</div>
+								<div class="flex items-center">
+									<Switch bind:state={RAGConfig.DATALAB_MARKER_FORCE_OCR} />
+								</div>
+							</div>
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t(
+											'Whether to paginate the output. Each page will be separated by a horizontal rule and page number. Defaults to False.'
+										)}
+										placement="top-start"
+									>
+										{$i18n.t('Paginate')}
+									</Tooltip>
+								</div>
+								<div class="flex items-center">
+									<Switch bind:state={RAGConfig.DATALAB_MARKER_PAGINATE} />
+								</div>
+							</div>
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t(
+											'Strip existing OCR text from the PDF and re-run OCR. Ignored if Force OCR is enabled. Defaults to False.'
+										)}
+										placement="top-start"
+									>
+										{$i18n.t('Strip Existing OCR')}
+									</Tooltip>
+								</div>
+								<div class="flex items-center">
+									<Switch bind:state={RAGConfig.DATALAB_MARKER_STRIP_EXISTING_OCR} />
+								</div>
+							</div>
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t(
+											'Disable image extraction from the PDF. If Use LLM is enabled, images will be automatically captioned. Defaults to False.'
+										)}
+										placement="top-start"
+									>
+										{$i18n.t('Disable Image Extraction')}
+									</Tooltip>
+								</div>
+								<div class="flex items-center">
+									<Switch bind:state={RAGConfig.DATALAB_MARKER_DISABLE_IMAGE_EXTRACTION} />
+								</div>
+							</div>
+							<div class="flex justify-between w-full mt-2">
+								<div class="self-center text-xs font-medium">
+									<Tooltip
+										content={$i18n.t(
+											"The output format for the text. Can be 'json', 'markdown', or 'html'. Defaults to 'markdown'."
+										)}
+										placement="top-start"
+									>
+										{$i18n.t('Output Format')}
+									</Tooltip>
+								</div>
+								<div class="">
+									<select
+										class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 text-xs bg-transparent outline-hidden text-right"
+										bind:value={RAGConfig.DATALAB_MARKER_OUTPUT_FORMAT}
+									>
+										<option value="markdown">{$i18n.t('Markdown')}</option>
+										<option value="json">{$i18n.t('JSON')}</option>
+										<option value="html">{$i18n.t('HTML')}</option>
+									</select>
+								</div>
+							</div>
 						{:else if RAGConfig.CONTENT_EXTRACTION_ENGINE === 'external'}
 							<div class="my-0.5 flex gap-2 pr-2">
 								<input
@@ -345,6 +525,71 @@
 									</div>
 								</div>
 							</div>
+							{#if RAGConfig.DOCLING_DO_PICTURE_DESCRIPTION}
+								<div class="flex justify-between w-full mt-2">
+									<div class="self-center text-xs font-medium">
+										<Tooltip content={''} placement="top-start">
+											{$i18n.t('Picture Description Mode')}
+										</Tooltip>
+									</div>
+									<div class="">
+										<select
+											class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 text-xs bg-transparent outline-hidden text-right"
+											bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE}
+										>
+											<option value="">{$i18n.t('Default')}</option>
+											<option value="local">{$i18n.t('Local')}</option>
+											<option value="api">{$i18n.t('API')}</option>
+										</select>
+									</div>
+								</div>
+
+								{#if RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE === 'local'}
+									<div class="flex flex-col gap-2 mt-2">
+										<div class=" flex flex-col w-full justify-between">
+											<div class=" mb-1 text-xs font-medium">
+												{$i18n.t('Picture Description Local Config')}
+											</div>
+											<div class="flex w-full items-center relative">
+												<Tooltip
+													content={$i18n.t(
+														'Options for running a local vision-language model in the picture description. The parameters refer to a model hosted on Hugging Face. This parameter is mutually exclusive with picture_description_api.'
+													)}
+													placement="top-start"
+													className="w-full"
+												>
+													<Textarea
+														bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_LOCAL}
+														placeholder={$i18n.t('Enter Config in JSON format')}
+													/>
+												</Tooltip>
+											</div>
+										</div>
+									</div>
+								{:else if RAGConfig.DOCLING_PICTURE_DESCRIPTION_MODE === 'api'}
+									<div class="flex flex-col gap-2 mt-2">
+										<div class=" flex flex-col w-full justify-between">
+											<div class=" mb-1 text-xs font-medium">
+												{$i18n.t('Picture Description API Config')}
+											</div>
+											<div class="flex w-full items-center relative">
+												<Tooltip
+													content={$i18n.t(
+														'API details for using a vision-language model in the picture description. This parameter is mutually exclusive with picture_description_local.'
+													)}
+													placement="top-start"
+													className="w-full"
+												>
+													<Textarea
+														bind:value={RAGConfig.DOCLING_PICTURE_DESCRIPTION_API}
+														placeholder={$i18n.t('Enter Config in JSON format')}
+													/>
+												</Tooltip>
+											</div>
+										</div>
+									</div>
+								{/if}
+							{/if}
 						{:else if RAGConfig.CONTENT_EXTRACTION_ENGINE === 'document_intelligence'}
 							<div class="my-0.5 flex gap-2 pr-2">
 								<input
@@ -462,6 +707,8 @@
 												embeddingModel = '';
 											} else if (e.target.value === 'openai') {
 												embeddingModel = 'text-embedding-3-small';
+											} else if (e.target.value === 'azure_openai') {
+												embeddingModel = 'text-embedding-3-small';
 											} else if (e.target.value === '') {
 												embeddingModel = 'sentence-transformers/all-MiniLM-L6-v2';
 											}
@@ -470,6 +717,7 @@
 										<option value="">{$i18n.t('Default (SentenceTransformers)')}</option>
 										<option value="ollama">{$i18n.t('Ollama')}</option>
 										<option value="openai">{$i18n.t('OpenAI')}</option>
+										<option value="azure_openai">Azure OpenAI</option>
 									</select>
 								</div>
 							</div>
@@ -500,6 +748,26 @@
 										required={false}
 									/>
 								</div>
+							{:else if embeddingEngine === 'azure_openai'}
+								<div class="my-0.5 flex flex-col gap-2 pr-2 w-full">
+									<div class="flex gap-2">
+										<input
+											class="flex-1 w-full text-sm bg-transparent outline-hidden"
+											placeholder={$i18n.t('API Base URL')}
+											bind:value={AzureOpenAIUrl}
+											required
+										/>
+										<SensitiveInput placeholder={$i18n.t('API Key')} bind:value={AzureOpenAIKey} />
+									</div>
+									<div class="flex gap-2">
+										<input
+											class="flex-1 w-full text-sm bg-transparent outline-hidden"
+											placeholder="Version"
+											bind:value={AzureOpenAIVersion}
+											required
+										/>
+									</div>
+								</div>
 							{/if}
 						</div>
 
@@ -596,7 +864,7 @@
 							</div>
 						</div>
 
-						{#if embeddingEngine === 'ollama' || embeddingEngine === 'openai'}
+						{#if embeddingEngine === 'ollama' || embeddingEngine === 'openai' || embeddingEngine === 'azure_openai'}
 							<div class="  mb-2.5 flex w-full justify-between">
 								<div class=" self-center text-xs font-medium">
 									{$i18n.t('Embedding Batch Size')}
@@ -642,12 +910,7 @@
 							<div class="  mb-2.5 flex w-full justify-between">
 								<div class=" self-center text-xs font-medium">{$i18n.t('Hybrid Search')}</div>
 								<div class="flex items-center relative">
-									<Switch
-										bind:state={RAGConfig.ENABLE_RAG_HYBRID_SEARCH}
-										on:change={() => {
-											submitHandler();
-										}}
-									/>
+									<Switch bind:state={RAGConfig.ENABLE_RAG_HYBRID_SEARCH} />
 								</div>
 							</div>
 

src/lib/components/admin/Settings/General.svelte

@@ -79,6 +79,7 @@
 	const updateHandler = async () => {
 		webhookUrl = await updateWebhookUrl(localStorage.token, webhookUrl);
 		const res = await updateAdminConfig(localStorage.token, adminConfig);
+		await updateLdapConfig(localStorage.token, ENABLE_LDAP);
 		await updateLdapServerHandler();
 
 		if (res) {
@@ -311,7 +312,6 @@
 							{$i18n.t('Pending User Overlay Title')}
 						</div>
 						<Textarea
-							rows={2}
 							placeholder={$i18n.t(
 								'Enter a title for the pending user info overlay. Leave empty for default.'
 							)}
@@ -401,12 +401,7 @@
 								<div class="  font-medium">{$i18n.t('LDAP')}</div>
 
 								<div class="mt-1">
-									<Switch
-										bind:state={ENABLE_LDAP}
-										on:change={async () => {
-											updateLdapConfig(localStorage.token, ENABLE_LDAP);
-										}}
-									/>
+									<Switch bind:state={ENABLE_LDAP} />
 								</div>
 							</div>
 

src/lib/components/admin/Settings/Interface.svelte

@@ -31,6 +31,8 @@
 		TASK_MODEL_EXTERNAL: '',
 		ENABLE_TITLE_GENERATION: true,
 		TITLE_GENERATION_PROMPT_TEMPLATE: '',
+		ENABLE_FOLLOW_UP_GENERATION: true,
+		FOLLOW_UP_GENERATION_PROMPT_TEMPLATE: '',
 		IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE: '',
 		ENABLE_AUTOCOMPLETE_GENERATION: true,
 		AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH: -1,
@@ -235,6 +237,32 @@
 					</div>
 				{/if}
 
+				<div class="mb-2.5 flex w-full items-center justify-between">
+					<div class=" self-center text-xs font-medium">
+						{$i18n.t('Follow Up Generation')}
+					</div>
+
+					<Switch bind:state={taskConfig.ENABLE_FOLLOW_UP_GENERATION} />
+				</div>
+
+				{#if taskConfig.ENABLE_FOLLOW_UP_GENERATION}
+					<div class="mb-2.5">
+						<div class=" mb-1 text-xs font-medium">{$i18n.t('Follow Up Generation Prompt')}</div>
+
+						<Tooltip
+							content={$i18n.t('Leave empty to use the default prompt, or enter a custom prompt')}
+							placement="top-start"
+						>
+							<Textarea
+								bind:value={taskConfig.FOLLOW_UP_GENERATION_PROMPT_TEMPLATE}
+								placeholder={$i18n.t(
+									'Leave empty to use the default prompt, or enter a custom prompt'
+								)}
+							/>
+						</Tooltip>
+					</div>
+				{/if}
+
 				<div class="mb-2.5 flex w-full items-center justify-between">
 					<div class=" self-center text-xs font-medium">
 						{$i18n.t('Tags Generation')}

src/lib/components/admin/Settings/Models.svelte

@@ -75,6 +75,8 @@
 	};
 
 	const init = async () => {
+		models = null;
+
 		workspaceModels = await getBaseModels(localStorage.token);
 		baseModels = await getModels(localStorage.token, null, true);
 
@@ -126,6 +128,7 @@
 				toast.success($i18n.t('Model updated successfully'));
 			}
 		}
+		await init();
 
 		_models.set(
 			await getModels(
@@ -133,7 +136,6 @@
 				$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
 			)
 		);
-		await init();
 	};
 
 	const toggleModelHandler = async (model) => {

src/lib/components/admin/Settings/Models/ModelMenu.svelte

@@ -45,7 +45,7 @@
 
 	<div slot="content">
 		<DropdownMenu.Content
-			class="w-full max-w-[160px] rounded-xl px-1 py-1.5 border border-gray-300/30 dark:border-gray-700/50 z-50 bg-white dark:bg-gray-850 dark:text-white shadow-sm"
+			class="w-full max-w-[170px] rounded-xl px-1 py-1.5 border border-gray-300/30 dark:border-gray-700/50 z-50 bg-white dark:bg-gray-850 dark:text-white shadow-sm"
 			sideOffset={-2}
 			side="bottom"
 			align="start"

src/lib/components/admin/Settings/WebSearch.svelte

@@ -446,15 +446,54 @@
 								</div>
 							</div>
 						{:else if webConfig.WEB_SEARCH_ENGINE === 'perplexity'}
-							<div>
-								<div class=" self-center text-xs font-medium mb-1">
-									{$i18n.t('Perplexity API Key')}
-								</div>
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Perplexity API Key')}
+									</div>
 
-								<SensitiveInput
-									placeholder={$i18n.t('Enter Perplexity API Key')}
-									bind:value={webConfig.PERPLEXITY_API_KEY}
-								/>
+									<SensitiveInput
+										placeholder={$i18n.t('Enter Perplexity API Key')}
+										bind:value={webConfig.PERPLEXITY_API_KEY}
+									/>
+								</div>
+							</div>
+
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class="self-center text-xs font-medium mb-1">
+										{$i18n.t('Perplexity Model')}
+									</div>
+									<input
+										list="perplexity-model-list"
+										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+										bind:value={webConfig.PERPLEXITY_MODEL}
+									/>
+
+									<datalist id="perplexity-model-list">
+										<option value="sonar">Sonar</option>
+										<option value="sonar-pro">Sonar Pro</option>
+										<option value="sonar-reasoning">Sonar Reasoning</option>
+										<option value="sonar-reasoning-pro">Sonar Reasoning Pro</option>
+										<option value="sonar-deep-research">Sonar Deep Research</option>
+									</datalist>
+								</div>
+							</div>
+
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Perplexity Search Context Usage')}
+									</div>
+									<select
+										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+										bind:value={webConfig.PERPLEXITY_SEARCH_CONTEXT_USAGE}
+									>
+										<option value="low">Low</option>
+										<option value="medium">Medium</option>
+										<option value="high">High</option>
+									</select>
+								</div>
 							</div>
 						{:else if webConfig.WEB_SEARCH_ENGINE === 'sougou'}
 							<div class="mb-2.5 flex w-full flex-col">

src/lib/components/admin/Users.svelte

@@ -4,13 +4,32 @@
 
 	import { goto } from '$app/navigation';
 	import { user } from '$lib/stores';
+	import { page } from '$app/stores';
 
 	import UserList from './Users/UserList.svelte';
 	import Groups from './Users/Groups.svelte';
 
 	const i18n = getContext('i18n');
 
-	let selectedTab = 'overview';
+	let selectedTab;
+	$: {
+		const pathParts = $page.url.pathname.split('/');
+		const tabFromPath = pathParts[pathParts.length - 1];
+		selectedTab = ['overview', 'groups'].includes(tabFromPath) ? tabFromPath : 'overview';
+	}
+
+	$: if (selectedTab) {
+		// scroll to selectedTab
+		scrollToTab(selectedTab);
+	}
+
+	const scrollToTab = (tabId) => {
+		const tabElement = document.getElementById(tabId);
+		if (tabElement) {
+			tabElement.scrollIntoView({ behavior: 'smooth', block: 'nearest', inline: 'start' });
+		}
+	};
+
 	let loaded = false;
 
 	onMount(async () => {
@@ -30,6 +49,9 @@
 				}
 			});
 		}
+
+		// Scroll to the selected tab on mount
+		scrollToTab(selectedTab);
 	});
 </script>
 
@@ -39,12 +61,13 @@
 		class=" flex flex-row overflow-x-auto gap-2.5 max-w-full lg:gap-1 lg:flex-col lg:flex-none lg:w-40 dark:text-gray-200 text-sm font-medium text-left scrollbar-none"
 	>
 		<button
+			id="overview"
 			class="px-0.5 py-1 min-w-fit rounded-lg lg:flex-none flex text-right transition {selectedTab ===
 			'overview'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'overview';
+				goto('/admin/users/overview');
 			}}
 		>
 			<div class=" self-center mr-2">
@@ -63,12 +86,13 @@
 		</button>
 
 		<button
+			id="groups"
 			class="px-0.5 py-1 min-w-fit rounded-lg lg:flex-none flex text-right transition {selectedTab ===
 			'groups'
 				? ''
 				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
 			on:click={() => {
-				selectedTab = 'groups';
+				goto('/admin/users/groups');
 			}}
 		>
 			<div class=" self-center mr-2">

src/lib/components/admin/Users/UserList.svelte

@@ -52,27 +52,6 @@
 
 	let showUserChatsModal = false;
 	let showEditUserModal = false;
-	let showUpdateRoleModal = false;
-
-	const onUpdateRole = (user) => {
-		if (user.role === 'user') {
-			updateRoleHandler(user.id, 'admin');
-		} else if (user.role === 'pending') {
-			updateRoleHandler(user.id, 'user');
-		} else {
-			updateRoleHandler(user.id, 'pending');
-		}
-	};
-	const updateRoleHandler = async (id, role) => {
-		const res = await updateUserRole(localStorage.token, id, role).catch((error) => {
-			toast.error(`${error}`);
-			return null;
-		});
-
-		if (res) {
-			getUserList();
-		}
-	};
 
 	const deleteUserHandler = async (id) => {
 		const res = await deleteUserById(localStorage.token, id).catch((error) => {
@@ -133,21 +112,6 @@
 	}}
 />
 
-<RoleUpdateConfirmDialog
-	bind:show={showUpdateRoleModal}
-	on:confirm={() => {
-		onUpdateRole(selectedUser);
-	}}
-	message={$i18n.t(`Are you sure you want to update this user\'s role to **{{ROLE}}**?`, {
-		ROLE:
-			selectedUser?.role === 'user'
-				? 'admin'
-				: selectedUser?.role === 'pending'
-					? 'user'
-					: 'pending'
-	})}
-/>
-
 {#key selectedUser}
 	<EditUserModal
 		bind:show={showEditUserModal}
@@ -415,7 +379,7 @@
 								class=" translate-y-0.5"
 								on:click={() => {
 									selectedUser = user;
-									showUpdateRoleModal = true;
+									showEditUserModal = !showEditUserModal;
 								}}
 							>
 								<Badge

src/lib/components/admin/Users/UserList/EditUserModal.svelte

@@ -19,6 +19,7 @@
 
 	let _user = {
 		profile_image_url: '',
+		role: 'pending',
 		name: '',
 		email: '',
 		password: ''
@@ -95,6 +96,23 @@
 
 					<div class=" px-5 pt-3 pb-5">
 						<div class=" flex flex-col space-y-1.5">
+							<div class="flex flex-col w-full">
+								<div class=" mb-1 text-xs text-gray-500">{$i18n.t('Role')}</div>
+
+								<div class="flex-1">
+									<select
+										class="w-full dark:bg-gray-900 rounded-sm text-sm bg-transparent disabled:text-gray-500 dark:disabled:text-gray-500 outline-hidden"
+										bind:value={_user.role}
+										disabled={_user.id == sessionUser.id}
+										required
+									>
+										<option value="admin">{$i18n.t('Admin')}</option>
+										<option value="user">{$i18n.t('User')}</option>
+										<option value="pending">{$i18n.t('Pending')}</option>
+									</select>
+								</div>
+							</div>
+
 							<div class="flex flex-col w-full">
 								<div class=" mb-1 text-xs text-gray-500">{$i18n.t('Email')}</div>
 

src/lib/components/chat/Chat.svelte

@@ -239,6 +239,8 @@
 	};
 
 	const showMessage = async (message) => {
+		await tick();
+
 		const _chatId = JSON.parse(JSON.stringify($chatId));
 		let _messageId = JSON.parse(JSON.stringify(message.id));
 
@@ -298,6 +300,12 @@
 					message.content = data.content;
 				} else if (type === 'chat:message:files' || type === 'files') {
 					message.files = data.files;
+				} else if (type === 'chat:message:follow_ups') {
+					message.followUps = data.follow_ups;
+
+					if (autoScroll) {
+						scrollToBottom('smooth');
+					}
 				} else if (type === 'chat:title') {
 					chatTitle.set(data);
 					currentChatPage.set(1);
@@ -430,6 +438,12 @@
 		window.addEventListener('message', onMessageHandler);
 		$socket?.on('chat-events', chatEventHandler);
 
+		page.subscribe((page) => {
+			if (page.url.pathname === '/') {
+				initNewChat();
+			}
+		});
+
 		if (!$chatId) {
 			chatIdUnsubscriber = chatId.subscribe(async (value) => {
 				if (!value) {
@@ -774,6 +788,7 @@
 
 		autoScroll = true;
 
+		resetInput();
 		await chatId.set('');
 		await chatTitle.set('');
 
@@ -911,10 +926,13 @@
 		}
 	};
 
-	const scrollToBottom = async () => {
+	const scrollToBottom = async (behavior = 'auto') => {
 		await tick();
 		if (messagesContainerElement) {
-			messagesContainerElement.scrollTop = messagesContainerElement.scrollHeight;
+			messagesContainerElement.scrollTo({
+				top: messagesContainerElement.scrollHeight,
+				behavior
+			});
 		}
 	};
 	const chatCompletedHandler = async (chatId, modelId, responseMessageId, messages) => {
@@ -1624,9 +1642,6 @@
 				params: {
 					...$settings?.params,
 					...params,
-
-					format: $settings.requestFormat ?? undefined,
-					keep_alive: $settings.keepAlive ?? undefined,
 					stop:
 						(params?.stop ?? $settings?.params?.stop ?? undefined)
 							? (params?.stop.split(',').map((token) => token.trim()) ?? $settings.params.stop).map(
@@ -1676,19 +1691,20 @@
 				chat_id: $chatId,
 				id: responseMessageId,
 
-				...(!$temporaryChatEnabled &&
-				(messages.length == 1 ||
-					(messages.length == 2 &&
-						messages.at(0)?.role === 'system' &&
-						messages.at(1)?.role === 'user')) &&
-				(selectedModels[0] === model.id || atSelectedModel !== undefined)
-					? {
-							background_tasks: {
+				background_tasks: {
+					...(!$temporaryChatEnabled &&
+					(messages.length == 1 ||
+						(messages.length == 2 &&
+							messages.at(0)?.role === 'system' &&
+							messages.at(1)?.role === 'user')) &&
+					(selectedModels[0] === model.id || atSelectedModel !== undefined)
+						? {
 								title_generation: $settings?.title?.auto ?? true,
 								tags_generation: $settings?.autoTags ?? true
 							}
-						}
-					: {}),
+						: {}),
+					follow_up_generation: $settings?.autoFollowUps ?? true
+				},
 
 				...(stream && (model.info?.meta?.capabilities?.usage ?? false)
 					? {
@@ -2072,7 +2088,7 @@
 								</div>
 							</div>
 
-							<div class=" pb-[1rem]">
+							<div class=" pb-2">
 								<MessageInput
 									{history}
 									{taskIds}

src/lib/components/chat/ChatPlaceholder.svelte

@@ -69,10 +69,10 @@
 		{#if $temporaryChatEnabled}
 			<Tooltip
 				content={$i18n.t('This chat won’t appear in history and your messages will not be saved.')}
-				className="w-full flex justify-center mb-0.5"
+				className="w-full flex justify-start mb-0.5"
 				placement="top"
 			>
-				<div class="flex items-center gap-2 text-gray-500 font-medium text-lg my-2 w-fit">
+				<div class="flex items-center gap-2 text-gray-500 font-medium text-lg mt-2 w-fit">
 					<EyeSlash strokeWidth="2.5" className="size-5" />{$i18n.t('Temporary Chat')}
 				</div>
 			</Tooltip>

src/lib/components/chat/Controls/Controls.svelte

@@ -86,7 +86,7 @@
 			<Collapsible title={$i18n.t('Advanced Params')} open={true} buttonClassName="w-full">
 				<div class="text-sm mt-1.5" slot="content">
 					<div>
-						<AdvancedParams admin={$user?.role === 'admin'} bind:params />
+						<AdvancedParams admin={$user?.role === 'admin'} custom={true} bind:params />
 					</div>
 				</div>
 			</Collapsible>

src/lib/components/chat/Controls/Valves.svelte

@@ -190,7 +190,7 @@
 			</div>
 
 			{#if selectedId}
-				<hr class="dark:border-gray-800 my-1 w-full" />
+				<hr class="border-gray-50 dark:border-gray-800 my-1 w-full" />
 
 				<div class="my-2 text-xs">
 					{#if !loading}

src/lib/components/chat/MessageInput.svelte

@@ -1,5 +1,9 @@
 <script lang="ts">
+	import DOMPurify from 'dompurify';
+	import { marked } from 'marked';
+
 	import { toast } from 'svelte-sonner';
+
 	import { v4 as uuidv4 } from 'uuid';
 	import { createPicker, getAuthToken } from '$lib/utils/google-drive-picker';
 	import { pickAndDownloadFile } from '$lib/utils/onedrive-file-picker';
@@ -405,7 +409,6 @@
 
 	const onBlur = () => {
 		shiftKey = false;
-		selectedChatId = null;
 	};
 
 	onMount(async () => {
@@ -596,7 +599,7 @@
 						/>
 					{:else}
 						<form
-							class="w-full flex gap-1.5"
+							class="w-full flex flex-col gap-1.5"
 							on:submit|preventDefault={() => {
 								// check if selectedModels support image input
 								dispatch('submit', prompt);
@@ -702,7 +705,7 @@
 								<div class="px-2.5">
 									{#if $settings?.richTextInput ?? true}
 										<div
-											class="scrollbar-hidden rtl:text-right ltr:text-left bg-transparent dark:text-gray-100 outline-hidden w-full pt-2.5 pb-1 px-1 resize-none h-fit max-h-80 overflow-auto"
+											class="scrollbar-hidden rtl:text-right ltr:text-left bg-transparent dark:text-gray-100 outline-hidden w-full pt-2.5 pb-[5px] px-1 resize-none h-fit max-h-80 overflow-auto"
 											id="chat-input-container"
 										>
 											<RichTextInput
@@ -1521,6 +1524,14 @@
 									</div>
 								</div>
 							</div>
+
+							{#if $config?.license_metadata?.input_footer}
+								<div class=" text-xs text-gray-500 text-center line-clamp-1 marked">
+									{@html DOMPurify.sanitize(marked($config?.license_metadata?.input_footer))}
+								</div>
+							{:else}
+								<div class="mb-1" />
+							{/if}
 						</form>
 					{/if}
 				</div>

src/lib/components/chat/MessageInput/VoiceRecording.svelte

@@ -205,8 +205,10 @@
 			return;
 		}
 
+		const mineTypes = ['audio/webm; codecs=opus', 'audio/mp4'];
+
 		mediaRecorder = new MediaRecorder(stream, {
-			mimeType: 'audio/webm; codecs=opus'
+			mimeType: mineTypes.find((type) => MediaRecorder.isTypeSupported(type))
 		});
 
 		mediaRecorder.onstart = () => {

src/lib/components/chat/Messages/Markdown/HTMLToken.svelte

@@ -28,7 +28,7 @@
 			<!-- svelte-ignore a11y-media-has-caption -->
 			<video
 				class="w-full my-2"
-				src={videoSrc}
+				src={videoSrc.replaceAll('&amp;', '&')}
 				title="Video player"
 				frameborder="0"
 				referrerpolicy="strict-origin-when-cross-origin"
@@ -38,6 +38,20 @@
 		{:else}
 			{token.text}
 		{/if}
+	{:else if html && html.includes('<audio')}
+		{@const audio = html.match(/<audio[^>]*>([\s\S]*?)<\/audio>/)}
+		{@const audioSrc = audio && audio[1]}
+		{#if audioSrc}
+			<!-- svelte-ignore a11y-media-has-caption -->
+			<audio
+				class="w-full my-2"
+				src={audioSrc.replaceAll('&amp;', '&')}
+				title="Audio player"
+				controls
+			></audio>
+		{:else}
+			{token.text}
+		{/if}
 	{:else if token.text && token.text.match(/<iframe\s+[^>]*src="https:\/\/www\.youtube\.com\/embed\/([a-zA-Z0-9_-]{11})(?:\?[^"]*)?"[^>]*><\/iframe>/)}
 		{@const match = token.text.match(
 			/<iframe\s+[^>]*src="https:\/\/www\.youtube\.com\/embed\/([a-zA-Z0-9_-]{11})(?:\?[^"]*)?"[^>]*><\/iframe>/

src/lib/components/chat/Messages/ProfileImage.svelte

@@ -15,7 +15,7 @@
 			  src.startsWith('/')
 			? src
 			: `/user.png`}
-	class=" {className} object-cover rounded-full -translate-y-[1px]"
+	class=" {className} object-cover rounded-full"
 	alt="profile"
 	draggable="false"
 />

src/lib/components/chat/Messages/RateComment.svelte

@@ -137,8 +137,8 @@
 	</div>
 
 	<div class="w-full flex justify-center">
-		<div class=" relative w-fit">
-			<div class="mt-1.5 w-fit flex gap-1 pb-5">
+		<div class=" relative w-fit overflow-x-auto scrollbar-none">
+			<div class="mt-1.5 w-fit flex gap-1 pb-2">
 				<!-- 1-10 scale -->
 				{#each Array.from({ length: 10 }).map((_, i) => i + 1) as rating}
 					<button
@@ -156,7 +156,7 @@
 				{/each}
 			</div>
 
-			<div class="absolute bottom-0 left-0 right-0 flex justify-between text-xs">
+			<div class="sticky top-0 bottom-0 left-0 right-0 flex justify-between text-xs">
 				<div>
 					1 - {$i18n.t('Awful')}
 				</div>

src/lib/components/chat/Messages/ResponseMessage.svelte

@@ -48,6 +48,9 @@
 	import ContentRenderer from './ContentRenderer.svelte';
 	import { KokoroWorker } from '$lib/workers/KokoroWorker';
 	import FileItem from '$lib/components/common/FileItem.svelte';
+	import FollowUps from './ResponseMessage/FollowUps.svelte';
+	import { fade } from 'svelte/transition';
+	import { flyAndScale } from '$lib/utils/transitions';
 
 	interface MessageType {
 		id: string;
@@ -212,7 +215,7 @@
 
 		speaking = true;
 
-		const content = removeAllDetails(content);
+		const content = removeAllDetails(message.content);
 
 		if ($config.audio.tts.engine === '') {
 			let voices = [];
@@ -582,14 +585,6 @@
 			});
 		}
 	});
-
-	let screenReaderDiv: HTMLDivElement;
-
-	$: if (message.done) {
-		if (screenReaderDiv) {
-			screenReaderDiv.textContent = message.content;
-		}
-	}
 </script>
 
 <DeleteConfirmDialog
@@ -600,17 +595,13 @@
 	}}
 />
 
-<div bind:this={screenReaderDiv} aria-live="polite" class="sr-only">
-	{message.done ? message.content : ''}
-</div>
-
 {#key message.id}
 	<div
 		class=" flex w-full message-{message.id}"
 		id="message-{message.id}"
 		dir={$settings.chatDirection}
 	>
-		<div class={`shrink-0 ltr:mr-3 rtl:ml-3`}>
+		<div class={`shrink-0 ltr:mr-3 rtl:ml-3 hidden @lg:flex `}>
 			<ProfileImage
 				src={model?.info?.meta?.profile_image_url ??
 					($i18n.language === 'dg-DG' ? `/doge.png` : `${WEBUI_BASE_URL}/static/favicon.png`)}
@@ -618,7 +609,7 @@
 			/>
 		</div>
 
-		<div class="flex-auto w-0 pl-1">
+		<div class="flex-auto w-0 pl-1 relative -translate-y-0.5">
 			<Name>
 				<Tooltip content={model?.name ?? message.model} placement="top-start">
 					<span class="line-clamp-1 text-black dark:text-white">
@@ -878,12 +869,14 @@
 							{#if siblings.length > 1}
 								<div class="flex self-center min-w-fit" dir="ltr">
 									<button
+										aria-label={$i18n.t('Previous message')}
 										class="self-center p-1 hover:bg-black/5 dark:hover:bg-white/5 dark:hover:text-white hover:text-black rounded-md transition"
 										on:click={() => {
 											showPreviousMessage(message);
 										}}
 									>
 										<svg
+											aria-hidden="true"
 											xmlns="http://www.w3.org/2000/svg"
 											fill="none"
 											viewBox="0 0 24 24"
@@ -949,10 +942,12 @@
 										on:click={() => {
 											showNextMessage(message);
 										}}
+										aria-label={$i18n.t('Next message')}
 									>
 										<svg
 											xmlns="http://www.w3.org/2000/svg"
 											fill="none"
+											aria-hidden="true"
 											viewBox="0 0 24 24"
 											stroke="currentColor"
 											stroke-width="2.5"
@@ -973,6 +968,7 @@
 									{#if $user?.role === 'user' ? ($user?.permissions?.chat?.edit ?? true) : true}
 										<Tooltip content={$i18n.t('Edit')} placement="bottom">
 											<button
+												aria-label={$i18n.t('Edit')}
 												class="{isLastMessage
 													? 'visible'
 													: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg dark:hover:text-white hover:text-black transition"
@@ -985,6 +981,7 @@
 													fill="none"
 													viewBox="0 0 24 24"
 													stroke-width="2.3"
+													aria-hidden="true"
 													stroke="currentColor"
 													class="w-4 h-4"
 												>
@@ -1001,6 +998,7 @@
 
 								<Tooltip content={$i18n.t('Copy')} placement="bottom">
 									<button
+										aria-label={$i18n.t('Copy')}
 										class="{isLastMessage
 											? 'visible'
 											: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg dark:hover:text-white hover:text-black transition copy-response-button"
@@ -1011,6 +1009,7 @@
 										<svg
 											xmlns="http://www.w3.org/2000/svg"
 											fill="none"
+											aria-hidden="true"
 											viewBox="0 0 24 24"
 											stroke-width="2.3"
 											stroke="currentColor"
@@ -1028,6 +1027,7 @@
 								{#if $user?.role === 'admin' || ($user?.permissions?.chat?.tts ?? true)}
 									<Tooltip content={$i18n.t('Read Aloud')} placement="bottom">
 										<button
+											aria-label={$i18n.t('Read Aloud')}
 											id="speak-button-{message.id}"
 											class="{isLastMessage
 												? 'visible'
@@ -1043,6 +1043,7 @@
 													class=" w-4 h-4"
 													fill="currentColor"
 													viewBox="0 0 24 24"
+													aria-hidden="true"
 													xmlns="http://www.w3.org/2000/svg"
 												>
 													<style>
@@ -1075,6 +1076,7 @@
 													xmlns="http://www.w3.org/2000/svg"
 													fill="none"
 													viewBox="0 0 24 24"
+													aria-hidden="true"
 													stroke-width="2.3"
 													stroke="currentColor"
 													class="w-4 h-4"
@@ -1090,6 +1092,7 @@
 													xmlns="http://www.w3.org/2000/svg"
 													fill="none"
 													viewBox="0 0 24 24"
+													aria-hidden="true"
 													stroke-width="2.3"
 													stroke="currentColor"
 													class="w-4 h-4"
@@ -1108,6 +1111,7 @@
 								{#if $config?.features.enable_image_generation && ($user?.role === 'admin' || $user?.permissions?.features?.image_generation) && !readOnly}
 									<Tooltip content={$i18n.t('Generate Image')} placement="bottom">
 										<button
+											aria-label={$i18n.t('Generate Image')}
 											class="{isLastMessage
 												? 'visible'
 												: 'invisible group-hover:visible'}  p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg dark:hover:text-white hover:text-black transition"
@@ -1119,6 +1123,7 @@
 										>
 											{#if generatingImage}
 												<svg
+													aria-hidden="true"
 													class=" w-4 h-4"
 													fill="currentColor"
 													viewBox="0 0 24 24"
@@ -1153,6 +1158,7 @@
 												<svg
 													xmlns="http://www.w3.org/2000/svg"
 													fill="none"
+													aria-hidden="true"
 													viewBox="0 0 24 24"
 													stroke-width="2.3"
 													stroke="currentColor"
@@ -1185,6 +1191,7 @@
 										placement="bottom"
 									>
 										<button
+											aria-hidden="true"
 											class=" {isLastMessage
 												? 'visible'
 												: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg dark:hover:text-white hover:text-black transition whitespace-pre-wrap"
@@ -1194,6 +1201,7 @@
 											id="info-{message.id}"
 										>
 											<svg
+												aria-hidden="true"
 												xmlns="http://www.w3.org/2000/svg"
 												fill="none"
 												viewBox="0 0 24 24"
@@ -1215,6 +1223,7 @@
 									{#if !$temporaryChatEnabled && ($config?.features.enable_message_rating ?? true)}
 										<Tooltip content={$i18n.t('Good Response')} placement="bottom">
 											<button
+												aria-label={$i18n.t('Good Response')}
 												class="{isLastMessage
 													? 'visible'
 													: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg {(
@@ -1233,6 +1242,7 @@
 												}}
 											>
 												<svg
+													aria-hidden="true"
 													stroke="currentColor"
 													fill="none"
 													stroke-width="2.3"
@@ -1251,6 +1261,7 @@
 
 										<Tooltip content={$i18n.t('Bad Response')} placement="bottom">
 											<button
+												aria-label={$i18n.t('Bad Response')}
 												class="{isLastMessage
 													? 'visible'
 													: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg {(
@@ -1269,6 +1280,7 @@
 												}}
 											>
 												<svg
+													aria-hidden="true"
 													stroke="currentColor"
 													fill="none"
 													stroke-width="2.3"
@@ -1289,6 +1301,7 @@
 									{#if isLastMessage}
 										<Tooltip content={$i18n.t('Continue Response')} placement="bottom">
 											<button
+												aria-label={$i18n.t('Continue Response')}
 												type="button"
 												id="continue-response-button"
 												class="{isLastMessage
@@ -1299,6 +1312,7 @@
 												}}
 											>
 												<svg
+													aria-hidden="true"
 													xmlns="http://www.w3.org/2000/svg"
 													fill="none"
 													viewBox="0 0 24 24"
@@ -1324,6 +1338,7 @@
 									<Tooltip content={$i18n.t('Regenerate')} placement="bottom">
 										<button
 											type="button"
+											aria-label={$i18n.t('Regenerate')}
 											class="{isLastMessage
 												? 'visible'
 												: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg dark:hover:text-white hover:text-black transition regenerate-response-button"
@@ -1349,6 +1364,7 @@
 												fill="none"
 												viewBox="0 0 24 24"
 												stroke-width="2.3"
+												aria-hidden="true"
 												stroke="currentColor"
 												class="w-4 h-4"
 											>
@@ -1365,6 +1381,7 @@
 										<Tooltip content={$i18n.t('Delete')} placement="bottom">
 											<button
 												type="button"
+												aria-label={$i18n.t('Delete')}
 												id="delete-response-button"
 												class="{isLastMessage
 													? 'visible'
@@ -1379,6 +1396,7 @@
 													viewBox="0 0 24 24"
 													stroke-width="2"
 													stroke="currentColor"
+													aria-hidden="true"
 													class="w-4 h-4"
 												>
 													<path
@@ -1396,6 +1414,7 @@
 											<Tooltip content={action.name} placement="bottom">
 												<button
 													type="button"
+													aria-label={action.name}
 													class="{isLastMessage
 														? 'visible'
 														: 'invisible group-hover:visible'} p-1.5 hover:bg-black/5 dark:hover:bg-white/5 rounded-lg dark:hover:text-white hover:text-black transition"
@@ -1437,8 +1456,25 @@
 							}}
 						/>
 					{/if}
+
+					{#if isLastMessage && message.done && !readOnly && (message?.followUps ?? []).length > 0}
+						<div class="mt-2.5" in:fade={{ duration: 100 }}>
+							<FollowUps
+								followUps={message?.followUps}
+								onClick={(prompt) => {
+									submitMessage(message?.id, prompt);
+								}}
+							/>
+						</div>
+					{/if}
 				{/if}
 			</div>
+
+			{#if message?.done}
+				<div aria-live="polite" class="sr-only">
+					{message?.content ?? ''}
+				</div>
+			{/if}
 		</div>
 	</div>
 {/key}

src/lib/components/chat/Messages/ResponseMessage/FollowUps.svelte

@@ -0,0 +1,36 @@
+<script lang="ts">
+	import ArrowTurnDownRight from '$lib/components/icons/ArrowTurnDownRight.svelte';
+	import { onMount, tick, getContext } from 'svelte';
+
+	const i18n = getContext('i18n');
+
+	export let followUps: string[] = [];
+	export let onClick: (followUp: string) => void = () => {};
+</script>
+
+<div class="mt-4">
+	<div class="text-sm font-medium">
+		{$i18n.t('Follow up')}
+	</div>
+
+	<div class="flex flex-col text-left gap-1 mt-1.5">
+		{#each followUps as followUp, idx (idx)}
+			<button
+				class=" mr-2 py-1.5 bg-transparent text-left text-sm flex items-center gap-2 px-1.5 text-gray-500 dark:text-gray-400 hover:text-black dark:hover:text-white transition"
+				on:click={() => onClick(followUp)}
+				title={followUp}
+				aria-label={followUp}
+			>
+				<ArrowTurnDownRight className="size-3.5" />
+
+				<div class="line-clamp-1">
+					{followUp}
+				</div>
+			</button>
+
+			{#if idx < followUps.length - 1}
+				<hr class="border-gray-100 dark:border-gray-850" />
+			{/if}
+		{/each}
+	</div>
+</div>

src/lib/components/chat/ModelSelector.svelte

@@ -25,6 +25,19 @@
 		toast.success($i18n.t('Default model updated'));
 	};
 
+	const pinModelHandler = async (modelId) => {
+		let pinnedModels = $settings?.pinnedModels ?? [];
+
+		if (pinnedModels.includes(modelId)) {
+			pinnedModels = pinnedModels.filter((id) => id !== modelId);
+		} else {
+			pinnedModels = [...new Set([...pinnedModels, modelId])];
+		}
+
+		settings.set({ ...$settings, pinnedModels: pinnedModels });
+		await updateUserSettings(localStorage.token, { ui: $settings });
+	};
+
 	$: if (selectedModels.length > 0 && $models.length > 0) {
 		selectedModels = selectedModels.map((model) =>
 			$models.map((m) => m.id).includes(model) ? model : ''
@@ -49,6 +62,7 @@
 							? ($user?.permissions?.chat?.temporary ?? true) &&
 								!($user?.permissions?.chat?.temporary_enforced ?? false)
 							: true}
+						{pinModelHandler}
 						bind:value={selectedModel}
 					/>
 				</div>

src/lib/components/chat/ModelSelector/ModelItem.svelte

@@ -0,0 +1,250 @@
+<script lang="ts">
+	import { marked } from 'marked';
+
+	import { getContext, tick } from 'svelte';
+	import dayjs from '$lib/dayjs';
+
+	import { mobile, settings, user } from '$lib/stores';
+
+	import Tooltip from '$lib/components/common/Tooltip.svelte';
+	import { copyToClipboard, sanitizeResponseContent } from '$lib/utils';
+	import ArrowUpTray from '$lib/components/icons/ArrowUpTray.svelte';
+	import Check from '$lib/components/icons/Check.svelte';
+	import ModelItemMenu from './ModelItemMenu.svelte';
+	import EllipsisHorizontal from '$lib/components/icons/EllipsisHorizontal.svelte';
+	import { toast } from 'svelte-sonner';
+
+	const i18n = getContext('i18n');
+
+	export let selectedModelIdx: number = -1;
+	export let item: any = {};
+	export let index: number = -1;
+	export let value: string = '';
+
+	export let unloadModelHandler: (modelValue: string) => void = () => {};
+	export let pinModelHandler: (modelId: string) => void = () => {};
+
+	export let onClick: () => void = () => {};
+
+	const copyLinkHandler = async (model) => {
+		const baseUrl = window.location.origin;
+		const res = await copyToClipboard(`${baseUrl}/?model=${encodeURIComponent(model.id)}`);
+
+		if (res) {
+			toast.success($i18n.t('Copied link to clipboard'));
+		} else {
+			toast.error($i18n.t('Failed to copy link'));
+		}
+	};
+
+	let showMenu = false;
+</script>
+
+<button
+	aria-label="model-item"
+	class="flex group/item w-full text-left font-medium line-clamp-1 select-none items-center rounded-button py-2 pl-3 pr-1.5 text-sm text-gray-700 dark:text-gray-100 outline-hidden transition-all duration-75 hover:bg-gray-100 dark:hover:bg-gray-800 rounded-lg cursor-pointer data-highlighted:bg-muted {index ===
+	selectedModelIdx
+		? 'bg-gray-100 dark:bg-gray-800 group-hover:bg-transparent'
+		: ''}"
+	data-arrow-selected={index === selectedModelIdx}
+	data-value={item.value}
+	on:click={() => {
+		onClick();
+	}}
+>
+	<div class="flex flex-col flex-1 gap-1.5">
+		{#if (item?.model?.tags ?? []).length > 0}
+			<div
+				class="flex gap-0.5 self-center items-start h-full w-full translate-y-[0.5px] overflow-x-auto scrollbar-none"
+			>
+				{#each item.model?.tags.sort((a, b) => a.name.localeCompare(b.name)) as tag}
+					<Tooltip content={tag.name} className="flex-shrink-0">
+						<div
+							class=" text-xs font-bold px-1 rounded-sm uppercase bg-gray-500/20 text-gray-700 dark:text-gray-200"
+						>
+							{tag.name}
+						</div>
+					</Tooltip>
+				{/each}
+			</div>
+		{/if}
+
+		<div class="flex items-center gap-2">
+			<div class="flex items-center min-w-fit">
+				<Tooltip content={$user?.role === 'admin' ? (item?.value ?? '') : ''} placement="top-start">
+					<img
+						src={item.model?.info?.meta?.profile_image_url ?? '/static/favicon.png'}
+						alt="Model"
+						class="rounded-full size-5 flex items-center"
+					/>
+				</Tooltip>
+			</div>
+
+			<div class="flex items-center">
+				<Tooltip content={`${item.label} (${item.value})`} placement="top-start">
+					<div class="line-clamp-1">
+						{item.label}
+					</div>
+				</Tooltip>
+			</div>
+
+			<div class=" shrink-0 flex items-center gap-2">
+				{#if item.model.owned_by === 'ollama'}
+					{#if (item.model.ollama?.details?.parameter_size ?? '') !== ''}
+						<div class="flex items-center translate-y-[0.5px]">
+							<Tooltip
+								content={`${
+									item.model.ollama?.details?.quantization_level
+										? item.model.ollama?.details?.quantization_level + ' '
+										: ''
+								}${
+									item.model.ollama?.size
+										? `(${(item.model.ollama?.size / 1024 ** 3).toFixed(1)}GB)`
+										: ''
+								}`}
+								className="self-end"
+							>
+								<span class=" text-xs font-medium text-gray-600 dark:text-gray-400 line-clamp-1"
+									>{item.model.ollama?.details?.parameter_size ?? ''}</span
+								>
+							</Tooltip>
+						</div>
+					{/if}
+					{#if item.model.ollama?.expires_at && new Date(item.model.ollama?.expires_at * 1000) > new Date()}
+						<div class="flex items-center translate-y-[0.5px] px-0.5">
+							<Tooltip
+								content={`${$i18n.t('Unloads {{FROM_NOW}}', {
+									FROM_NOW: dayjs(item.model.ollama?.expires_at * 1000).fromNow()
+								})}`}
+								className="self-end"
+							>
+								<div class=" flex items-center">
+									<span class="relative flex size-2">
+										<span
+											class="animate-ping absolute inline-flex h-full w-full rounded-full bg-green-400 opacity-75"
+										/>
+										<span class="relative inline-flex rounded-full size-2 bg-green-500" />
+									</span>
+								</div>
+							</Tooltip>
+						</div>
+					{/if}
+				{/if}
+
+				<!-- {JSON.stringify(item.info)} -->
+
+				{#if item.model?.direct}
+					<Tooltip content={`${$i18n.t('Direct')}`}>
+						<div class="translate-y-[1px]">
+							<svg
+								xmlns="http://www.w3.org/2000/svg"
+								viewBox="0 0 16 16"
+								fill="currentColor"
+								class="size-3"
+							>
+								<path
+									fill-rule="evenodd"
+									d="M2 2.75A.75.75 0 0 1 2.75 2C8.963 2 14 7.037 14 13.25a.75.75 0 0 1-1.5 0c0-5.385-4.365-9.75-9.75-9.75A.75.75 0 0 1 2 2.75Zm0 4.5a.75.75 0 0 1 .75-.75 6.75 6.75 0 0 1 6.75 6.75.75.75 0 0 1-1.5 0C8 10.35 5.65 8 2.75 8A.75.75 0 0 1 2 7.25ZM3.5 11a1.5 1.5 0 1 0 0 3 1.5 1.5 0 0 0 0-3Z"
+									clip-rule="evenodd"
+								/>
+							</svg>
+						</div>
+					</Tooltip>
+				{:else if item.model.connection_type === 'external'}
+					<Tooltip content={`${$i18n.t('External')}`}>
+						<div class="translate-y-[1px]">
+							<svg
+								xmlns="http://www.w3.org/2000/svg"
+								viewBox="0 0 16 16"
+								fill="currentColor"
+								class="size-3"
+							>
+								<path
+									fill-rule="evenodd"
+									d="M8.914 6.025a.75.75 0 0 1 1.06 0 3.5 3.5 0 0 1 0 4.95l-2 2a3.5 3.5 0 0 1-5.396-4.402.75.75 0 0 1 1.251.827 2 2 0 0 0 3.085 2.514l2-2a2 2 0 0 0 0-2.828.75.75 0 0 1 0-1.06Z"
+									clip-rule="evenodd"
+								/>
+								<path
+									fill-rule="evenodd"
+									d="M7.086 9.975a.75.75 0 0 1-1.06 0 3.5 3.5 0 0 1 0-4.95l2-2a3.5 3.5 0 0 1 5.396 4.402.75.75 0 0 1-1.251-.827 2 2 0 0 0-3.085-2.514l-2 2a2 2 0 0 0 0 2.828.75.75 0 0 1 0 1.06Z"
+									clip-rule="evenodd"
+								/>
+							</svg>
+						</div>
+					</Tooltip>
+				{/if}
+
+				{#if item.model?.info?.meta?.description}
+					<Tooltip
+						content={`${marked.parse(
+							sanitizeResponseContent(item.model?.info?.meta?.description).replaceAll('\n', '<br>')
+						)}`}
+					>
+						<div class=" translate-y-[1px]">
+							<svg
+								xmlns="http://www.w3.org/2000/svg"
+								fill="none"
+								viewBox="0 0 24 24"
+								stroke-width="1.5"
+								stroke="currentColor"
+								class="w-4 h-4"
+							>
+								<path
+									stroke-linecap="round"
+									stroke-linejoin="round"
+									d="m11.25 11.25.041-.02a.75.75 0 0 1 1.063.852l-.708 2.836a.75.75 0 0 0 1.063.853l.041-.021M21 12a9 9 0 1 1-18 0 9 9 0 0 1 18 0Zm-9-3.75h.008v.008H12V8.25Z"
+								/>
+							</svg>
+						</div>
+					</Tooltip>
+				{/if}
+			</div>
+		</div>
+	</div>
+
+	<div class="ml-auto pl-2 pr-1 flex items-center gap-1.5 shrink-0">
+		{#if $user?.role === 'admin' && item.model.owned_by === 'ollama' && item.model.ollama?.expires_at && new Date(item.model.ollama?.expires_at * 1000) > new Date()}
+			<Tooltip
+				content={`${$i18n.t('Eject')}`}
+				className="flex-shrink-0 group-hover/item:opacity-100 opacity-0 "
+			>
+				<button
+					class="flex"
+					on:click={(e) => {
+						e.preventDefault();
+						e.stopPropagation();
+						unloadModelHandler(item.value);
+					}}
+				>
+					<ArrowUpTray className="size-3" />
+				</button>
+			</Tooltip>
+		{/if}
+
+		<ModelItemMenu
+			bind:show={showMenu}
+			model={item.model}
+			{pinModelHandler}
+			copyLinkHandler={() => {
+				copyLinkHandler(item.model);
+			}}
+		>
+			<button
+				class="flex"
+				on:click={(e) => {
+					e.preventDefault();
+					e.stopPropagation();
+					showMenu = !showMenu;
+				}}
+			>
+				<EllipsisHorizontal />
+			</button>
+		</ModelItemMenu>
+
+		{#if value === item.value}
+			<div>
+				<Check className="size-3" />
+			</div>
+		{/if}
+	</div>
+</button>

src/lib/components/chat/ModelSelector/ModelItemMenu.svelte

@@ -0,0 +1,90 @@
+<script lang="ts">
+	import { DropdownMenu } from 'bits-ui';
+	import { flyAndScale } from '$lib/utils/transitions';
+
+	import { getContext } from 'svelte';
+
+	import Tooltip from '$lib/components/common/Tooltip.svelte';
+	import Link from '$lib/components/icons/Link.svelte';
+	import Eye from '$lib/components/icons/Eye.svelte';
+	import EyeSlash from '$lib/components/icons/EyeSlash.svelte';
+	import { settings } from '$lib/stores';
+
+	const i18n = getContext('i18n');
+
+	export let show = false;
+	export let model;
+
+	export let pinModelHandler: (modelId: string) => void = () => {};
+	export let copyLinkHandler: Function = () => {};
+
+	export let onClose: Function = () => {};
+</script>
+
+<DropdownMenu.Root
+	bind:open={show}
+	closeFocus={false}
+	onOpenChange={(state) => {
+		if (state === false) {
+			onClose();
+		}
+	}}
+	typeahead={false}
+>
+	<DropdownMenu.Trigger>
+		<Tooltip content={$i18n.t('More')} className=" group-hover/item:opacity-100  opacity-0">
+			<slot />
+		</Tooltip>
+	</DropdownMenu.Trigger>
+
+	<DropdownMenu.Content
+		strategy="fixed"
+		class="w-full max-w-[180px] text-sm rounded-xl px-1 py-1.5 z-[9999999] bg-white dark:bg-gray-850 dark:text-white shadow-lg"
+		sideOffset={-2}
+		side="bottom"
+		align="end"
+		transition={flyAndScale}
+	>
+		<button
+			type="button"
+			class="flex rounded-md py-1.5 px-3 w-full hover:bg-gray-50 dark:hover:bg-gray-800 transition items-center gap-2"
+			on:click={(e) => {
+				e.stopPropagation();
+				e.preventDefault();
+
+				pinModelHandler(model?.id);
+				show = false;
+			}}
+		>
+			{#if ($settings?.pinnedModels ?? []).includes(model?.id)}
+				<EyeSlash />
+			{:else}
+				<Eye />
+			{/if}
+
+			<div class="flex items-center">
+				{#if ($settings?.pinnedModels ?? []).includes(model?.id)}
+					{$i18n.t('Hide from Sidebar')}
+				{:else}
+					{$i18n.t('Keep in Sidebar')}
+				{/if}
+			</div>
+		</button>
+
+		<button
+			type="button"
+			class="flex rounded-md py-1.5 px-3 w-full hover:bg-gray-50 dark:hover:bg-gray-800 transition items-center gap-2"
+			on:click={(e) => {
+				e.stopPropagation();
+				e.preventDefault();
+
+				copyLinkHandler();
+				show = false;
+			}}
+		>
+			<Link />
+
+			<div class="flex items-center">{$i18n.t('Copy Link')}</div>
+		</button>
+	</DropdownMenu.Content>
+</DropdownMenu.Root>

src/lib/components/chat/ModelSelector/Selector.svelte

@@ -3,12 +3,13 @@
 	import { marked } from 'marked';
 	import Fuse from 'fuse.js';
 
+	import dayjs from '$lib/dayjs';
+	import relativeTime from 'dayjs/plugin/relativeTime';
+	dayjs.extend(relativeTime);
+
 	import { flyAndScale } from '$lib/utils/transitions';
 	import { createEventDispatcher, onMount, getContext, tick } from 'svelte';
-
-	import ChevronDown from '$lib/components/icons/ChevronDown.svelte';
-	import Check from '$lib/components/icons/Check.svelte';
-	import Search from '$lib/components/icons/Search.svelte';
+	import { goto } from '$app/navigation';
 
 	import { deleteModel, getOllamaVersion, pullModel, unloadModel } from '$lib/apis/ollama';
 
@@ -25,14 +26,14 @@
 	import { capitalizeFirstLetter, sanitizeResponseContent, splitStream } from '$lib/utils';
 	import { getModels } from '$lib/apis';
 
+	import ChevronDown from '$lib/components/icons/ChevronDown.svelte';
+	import Check from '$lib/components/icons/Check.svelte';
+	import Search from '$lib/components/icons/Search.svelte';
 	import Tooltip from '$lib/components/common/Tooltip.svelte';
 	import Switch from '$lib/components/common/Switch.svelte';
 	import ChatBubbleOval from '$lib/components/icons/ChatBubbleOval.svelte';
-	import { goto } from '$app/navigation';
-	import dayjs from '$lib/dayjs';
-	import relativeTime from 'dayjs/plugin/relativeTime';
-	import ArrowUpTray from '$lib/components/icons/ArrowUpTray.svelte';
-	dayjs.extend(relativeTime);
+
+	import ModelItem from './ModelItem.svelte';
 
 	const i18n = getContext('i18n');
 	const dispatch = createEventDispatcher();
@@ -56,6 +57,8 @@
 	export let className = 'w-[32rem]';
 	export let triggerClassName = 'text-lg';
 
+	export let pinModelHandler: (modelId: string) => void = () => {};
+
 	let tagsContainerElement;
 
 	let show = false;
@@ -407,10 +410,10 @@
 				</div>
 			{/if}
 
-			<div class="px-3 max-h-64 overflow-y-auto scrollbar-hidden group relative">
+			<div class="px-3">
 				{#if tags && items.filter((item) => !(item.model?.info?.meta?.hidden ?? false)).length > 0}
 					<div
-						class=" flex w-full sticky top-0 z-10 bg-white dark:bg-gray-850 overflow-x-auto scrollbar-none"
+						class=" flex w-full bg-white dark:bg-gray-850 overflow-x-auto scrollbar-none"
 						on:wheel={(e) => {
 							if (e.deltaY !== 0) {
 								e.preventDefault();
@@ -492,212 +495,24 @@
 						</div>
 					</div>
 				{/if}
+			</div>
 
+			<div class="px-3 max-h-64 overflow-y-auto group relative">
 				{#each filteredItems as item, index}
-					<button
-						aria-label="model-item"
-						class="flex w-full text-left font-medium line-clamp-1 select-none items-center rounded-button py-2 pl-3 pr-1.5 text-sm text-gray-700 dark:text-gray-100 outline-hidden transition-all duration-75 hover:bg-gray-100 dark:hover:bg-gray-800 rounded-lg cursor-pointer data-highlighted:bg-muted {index ===
-						selectedModelIdx
-							? 'bg-gray-100 dark:bg-gray-800 group-hover:bg-transparent'
-							: ''}"
-						data-arrow-selected={index === selectedModelIdx}
-						data-value={item.value}
-						on:click={() => {
+					<ModelItem
+						{selectedModelIdx}
+						{item}
+						{index}
+						{value}
+						{pinModelHandler}
+						{unloadModelHandler}
+						onClick={() => {
 							value = item.value;
 							selectedModelIdx = index;
 
 							show = false;
 						}}
-					>
-						<div class="flex flex-col">
-							{#if $mobile && (item?.model?.tags ?? []).length > 0}
-								<div class="flex gap-0.5 self-start h-full mb-1.5 -translate-x-1">
-									{#each item.model?.tags.sort((a, b) => a.name.localeCompare(b.name)) as tag}
-										<div
-											class=" text-xs font-bold px-1 rounded-sm uppercase line-clamp-1 bg-gray-500/20 text-gray-700 dark:text-gray-200"
-										>
-											{tag.name}
-										</div>
-									{/each}
-								</div>
-							{/if}
-							<div class="flex items-center gap-2">
-								<div class="flex items-center min-w-fit">
-									<div class="line-clamp-1">
-										<div class="flex items-center min-w-fit">
-											<Tooltip
-												content={$user?.role === 'admin' ? (item?.value ?? '') : ''}
-												placement="top-start"
-											>
-												<img
-													src={item.model?.info?.meta?.profile_image_url ?? '/static/favicon.png'}
-													alt="Model"
-													class="rounded-full size-5 flex items-center mr-2"
-												/>
-
-												<div class="flex items-center line-clamp-1">
-													<div class="line-clamp-1">
-														{item.label}
-													</div>
-												</div>
-											</Tooltip>
-										</div>
-									</div>
-								</div>
-
-								{#if item.model.owned_by === 'ollama'}
-									{#if (item.model.ollama?.details?.parameter_size ?? '') !== ''}
-										<div class="flex items-center translate-y-[0.5px]">
-											<Tooltip
-												content={`${
-													item.model.ollama?.details?.quantization_level
-														? item.model.ollama?.details?.quantization_level + ' '
-														: ''
-												}${
-													item.model.ollama?.size
-														? `(${(item.model.ollama?.size / 1024 ** 3).toFixed(1)}GB)`
-														: ''
-												}`}
-												className="self-end"
-											>
-												<span
-													class=" text-xs font-medium text-gray-600 dark:text-gray-400 line-clamp-1"
-													>{item.model.ollama?.details?.parameter_size ?? ''}</span
-												>
-											</Tooltip>
-										</div>
-									{/if}
-									{#if item.model.ollama?.expires_at && new Date(item.model.ollama?.expires_at * 1000) > new Date()}
-										<div class="flex items-center translate-y-[0.5px] px-0.5">
-											<Tooltip
-												content={`${$i18n.t('Unloads {{FROM_NOW}}', {
-													FROM_NOW: dayjs(item.model.ollama?.expires_at * 1000).fromNow()
-												})}`}
-												className="self-end"
-											>
-												<div class=" flex items-center">
-													<span class="relative flex size-2">
-														<span
-															class="animate-ping absolute inline-flex h-full w-full rounded-full bg-green-400 opacity-75"
-														/>
-														<span class="relative inline-flex rounded-full size-2 bg-green-500" />
-													</span>
-												</div>
-											</Tooltip>
-										</div>
-									{/if}
-								{/if}
-
-								<!-- {JSON.stringify(item.info)} -->
-
-								{#if item.model?.direct}
-									<Tooltip content={`${$i18n.t('Direct')}`}>
-										<div class="translate-y-[1px]">
-											<svg
-												xmlns="http://www.w3.org/2000/svg"
-												viewBox="0 0 16 16"
-												fill="currentColor"
-												class="size-3"
-											>
-												<path
-													fill-rule="evenodd"
-													d="M2 2.75A.75.75 0 0 1 2.75 2C8.963 2 14 7.037 14 13.25a.75.75 0 0 1-1.5 0c0-5.385-4.365-9.75-9.75-9.75A.75.75 0 0 1 2 2.75Zm0 4.5a.75.75 0 0 1 .75-.75 6.75 6.75 0 0 1 6.75 6.75.75.75 0 0 1-1.5 0C8 10.35 5.65 8 2.75 8A.75.75 0 0 1 2 7.25ZM3.5 11a1.5 1.5 0 1 0 0 3 1.5 1.5 0 0 0 0-3Z"
-													clip-rule="evenodd"
-												/>
-											</svg>
-										</div>
-									</Tooltip>
-								{:else if item.model.connection_type === 'external'}
-									<Tooltip content={`${$i18n.t('External')}`}>
-										<div class="translate-y-[1px]">
-											<svg
-												xmlns="http://www.w3.org/2000/svg"
-												viewBox="0 0 16 16"
-												fill="currentColor"
-												class="size-3"
-											>
-												<path
-													fill-rule="evenodd"
-													d="M8.914 6.025a.75.75 0 0 1 1.06 0 3.5 3.5 0 0 1 0 4.95l-2 2a3.5 3.5 0 0 1-5.396-4.402.75.75 0 0 1 1.251.827 2 2 0 0 0 3.085 2.514l2-2a2 2 0 0 0 0-2.828.75.75 0 0 1 0-1.06Z"
-													clip-rule="evenodd"
-												/>
-												<path
-													fill-rule="evenodd"
-													d="M7.086 9.975a.75.75 0 0 1-1.06 0 3.5 3.5 0 0 1 0-4.95l2-2a3.5 3.5 0 0 1 5.396 4.402.75.75 0 0 1-1.251-.827 2 2 0 0 0-3.085-2.514l-2 2a2 2 0 0 0 0 2.828.75.75 0 0 1 0 1.06Z"
-													clip-rule="evenodd"
-												/>
-											</svg>
-										</div>
-									</Tooltip>
-								{/if}
-
-								{#if item.model?.info?.meta?.description}
-									<Tooltip
-										content={`${marked.parse(
-											sanitizeResponseContent(item.model?.info?.meta?.description).replaceAll(
-												'\n',
-												'<br>'
-											)
-										)}`}
-									>
-										<div class=" translate-y-[1px]">
-											<svg
-												xmlns="http://www.w3.org/2000/svg"
-												fill="none"
-												viewBox="0 0 24 24"
-												stroke-width="1.5"
-												stroke="currentColor"
-												class="w-4 h-4"
-											>
-												<path
-													stroke-linecap="round"
-													stroke-linejoin="round"
-													d="m11.25 11.25.041-.02a.75.75 0 0 1 1.063.852l-.708 2.836a.75.75 0 0 0 1.063.853l.041-.021M21 12a9 9 0 1 1-18 0 9 9 0 0 1 18 0Zm-9-3.75h.008v.008H12V8.25Z"
-												/>
-											</svg>
-										</div>
-									</Tooltip>
-								{/if}
-
-								{#if !$mobile && (item?.model?.tags ?? []).length > 0}
-									<div
-										class="flex gap-0.5 self-center items-center h-full translate-y-[0.5px] overflow-x-auto scrollbar-none"
-									>
-										{#each item.model?.tags.sort((a, b) => a.name.localeCompare(b.name)) as tag}
-											<Tooltip content={tag.name} className="flex-shrink-0">
-												<div
-													class=" text-xs font-bold px-1 rounded-sm uppercase bg-gray-500/20 text-gray-700 dark:text-gray-200"
-												>
-													{tag.name}
-												</div>
-											</Tooltip>
-										{/each}
-									</div>
-								{/if}
-							</div>
-						</div>
-
-						<div class="ml-auto pl-2 pr-1 flex gap-1.5 items-center">
-							{#if $user?.role === 'admin' && item.model.owned_by === 'ollama' && item.model.ollama?.expires_at && new Date(item.model.ollama?.expires_at * 1000) > new Date()}
-								<Tooltip content={`${$i18n.t('Eject')}`} className="flex-shrink-0">
-									<button
-										class="flex"
-										on:click={() => {
-											unloadModelHandler(item.value);
-										}}
-									>
-										<ArrowUpTray className="size-3" />
-									</button>
-								</Tooltip>
-							{/if}
-
-							{#if value === item.value}
-								<div>
-									<Check className="size-3" />
-								</div>
-							{/if}
-						</div>
-					</button>
+					/>
 				{:else}
 					<div class="">
 						<div class="block px-3 py-2 text-sm text-gray-700 dark:text-gray-100">

src/lib/components/chat/Settings/Advanced/AdvancedParams.svelte

@@ -1,15 +1,18 @@
 <script lang="ts">
 	import Switch from '$lib/components/common/Switch.svelte';
+	import Textarea from '$lib/components/common/Textarea.svelte';
 	import Tooltip from '$lib/components/common/Tooltip.svelte';
-	import { getContext, createEventDispatcher } from 'svelte';
-
-	const dispatch = createEventDispatcher();
+	import Plus from '$lib/components/icons/Plus.svelte';
+	import { getContext } from 'svelte';
 
 	const i18n = getContext('i18n');
 
-	export let admin = false;
+	export let onChange: (params: any) => void = () => {};
 
-	export let params = {
+	export let admin = false;
+	export let custom = false;
+
+	const defaultParams = {
 		// Advanced
 		stream_response: null, // Set stream responses for this model individually
 		function_calling: null,
@@ -18,31 +21,33 @@
 		temperature: null,
 		reasoning_effort: null,
 		logit_bias: null,
-		frequency_penalty: null,
-		repeat_last_n: null,
-		mirostat: null,
-		mirostat_eta: null,
-		mirostat_tau: null,
+		max_tokens: null,
 		top_k: null,
 		top_p: null,
 		min_p: null,
+		frequency_penalty: null,
+		presence_penalty: null,
+		mirostat: null,
+		mirostat_eta: null,
+		mirostat_tau: null,
+		repeat_last_n: null,
 		tfs_z: null,
-		num_ctx: null,
-		num_batch: null,
-		num_keep: null,
-		max_tokens: null,
+		repeat_penalty: null,
 		use_mmap: null,
 		use_mlock: null,
+		think: null,
+		format: null,
+		keep_alive: null,
+		num_keep: null,
+		num_ctx: null,
+		num_batch: null,
 		num_thread: null,
-		num_gpu: null,
-		template: null
+		num_gpu: null
 	};
 
-	let customFieldName = '';
-	let customFieldValue = '';
-
+	export let params = defaultParams;
 	$: if (params) {
-		dispatch('change', params);
+		onChange(params);
 	}
 </script>
 
@@ -145,7 +150,7 @@
 			<div class="flex mt-0.5 space-x-2">
 				<div class=" flex-1">
 					<input
-						class="w-full rounded-lg py-2 px-4 text-sm dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+						class="text-sm w-full bg-transparent outline-hidden outline-none"
 						type="number"
 						placeholder={$i18n.t('Enter Seed')}
 						bind:value={params.seed}
@@ -190,7 +195,7 @@
 			<div class="flex mt-0.5 space-x-2">
 				<div class=" flex-1">
 					<input
-						class="w-full rounded-lg py-2 px-1 text-sm dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+						class="text-sm w-full bg-transparent outline-hidden outline-none"
 						type="text"
 						placeholder={$i18n.t('Enter stop sequence')}
 						bind:value={params.stop}
@@ -288,7 +293,7 @@
 			<div class="flex mt-0.5 space-x-2">
 				<div class=" flex-1">
 					<input
-						class="w-full rounded-lg py-2 px-1 text-sm dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+						class="text-sm w-full bg-transparent outline-hidden outline-none"
 						type="text"
 						placeholder={$i18n.t('Enter reasoning effort')}
 						bind:value={params.reasoning_effort}
@@ -331,7 +336,7 @@
 			<div class="flex mt-0.5 space-x-2">
 				<div class=" flex-1">
 					<input
-						class="w-full rounded-lg pl-2 py-2 px-1 text-sm dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+						class="text-sm w-full bg-transparent outline-hidden outline-none"
 						type="text"
 						placeholder={$i18n.t(
 							'Enter comma-separated "token:bias_value" pairs (example: 5432:100, 413:-100)'
@@ -1091,6 +1096,74 @@
 		</div>
 	{/if}
 
+	<div class=" py-0.5 w-full justify-between">
+		<Tooltip
+			content={$i18n.t(
+				'This option enables or disables the use of the reasoning feature in Ollama, which allows the model to think before generating a response. When enabled, the model can take a moment to process the conversation context and generate a more thoughtful response.'
+			)}
+			placement="top-start"
+			className="inline-tooltip"
+		>
+			<div class=" py-0.5 flex w-full justify-between">
+				<div class=" self-center text-xs font-medium">
+					{'think'} ({$i18n.t('Ollama')})
+				</div>
+				<button
+					class="p-1 px-3 text-xs flex rounded-sm transition"
+					on:click={() => {
+						params.think = (params?.think ?? null) === null ? true : params.think ? false : null;
+					}}
+					type="button"
+				>
+					{#if params.think === true}
+						<span class="ml-2 self-center">{$i18n.t('On')}</span>
+					{:else if params.think === false}
+						<span class="ml-2 self-center">{$i18n.t('Off')}</span>
+					{:else}
+						<span class="ml-2 self-center">{$i18n.t('Default')}</span>
+					{/if}
+				</button>
+			</div>
+		</Tooltip>
+	</div>
+
+	<div class=" py-0.5 w-full justify-between">
+		<Tooltip
+			content={$i18n.t('The format to return a response in. Format can be json or a JSON schema.')}
+			placement="top-start"
+			className="inline-tooltip"
+		>
+			<div class=" py-0.5 flex w-full justify-between">
+				<div class=" self-center text-xs font-medium">
+					{'format'} ({$i18n.t('Ollama')})
+				</div>
+				<button
+					class="p-1 px-3 text-xs flex rounded-sm transition"
+					on:click={() => {
+						params.format = (params?.format ?? null) === null ? 'json' : null;
+					}}
+					type="button"
+				>
+					{#if (params?.format ?? null) === null}
+						<span class="ml-2 self-center">{$i18n.t('Default')}</span>
+					{:else}
+						<span class="ml-2 self-center">{$i18n.t('JSON')}</span>
+					{/if}
+				</button>
+			</div>
+		</Tooltip>
+
+		{#if (params?.format ?? null) !== null}
+			<div class="flex mt-0.5 space-x-2">
+				<Textarea
+					className="w-full  text-sm bg-transparent outline-hidden"
+					placeholder={$i18n.t('e.g. "json" or a JSON schema')}
+					bind:value={params.format}
+				/>
+			</div>
+		{/if}
+	</div>
+
 	<div class=" py-0.5 w-full justify-between">
 		<Tooltip
 			content={$i18n.t(
@@ -1367,37 +1440,111 @@
 			{/if}
 		</div>
 
-		<!-- <div class=" py-0.5 w-full justify-between">
-			<div class="flex w-full justify-between">
-				<div class=" self-center text-xs font-medium">{$i18n.t('Template')}</div>
-
-				<button
-					class="p-1 px-3 text-xs flex rounded-sm transition shrink-0 outline-hidden"
-					type="button"
-					on:click={() => {
-						params.template = (params?.template ?? null) === null ? '' : null;
-					}}
-				>
-					{#if (params?.template ?? null) === null}
-						<span class="ml-2 self-center">{$i18n.t('Default')}</span>
-					{:else}
-						<span class="ml-2 self-center">{$i18n.t('Custom')}</span>
-					{/if}
-				</button>
-			</div>
-
-			{#if (params?.template ?? null) !== null}
-				<div class="flex mt-0.5 space-x-2">
-					<div class=" flex-1">
-						<textarea
-							class="px-3 py-1.5 text-sm w-full bg-transparent border dark:border-gray-600 outline-hidden rounded-lg -mb-1"
-							placeholder={$i18n.t('Write your model template content here')}
-							rows="4"
-							bind:value={params.template}
-						/>
+		<div class=" py-0.5 w-full justify-between">
+			<Tooltip
+				content={$i18n.t(
+					'This option controls how long the model will stay loaded into memory following the request (default: 5m)'
+				)}
+				placement="top-start"
+				className="inline-tooltip"
+			>
+				<div class=" py-0.5 flex w-full justify-between">
+					<div class=" self-center text-xs font-medium">
+						{'keep_alive'} ({$i18n.t('Ollama')})
 					</div>
+					<button
+						class="p-1 px-3 text-xs flex rounded-sm transition"
+						on:click={() => {
+							params.keep_alive = (params?.keep_alive ?? null) === null ? '5m' : null;
+						}}
+						type="button"
+					>
+						{#if (params?.keep_alive ?? null) === null}
+							<span class="ml-2 self-center">{$i18n.t('Default')}</span>
+						{:else}
+							<span class="ml-2 self-center">{$i18n.t('Custom')}</span>
+						{/if}
+					</button>
+				</div>
+			</Tooltip>
+
+			{#if (params?.keep_alive ?? null) !== null}
+				<div class="flex mt-0.5 space-x-2">
+					<input
+						class="w-full text-sm bg-transparent outline-hidden"
+						type="text"
+						placeholder={$i18n.t("e.g. '30s','10m'. Valid time units are 's', 'm', 'h'.")}
+						bind:value={params.keep_alive}
+					/>
 				</div>
 			{/if}
-		</div> -->
+		</div>
+
+		{#if custom && admin}
+			<div class="flex flex-col justify-center">
+				{#each Object.keys(params?.custom_params ?? {}) as key}
+					<div class=" py-0.5 w-full justify-between mb-1">
+						<div class="flex w-full justify-between">
+							<div class=" self-center text-xs font-medium">
+								<input
+									type="text"
+									class=" text-xs w-full bg-transparent outline-none"
+									placeholder={$i18n.t('Custom Parameter Name')}
+									value={key}
+									on:change={(e) => {
+										const newKey = e.target.value.trim();
+										if (newKey && newKey !== key) {
+											params.custom_params[newKey] = params.custom_params[key];
+											delete params.custom_params[key];
+											params = {
+												...params,
+												custom_params: { ...params.custom_params }
+											};
+										}
+									}}
+								/>
+							</div>
+							<button
+								class="p-1 px-3 text-xs flex rounded-sm transition shrink-0 outline-hidden"
+								type="button"
+								on:click={() => {
+									delete params.custom_params[key];
+									params = {
+										...params,
+										custom_params: { ...params.custom_params }
+									};
+								}}
+							>
+								{$i18n.t('Remove')}
+							</button>
+						</div>
+						<div class="flex mt-0.5 space-x-2">
+							<div class=" flex-1">
+								<input
+									bind:value={params.custom_params[key]}
+									type="text"
+									class="text-sm w-full bg-transparent outline-hidden outline-none"
+									placeholder={$i18n.t('Custom Parameter Value')}
+								/>
+							</div>
+						</div>
+					</div>
+				{/each}
+
+				<button
+					class=" flex gap-2 items-center w-full text-center justify-center mt-1 mb-5"
+					type="button"
+					on:click={() => {
+						params.custom_params = (params?.custom_params ?? {}) || {};
+						params.custom_params['custom_param_name'] = 'custom_param_value';
+					}}
+				>
+					<div>
+						<Plus />
+					</div>
+					<div>{$i18n.t('Add Custom Parameter')}</div>
+				</button>
+			</div>
+		{/if}
 	{/if}
 </div>

src/lib/components/chat/Settings/General.svelte

@@ -10,12 +10,11 @@
 
 	import AdvancedParams from './Advanced/AdvancedParams.svelte';
 	import Textarea from '$lib/components/common/Textarea.svelte';
-
 	export let saveSettings: Function;
 	export let getModels: Function;
 
 	// General
-	let themes = ['dark', 'light', 'rose-pine dark', 'rose-pine-dawn light', 'oled-dark'];
+	let themes = ['dark', 'light', 'oled-dark'];
 	let selectedTheme = 'system';
 
 	let languages: Awaited<ReturnType<typeof getLanguages>> = [];
@@ -40,10 +39,6 @@
 		}
 	};
 
-	// Advanced
-	let requestFormat = null;
-	let keepAlive: string | null = null;
-
 	let params = {
 		// Advanced
 		stream_response: null,
@@ -71,37 +66,7 @@
 		num_gpu: null
 	};
 
-	const validateJSON = (json) => {
-		try {
-			const obj = JSON.parse(json);
-
-			if (obj && typeof obj === 'object') {
-				return true;
-			}
-		} catch (e) {}
-		return false;
-	};
-
-	const toggleRequestFormat = async () => {
-		if (requestFormat === null) {
-			requestFormat = 'json';
-		} else {
-			requestFormat = null;
-		}
-
-		saveSettings({ requestFormat: requestFormat !== null ? requestFormat : undefined });
-	};
-
 	const saveHandler = async () => {
-		if (requestFormat !== null && requestFormat !== 'json') {
-			if (validateJSON(requestFormat) === false) {
-				toast.error($i18n.t('Invalid JSON schema'));
-				return;
-			} else {
-				requestFormat = JSON.parse(requestFormat);
-			}
-		}
-
 		saveSettings({
 			system: system !== '' ? system : undefined,
 			params: {
@@ -130,15 +95,13 @@
 				use_mmap: params.use_mmap !== null ? params.use_mmap : undefined,
 				use_mlock: params.use_mlock !== null ? params.use_mlock : undefined,
 				num_thread: params.num_thread !== null ? params.num_thread : undefined,
-				num_gpu: params.num_gpu !== null ? params.num_gpu : undefined
-			},
-			keepAlive: keepAlive ? (isNaN(keepAlive) ? keepAlive : parseInt(keepAlive)) : undefined,
-			requestFormat: requestFormat !== null ? requestFormat : undefined
+				num_gpu: params.num_gpu !== null ? params.num_gpu : undefined,
+				think: params.think !== null ? params.think : undefined,
+				keep_alive: params.keep_alive !== null ? params.keep_alive : undefined,
+				format: params.format !== null ? params.format : undefined
+			}
 		});
 		dispatch('save');
-
-		requestFormat =
-			typeof requestFormat === 'object' ? JSON.stringify(requestFormat, null, 2) : requestFormat;
 	};
 
 	onMount(async () => {
@@ -149,14 +112,6 @@
 		notificationEnabled = $settings.notificationEnabled ?? false;
 		system = $settings.system ?? '';
 
-		requestFormat = $settings.requestFormat ?? null;
-		if (requestFormat !== null && requestFormat !== 'json') {
-			requestFormat =
-				typeof requestFormat === 'object' ? JSON.stringify(requestFormat, null, 2) : requestFormat;
-		}
-
-		keepAlive = $settings.keepAlive ?? null;
-
 		params = { ...params, ...$settings.params };
 		params.stop = $settings?.params?.stop ? ($settings?.params?.stop ?? []).join(',') : null;
 	});
@@ -335,77 +290,6 @@
 
 				{#if showAdvanced}
 					<AdvancedParams admin={$user?.role === 'admin'} bind:params />
-					<hr class=" border-gray-100 dark:border-gray-850" />
-
-					<div class=" w-full justify-between">
-						<div class="flex w-full justify-between">
-							<div class=" self-center text-xs font-medium">{$i18n.t('Keep Alive')}</div>
-
-							<button
-								class="p-1 px-3 text-xs flex rounded-sm transition"
-								type="button"
-								on:click={() => {
-									keepAlive = keepAlive === null ? '5m' : null;
-								}}
-							>
-								{#if keepAlive === null}
-									<span class="ml-2 self-center"> {$i18n.t('Default')} </span>
-								{:else}
-									<span class="ml-2 self-center"> {$i18n.t('Custom')} </span>
-								{/if}
-							</button>
-						</div>
-
-						{#if keepAlive !== null}
-							<div class="flex mt-1 space-x-2">
-								<input
-									class="w-full text-sm dark:text-gray-300 dark:bg-gray-850 outline-hidden"
-									type="text"
-									placeholder={$i18n.t("e.g. '30s','10m'. Valid time units are 's', 'm', 'h'.")}
-									bind:value={keepAlive}
-								/>
-							</div>
-						{/if}
-					</div>
-
-					<div>
-						<div class=" flex w-full justify-between">
-							<div class=" self-center text-xs font-medium">{$i18n.t('Request Mode')}</div>
-
-							<button
-								class="p-1 px-3 text-xs flex rounded-sm transition"
-								on:click={() => {
-									toggleRequestFormat();
-								}}
-							>
-								{#if requestFormat === null}
-									<span class="ml-2 self-center"> {$i18n.t('Default')} </span>
-								{:else}
-									<!-- <svg
-                            xmlns="http://www.w3.org/2000/svg"
-                            viewBox="0 0 20 20"
-                            fill="currentColor"
-                            class="w-4 h-4 self-center"
-                        >
-                            <path
-                                d="M10 2a.75.75 0 01.75.75v1.5a.75.75 0 01-1.5 0v-1.5A.75.75 0 0110 2zM10 15a.75.75 0 01.75.75v1.5a.75.75 0 01-1.5 0v-1.5A.75.75 0 0110 15zM10 7a3 3 0 100 6 3 3 0 000-6zM15.657 5.404a.75.75 0 10-1.06-1.06l-1.061 1.06a.75.75 0 001.06 1.06l1.06-1.06zM6.464 14.596a.75.75 0 10-1.06-1.06l-1.06 1.06a.75.75 0 001.06 1.06l1.06-1.06zM18 10a.75.75 0 01-.75.75h-1.5a.75.75 0 010-1.5h1.5A.75.75 0 0118 10zM5 10a.75.75 0 01-.75.75h-1.5a.75.75 0 010-1.5h1.5A.75.75 0 015 10zM14.596 15.657a.75.75 0 001.06-1.06l-1.06-1.061a.75.75 0 10-1.06 1.06l1.06 1.06zM5.404 6.464a.75.75 0 001.06-1.06l-1.06-1.06a.75.75 0 10-1.061 1.06l1.06 1.06z"
-                            />
-                        </svg> -->
-									<span class="ml-2 self-center"> {$i18n.t('JSON')} </span>
-								{/if}
-							</button>
-						</div>
-
-						{#if requestFormat !== null}
-							<div class="flex mt-1 space-x-2">
-								<Textarea
-									className="w-full  text-sm dark:text-gray-300 dark:bg-gray-900 outline-hidden"
-									placeholder={$i18n.t('e.g. "json" or a JSON schema')}
-									bind:value={requestFormat}
-								/>
-							</div>
-						{/if}
-					</div>
 				{/if}
 			</div>
 		{/if}

src/lib/components/chat/Settings/Interface.svelte

@@ -17,6 +17,7 @@
 
 	// Addons
 	let titleAutoGenerate = true;
+	let autoFollowUps = true;
 	let autoTags = true;
 
 	let responseAutoCopy = false;
@@ -197,6 +198,11 @@
 		});
 	};
 
+	const toggleAutoFollowUps = async () => {
+		autoFollowUps = !autoFollowUps;
+		saveSettings({ autoFollowUps });
+	};
+
 	const toggleAutoTags = async () => {
 		autoTags = !autoTags;
 		saveSettings({ autoTags });
@@ -286,35 +292,36 @@
 
 	onMount(async () => {
 		titleAutoGenerate = $settings?.title?.auto ?? true;
-		autoTags = $settings.autoTags ?? true;
+		autoTags = $settings?.autoTags ?? true;
+		autoFollowUps = $settings?.autoFollowUps ?? true;
 
-		highContrastMode = $settings.highContrastMode ?? false;
+		highContrastMode = $settings?.highContrastMode ?? false;
 
-		detectArtifacts = $settings.detectArtifacts ?? true;
-		responseAutoCopy = $settings.responseAutoCopy ?? false;
+		detectArtifacts = $settings?.detectArtifacts ?? true;
+		responseAutoCopy = $settings?.responseAutoCopy ?? false;
 
-		showUsername = $settings.showUsername ?? false;
-		showUpdateToast = $settings.showUpdateToast ?? true;
-		showChangelog = $settings.showChangelog ?? true;
+		showUsername = $settings?.showUsername ?? false;
+		showUpdateToast = $settings?.showUpdateToast ?? true;
+		showChangelog = $settings?.showChangelog ?? true;
 
-		showEmojiInCall = $settings.showEmojiInCall ?? false;
-		voiceInterruption = $settings.voiceInterruption ?? false;
+		showEmojiInCall = $settings?.showEmojiInCall ?? false;
+		voiceInterruption = $settings?.voiceInterruption ?? false;
 
-		richTextInput = $settings.richTextInput ?? true;
-		promptAutocomplete = $settings.promptAutocomplete ?? false;
-		largeTextAsFile = $settings.largeTextAsFile ?? false;
-		copyFormatted = $settings.copyFormatted ?? false;
+		richTextInput = $settings?.richTextInput ?? true;
+		promptAutocomplete = $settings?.promptAutocomplete ?? false;
+		largeTextAsFile = $settings?.largeTextAsFile ?? false;
+		copyFormatted = $settings?.copyFormatted ?? false;
 
-		collapseCodeBlocks = $settings.collapseCodeBlocks ?? false;
-		expandDetails = $settings.expandDetails ?? false;
+		collapseCodeBlocks = $settings?.collapseCodeBlocks ?? false;
+		expandDetails = $settings?.expandDetails ?? false;
 
-		landingPageMode = $settings.landingPageMode ?? '';
-		chatBubble = $settings.chatBubble ?? true;
-		widescreenMode = $settings.widescreenMode ?? false;
-		splitLargeChunks = $settings.splitLargeChunks ?? false;
-		scrollOnBranchChange = $settings.scrollOnBranchChange ?? true;
-		chatDirection = $settings.chatDirection ?? 'auto';
-		userLocation = $settings.userLocation ?? false;
+		landingPageMode = $settings?.landingPageMode ?? '';
+		chatBubble = $settings?.chatBubble ?? true;
+		widescreenMode = $settings?.widescreenMode ?? false;
+		splitLargeChunks = $settings?.splitLargeChunks ?? false;
+		scrollOnBranchChange = $settings?.scrollOnBranchChange ?? true;
+		chatDirection = $settings?.chatDirection ?? 'auto';
+		userLocation = $settings?.userLocation ?? false;
 
 		notificationSound = $settings?.notificationSound ?? true;
 		notificationSoundAlways = $settings?.notificationSoundAlways ?? false;
@@ -324,19 +331,19 @@
 
 		stylizedPdfExport = $settings?.stylizedPdfExport ?? true;
 
-		hapticFeedback = $settings.hapticFeedback ?? false;
-		ctrlEnterToSend = $settings.ctrlEnterToSend ?? false;
+		hapticFeedback = $settings?.hapticFeedback ?? false;
+		ctrlEnterToSend = $settings?.ctrlEnterToSend ?? false;
 
-		imageCompression = $settings.imageCompression ?? false;
-		imageCompressionSize = $settings.imageCompressionSize ?? { width: '', height: '' };
+		imageCompression = $settings?.imageCompression ?? false;
+		imageCompressionSize = $settings?.imageCompressionSize ?? { width: '', height: '' };
 
 		defaultModelId = $settings?.models?.at(0) ?? '';
 		if ($config?.default_models) {
 			defaultModelId = $config.default_models.split(',')[0];
 		}
 
-		backgroundImageUrl = $settings.backgroundImageUrl ?? null;
-		webSearch = $settings.webSearch ?? null;
+		backgroundImageUrl = $settings?.backgroundImageUrl ?? null;
+		webSearch = $settings?.webSearch ?? null;
 	});
 </script>
 
@@ -619,6 +626,26 @@
 				</div>
 			</div>
 
+			<div>
+				<div class=" py-0.5 flex w-full justify-between">
+					<div class=" self-center text-xs">{$i18n.t('Follow-Up Auto-Generation')}</div>
+
+					<button
+						class="p-1 px-3 text-xs flex rounded-sm transition"
+						on:click={() => {
+							toggleAutoFollowUps();
+						}}
+						type="button"
+					>
+						{#if autoFollowUps === true}
+							<span class="ml-2 self-center">{$i18n.t('On')}</span>
+						{:else}
+							<span class="ml-2 self-center">{$i18n.t('Off')}</span>
+						{/if}
+					</button>
+				</div>
+			</div>
+
 			<div>
 				<div class=" py-0.5 flex w-full justify-between">
 					<div class=" self-center text-xs">{$i18n.t('Chat Tags Auto-Generation')}</div>

src/lib/components/chat/SettingsModal.svelte

@@ -16,6 +16,7 @@
 	import User from '../icons/User.svelte';
 	import Personalization from './Settings/Personalization.svelte';
 	import Search from '../icons/Search.svelte';
+	import XMark from '../icons/XMark.svelte';
 	import Connections from './Settings/Connections.svelte';
 	import Tools from './Settings/Tools.svelte';
 
@@ -34,93 +35,160 @@
 			id: 'general',
 			title: 'General',
 			keywords: [
-				'general',
-				'theme',
-				'language',
-				'notifications',
-				'system',
-				'systemprompt',
-				'prompt',
-				'advanced',
-				'settings',
-				'defaultsettings',
+				'advancedparams',
+				'advancedparameters',
+				'advanced params',
+				'advanced parameters',
 				'configuration',
-				'systemsettings',
-				'notificationsettings',
-				'systempromptconfig',
-				'languageoptions',
 				'defaultparameters',
-				'systemparameters'
+				'default parameters',
+				'defaultsettings',
+				'default settings',
+				'general',
+				'keepalive',
+				'keep alive',
+				'languages',
+				'notifications',
+				'requestmode',
+				'request mode',
+				'systemparameters',
+				'system parameters',
+				'systemprompt',
+				'system prompt',
+				'systemsettings',
+				'system settings',
+				'theme',
+				'translate',
+				'webuisettings',
+				'webui settings'
 			]
 		},
 		{
 			id: 'interface',
 			title: 'Interface',
 			keywords: [
-				'defaultmodel',
-				'selectmodel',
-				'ui',
-				'userinterface',
-				'display',
-				'layout',
-				'design',
-				'landingpage',
-				'landingpagemode',
-				'default',
-				'chat',
-				'chatbubble',
-				'chatui',
-				'username',
-				'showusername',
-				'displayusername',
-				'widescreen',
-				'widescreenmode',
-				'fullscreen',
-				'expandmode',
+				'allow user location',
+				'allow voice interruption in call',
+				'allowuserlocation',
+				'allowvoiceinterruptionincall',
+				'always collapse codeblocks',
+				'always collapse code blocks',
+				'always expand details',
+				'always on web search',
+				'always play notification sound',
+				'alwayscollapsecodeblocks',
+				'alwaysexpanddetails',
+				'alwaysonwebsearch',
+				'alwaysplaynotificationsound',
+				'android',
+				'auto chat tags',
+				'auto copy response to clipboard',
+				'auto title',
+				'autochattags',
+				'autocopyresponsetoclipboard',
+				'autotitle',
+				'beta',
+				'call',
+				'chat background image',
+				'chat bubble ui',
+				'chat direction',
+				'chat tags autogen',
+				'chat tags autogeneration',
+				'chat ui',
+				'chatbackgroundimage',
+				'chatbubbleui',
 				'chatdirection',
+				'chat tags autogeneration',
+				'chattagsautogeneration',
+				'chatui',
+				'copy formatted text',
+				'copyformattedtext',
+				'default model',
+				'defaultmodel',
+				'design',
+				'detect artifacts automatically',
+				'detectartifactsautomatically',
+				'display emoji in call',
+				'display username',
+				'displayemojiincall',
+				'displayusername',
+				'enter key behavior',
+				'enterkeybehavior',
+				'expand mode',
+				'expandmode',
+				'file',
+				'followup autogeneration',
+				'followupautogeneration',
+				'fullscreen',
+				'fullwidthmode',
+				'full width mode',
+				'haptic feedback',
+				'hapticfeedback',
+				'high contrast mode',
+				'highcontrastmode',
+				'iframe sandbox allow forms',
+				'iframe sandbox allow same origin',
+				'iframesandboxallowforms',
+				'iframesandboxallowsameorigin',
+				'imagecompression',
+				'image compression',
+				'imagemaxcompressionsize',
+				'image max compression size',
+				'interface customization',
+				'interface options',
+				'interfacecustomization',
+				'interfaceoptions',
+				'landing page mode',
+				'landingpagemode',
+				'layout',
+				'left to right',
+				'left-to-right',
 				'lefttoright',
 				'ltr',
+				'paste large text as file',
+				'pastelargetextasfile',
+				'reset background',
+				'resetbackground',
+				'response auto copy',
+				'responseautocopy',
+				'rich text input for chat',
+				'richtextinputforchat',
+				'right to left',
+				'right-to-left',
 				'righttoleft',
 				'rtl',
-				'notifications',
-				'toast',
-				'toastnotifications',
-				'largechunks',
-				'streamlargechunks',
-				'scroll',
-				'scrollonbranchchange',
+				'scroll behavior',
+				'scroll on branch change',
 				'scrollbehavior',
-				'richtext',
-				'richtextinput',
-				'background',
-				'chatbackground',
-				'chatbackgroundimage',
-				'backgroundimage',
-				'uploadbackground',
-				'resetbackground',
-				'titleautogen',
+				'scrollonbranchchange',
+				'select model',
+				'selectmodel',
+				'settings',
+				'show username',
+				'showusername',
+				'stream large chunks',
+				'streamlargechunks',
+				'stylized pdf export',
+				'stylizedpdfexport',
+				'title autogeneration',
 				'titleautogeneration',
-				'autotitle',
-				'chattags',
-				'autochattags',
-				'responseautocopy',
-				'clipboard',
-				'location',
-				'userlocation',
+				'toast notifications for new updates',
+				'toastnotificationsfornewupdates',
+				'upload background',
+				'uploadbackground',
+				'user interface',
+				'user location access',
+				'userinterface',
 				'userlocationaccess',
-				'haptic',
-				'hapticfeedback',
 				'vibration',
-				'voice',
+				'voice control',
 				'voicecontrol',
-				'voiceinterruption',
-				'call',
-				'emojis',
-				'displayemoji',
-				'save',
-				'interfaceoptions',
-				'interfacecustomization',
-				'alwaysonwebsearch'
+				'widescreen mode',
+				'widescreenmode',
+				'whatsnew',
+				'whats new',
+				'websearchinchat',
+				'web search in chat'
 			]
 		},
 		...($user?.role === 'admin' ||
@@ -129,7 +197,15 @@
 					{
 						id: 'connections',
 						title: 'Connections',
-						keywords: []
+						keywords: [
+							'addconnection',
+							'add connection',
+							'manageconnections',
+							'manage connections',
+							'manage direct connections',
+							'managedirectconnections',
+							'settings'
+						]
 					}
 				]
 			: []),
@@ -140,7 +216,15 @@
 					{
 						id: 'tools',
 						title: 'Tools',
-						keywords: []
+						keywords: [
+							'addconnection',
+							'add connection',
+							'managetools',
+							'manage tools',
+							'manage tool servers',
+							'managetoolservers',
+							'settings'
+						]
 					}
 				]
 			: []),
@@ -149,159 +233,233 @@
 			id: 'personalization',
 			title: 'Personalization',
 			keywords: [
-				'personalization',
-				'memory',
-				'personalize',
-				'preferences',
-				'profile',
-				'personalsettings',
+				'account preferences',
+				'account settings',
+				'accountpreferences',
+				'accountsettings',
+				'custom settings',
 				'customsettings',
-				'userpreferences',
-				'accountpreferences'
+				'experimental',
+				'memories',
+				'memory',
+				'personalization',
+				'personalize',
+				'personal settings',
+				'personalsettings',
+				'profile',
+				'user preferences',
+				'userpreferences'
 			]
 		},
 		{
 			id: 'audio',
 			title: 'Audio',
 			keywords: [
-				'audio',
-				'sound',
-				'soundsettings',
-				'audiocontrol',
-				'volume',
-				'speech',
-				'speechrecognition',
-				'stt',
-				'speechtotext',
-				'tts',
-				'texttospeech',
-				'playback',
-				'playbackspeed',
-				'voiceplayback',
-				'speechplayback',
-				'audiooutput',
-				'speechengine',
-				'voicecontrol',
-				'audioplayback',
-				'transcription',
-				'autotranscribe',
-				'autosend',
-				'speechsettings',
-				'audiovoice',
-				'voiceoptions',
-				'setvoice',
-				'nonlocalvoices',
-				'savesettings',
+				'audio config',
+				'audio control',
+				'audio features',
+				'audio input',
+				'audio output',
+				'audio playback',
+				'audio voice',
 				'audioconfig',
-				'speechconfig',
-				'voicerecognition',
-				'speechsynthesis',
-				'speechmode',
-				'voicespeed',
-				'speechrate',
-				'speechspeed',
-				'audioinput',
+				'audiocontrol',
 				'audiofeatures',
-				'voicemodes'
+				'audioinput',
+				'audiooutput',
+				'audioplayback',
+				'audiovoice',
+				'auto playback response',
+				'autoplaybackresponse',
+				'auto transcribe',
+				'autotranscribe',
+				'instant auto send after voice transcription',
+				'instantautosendaftervoicetranscription',
+				'language',
+				'non local voices',
+				'nonlocalvoices',
+				'save settings',
+				'savesettings',
+				'set voice',
+				'setvoice',
+				'sound settings',
+				'soundsettings',
+				'speech config',
+				'speech mode',
+				'speech playback speed',
+				'speech rate',
+				'speech recognition',
+				'speech settings',
+				'speech speed',
+				'speech synthesis',
+				'speech to text engine',
+				'speechconfig',
+				'speechmode',
+				'speechplaybackspeed',
+				'speechrate',
+				'speechrecognition',
+				'speechsettings',
+				'speechspeed',
+				'speechsynthesis',
+				'speechtotextengine',
+				'speedch playback rate',
+				'speedchplaybackrate',
+				'stt settings',
+				'sttsettings',
+				'text to speech engine',
+				'text to speech',
+				'textospeechengine',
+				'texttospeech',
+				'texttospeechvoice',
+				'text to speech voice',
+				'voice control',
+				'voice modes',
+				'voice options',
+				'voice playback',
+				'voice recognition',
+				'voice speed',
+				'voicecontrol',
+				'voicemodes',
+				'voiceoptions',
+				'voiceplayback',
+				'voicerecognition',
+				'voicespeed',
+				'volume'
 			]
 		},
 		{
 			id: 'chats',
 			title: 'Chats',
 			keywords: [
-				'chat',
-				'messages',
-				'conversations',
-				'chatsettings',
-				'history',
+				'archive all chats',
+				'archive chats',
+				'archiveallchats',
+				'archivechats',
+				'archived chats',
+				'archivedchats',
+				'chat activity',
+				'chat history',
+				'chat settings',
+				'chatactivity',
 				'chathistory',
-				'messagehistory',
-				'messagearchive',
-				'convo',
-				'chats',
+				'chatsettings',
+				'conversation activity',
+				'conversation history',
+				'conversationactivity',
 				'conversationhistory',
-				'exportmessages',
-				'chatactivity'
+				'conversations',
+				'convos',
+				'delete all chats',
+				'delete chats',
+				'deleteallchats',
+				'deletechats',
+				'export chats',
+				'exportchats',
+				'import chats',
+				'importchats',
+				'message activity',
+				'message archive',
+				'message history',
+				'messagearchive',
+				'messagehistory'
 			]
 		},
 		{
 			id: 'account',
 			title: 'Account',
 			keywords: [
-				'account',
-				'profile',
-				'security',
-				'privacy',
-				'settings',
+				'account preferences',
+				'account settings',
+				'accountpreferences',
+				'accountsettings',
+				'api keys',
+				'apikeys',
+				'change password',
+				'changepassword',
+				'jwt token',
+				'jwttoken',
 				'login',
+				'new password',
+				'newpassword',
+				'notification webhook url',
+				'notificationwebhookurl',
+				'personal settings',
+				'personalsettings',
+				'privacy settings',
+				'privacysettings',
+				'profileavatar',
+				'profile avatar',
+				'profile details',
+				'profile image',
+				'profile picture',
+				'profiledetails',
+				'profileimage',
+				'profilepicture',
+				'security settings',
+				'securitysettings',
+				'update account',
+				'update password',
+				'updateaccount',
+				'updatepassword',
+				'user account',
+				'user data',
+				'user preferences',
+				'user profile',
 				'useraccount',
 				'userdata',
-				'api',
-				'apikey',
+				'username',
+				'userpreferences',
 				'userprofile',
-				'profiledetails',
-				'accountsettings',
-				'accountpreferences',
-				'securitysettings',
-				'privacysettings'
-			]
-		},
-		{
-			id: 'admin',
-			title: 'Admin',
-			keywords: [
-				'admin',
-				'administrator',
-				'adminsettings',
-				'adminpanel',
-				'systemadmin',
-				'administratoraccess',
-				'systemcontrol',
-				'manage',
-				'management',
-				'admincontrols',
-				'adminfeatures',
-				'usercontrol',
-				'arenamodel',
-				'evaluations',
-				'websearch',
-				'database',
-				'pipelines',
-				'images',
-				'audio',
-				'documents',
-				'rag',
-				'models',
-				'ollama',
-				'openai',
-				'users'
+				'webhook url',
+				'webhookurl'
 			]
 		},
 		{
 			id: 'about',
 			title: 'About',
 			keywords: [
-				'about',
-				'info',
-				'information',
-				'version',
-				'documentation',
-				'help',
-				'support',
-				'details',
-				'aboutus',
-				'softwareinfo',
-				'timothyjaeryangbaek',
-				'openwebui',
-				'release',
-				'updates',
-				'updateinfo',
-				'versioninfo',
+				'about app',
+				'about me',
+				'about open webui',
+				'about page',
+				'about us',
 				'aboutapp',
-				'terms',
-				'termsandconditions',
+				'aboutme',
+				'aboutopenwebui',
+				'aboutpage',
+				'aboutus',
+				'check for updates',
+				'checkforupdates',
 				'contact',
-				'aboutpage'
+				'copyright',
+				'details',
+				'discord',
+				'documentation',
+				'github',
+				'help',
+				'information',
+				'license',
+				'redistributions',
+				'release',
+				'see whats new',
+				'seewhatsnew',
+				'settings',
+				'software info',
+				'softwareinfo',
+				'support',
+				'terms and conditions',
+				'terms of use',
+				'termsandconditions',
+				'termsofuse',
+				'timothy jae ryang baek',
+				'timothy j baek',
+				'timothyjaeryangbaek',
+				'timothyjbaek',
+				'twitter',
+				'update info',
+				'updateinfo',
+				'version info',
+				'versioninfo'
 			]
 		}
 	];
@@ -384,36 +542,34 @@
 		<div class=" flex justify-between dark:text-gray-300 px-5 pt-4 pb-1">
 			<div class=" text-lg font-medium self-center">{$i18n.t('Settings')}</div>
 			<button
+				aria-label={$i18n.t('Close settings modal')}
 				class="self-center"
 				on:click={() => {
 					show = false;
 				}}
 			>
-				<svg
-					xmlns="http://www.w3.org/2000/svg"
-					viewBox="0 0 20 20"
-					fill="currentColor"
-					class="w-5 h-5"
-				>
-					<path
-						d="M6.28 5.22a.75.75 0 00-1.06 1.06L8.94 10l-3.72 3.72a.75.75 0 101.06 1.06L10 11.06l3.72 3.72a.75.75 0 101.06-1.06L11.06 10l3.72-3.72a.75.75 0 00-1.06-1.06L10 8.94 6.28 5.22z"
-					/>
-				</svg>
+				<XMark className="w-5 h-5"></XMark>
 			</button>
 		</div>
 
 		<div class="flex flex-col md:flex-row w-full px-4 pt-1 pb-4 md:space-x-4">
 			<div
 				id="settings-tabs-container"
-				class="tabs flex flex-row overflow-x-auto gap-2.5 md:gap-1 md:flex-col flex-1 md:flex-none md:w-40 dark:text-gray-200 text-sm font-medium text-left mb-1 md:mb-0 -translate-y-1"
+				class="tabs flex flex-row overflow-x-auto gap-2.5 md:gap-1 md:flex-col flex-1 md:flex-none md:w-40 md:min-h-[32rem] md:max-h-[32rem] dark:text-gray-200 text-sm font-medium text-left mb-1 md:mb-0 -translate-y-1"
 			>
 				<div class="hidden md:flex w-full rounded-xl -mb-1 px-0.5 gap-2" id="settings-search">
 					<div class="self-center rounded-l-xl bg-transparent">
-						<Search className="size-3.5" />
+						<Search
+							className="size-3.5"
+							strokeWidth={($settings?.highContrastMode ?? false) ? '3' : '1.5'}
+						/>
 					</div>
+					<label class="sr-only" for="search-input-settings-modal">{$i18n.t('Search')}</label>
 					<input
-						class="w-full py-1.5 text-sm bg-transparent dark:text-gray-300 outline-hidden"
+						class={`w-full py-1.5 text-sm bg-transparent dark:text-gray-300 outline-hidden
+								${($settings?.highContrastMode ?? false) ? 'placeholder-gray-800' : ''}`}
 						bind:value={search}
+						id="search-input-settings-modal"
 						on:input={searchDebounceHandler}
 						placeholder={$i18n.t('Search')}
 					/>
@@ -423,10 +579,16 @@
 					{#each visibleTabs as tabId (tabId)}
 						{#if tabId === 'general'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'general'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'general'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'general';
 								}}
@@ -449,10 +611,16 @@
 							</button>
 						{:else if tabId === 'interface'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'interface'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'interface'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'interface';
 								}}
@@ -476,10 +644,16 @@
 						{:else if tabId === 'connections'}
 							{#if $user?.role === 'admin' || ($user?.role === 'user' && $config?.features?.enable_direct_connections)}
 								<button
-									class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-									'connections'
-										? ''
-										: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+									class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'connections'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 									on:click={() => {
 										selectedTab = 'connections';
 									}}
@@ -502,10 +676,16 @@
 						{:else if tabId === 'tools'}
 							{#if $user?.role === 'admin' || ($user?.role === 'user' && $user?.permissions?.features?.direct_tool_servers)}
 								<button
-									class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-									'tools'
-										? ''
-										: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+									class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'tools'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 									on:click={() => {
 										selectedTab = 'tools';
 									}}
@@ -529,10 +709,16 @@
 							{/if}
 						{:else if tabId === 'personalization'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'personalization'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'personalization'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'personalization';
 								}}
@@ -544,10 +730,16 @@
 							</button>
 						{:else if tabId === 'audio'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'audio'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'audio'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'audio';
 								}}
@@ -571,10 +763,16 @@
 							</button>
 						{:else if tabId === 'chats'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'chats'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'chats'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'chats';
 								}}
@@ -597,10 +795,16 @@
 							</button>
 						{:else if tabId === 'account'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'account'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'account'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'account';
 								}}
@@ -623,10 +827,16 @@
 							</button>
 						{:else if tabId === 'about'}
 							<button
-								class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-								'about'
-									? ''
-									: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+								class={`px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition
+								${
+									selectedTab === 'about'
+										? ($settings?.highContrastMode ?? false)
+											? 'dark:bg-gray-800 bg-gray-200'
+											: ''
+										: ($settings?.highContrastMode ?? false)
+											? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+											: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'
+								}`}
 								on:click={() => {
 									selectedTab = 'about';
 								}}
@@ -647,35 +857,6 @@
 								</div>
 								<div class=" self-center">{$i18n.t('About')}</div>
 							</button>
-						{:else if tabId === 'admin'}
-							{#if $user?.role === 'admin'}
-								<button
-									class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-left transition {selectedTab ===
-									'admin'
-										? ''
-										: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
-									on:click={async () => {
-										await goto('/admin/settings');
-										show = false;
-									}}
-								>
-									<div class=" self-center mr-2">
-										<svg
-											xmlns="http://www.w3.org/2000/svg"
-											viewBox="0 0 24 24"
-											fill="currentColor"
-											class="size-4"
-										>
-											<path
-												fill-rule="evenodd"
-												d="M4.5 3.75a3 3 0 0 0-3 3v10.5a3 3 0 0 0 3 3h15a3 3 0 0 0 3-3V6.75a3 3 0 0 0-3-3h-15Zm4.125 3a2.25 2.25 0 1 0 0 4.5 2.25 2.25 0 0 0 0-4.5Zm-3.873 8.703a4.126 4.126 0 0 1 7.746 0 .75.75 0 0 1-.351.92 7.47 7.47 0 0 1-3.522.877 7.47 7.47 0 0 1-3.522-.877.75.75 0 0 1-.351-.92ZM15 8.25a.75.75 0 0 0 0 1.5h3.75a.75.75 0 0 0 0-1.5H15ZM14.25 12a.75.75 0 0 1 .75-.75h3.75a.75.75 0 0 1 0 1.5H15a.75.75 0 0 1-.75-.75Zm.75 2.25a.75.75 0 0 0 0 1.5h3.75a.75.75 0 0 0 0-1.5H15Z"
-												clip-rule="evenodd"
-											/>
-										</svg>
-									</div>
-									<div class=" self-center">{$i18n.t('Admin Settings')}</div>
-								</button>
-							{/if}
 						{/if}
 					{/each}
 				{:else}
@@ -683,6 +864,37 @@
 						{$i18n.t('No results found')}
 					</div>
 				{/if}
+
+				{#if $user?.role === 'admin'}
+					<a
+						href="/admin/settings"
+						class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none md:mt-auto flex text-left transition {$settings?.highContrastMode
+							? 'hover:bg-gray-200 dark:hover:bg-gray-800'
+							: 'text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+						on:click={async (e) => {
+							e.preventDefault();
+							await goto('/admin/settings');
+							show = false;
+						}}
+					>
+						<div class=" self-center mr-2">
+							<svg
+								xmlns="http://www.w3.org/2000/svg"
+								viewBox="0 0 24 24"
+								fill="currentColor"
+								aria-hidden="true"
+								class="size-4"
+							>
+								<path
+									fill-rule="evenodd"
+									d="M4.5 3.75a3 3 0 0 0-3 3v10.5a3 3 0 0 0 3 3h15a3 3 0 0 0 3-3V6.75a3 3 0 0 0-3-3h-15Zm4.125 3a2.25 2.25 0 1 0 0 4.5 2.25 2.25 0 0 0 0-4.5Zm-3.873 8.703a4.126 4.126 0 0 1 7.746 0 .75.75 0 0 1-.351.92 7.47 7.47 0 0 1-3.522.877 7.47 7.47 0 0 1-3.522-.877.75.75 0 0 1-.351-.92ZM15 8.25a.75.75 0 0 0 0 1.5h3.75a.75.75 0 0 0 0-1.5H15ZM14.25 12a.75.75 0 0 1 .75-.75h3.75a.75.75 0 0 1 0 1.5H15a.75.75 0 0 1-.75-.75Zm.75 2.25a.75.75 0 0 0 0 1.5h3.75a.75.75 0 0 0 0-1.5H15Z"
+									clip-rule="evenodd"
+								/>
+							</svg>
+						</div>
+						<div class=" self-center">{$i18n.t('Admin Settings')}</div>
+					</a>
+				{/if}
 			</div>
 			<div class="flex-1 md:min-h-[32rem] max-h-[32rem]">
 				{#if selectedTab === 'general'}
@@ -763,6 +975,7 @@
 	}
 
 	input[type='number'] {
+		appearance: textfield;
 		-moz-appearance: textfield; /* Firefox */
 	}
 </style>

src/lib/components/common/CodeEditor.svelte

@@ -11,10 +11,13 @@
 
 	import { oneDark } from '@codemirror/theme-one-dark';
 
-	import { onMount, createEventDispatcher, getContext, tick } from 'svelte';
+	import { onMount, createEventDispatcher, getContext, tick, onDestroy } from 'svelte';
+
+	import PyodideWorker from '$lib/workers/pyodide.worker?worker';
 
 	import { formatPythonCode } from '$lib/apis/utils';
 	import { toast } from 'svelte-sonner';
+	import { user } from '$lib/stores';
 
 	const dispatch = createEventDispatcher();
 	const i18n = getContext('i18n');
@@ -113,13 +116,82 @@
 		return await language?.load();
 	};
 
+	let pyodideWorkerInstance = null;
+
+	const getPyodideWorker = () => {
+		if (!pyodideWorkerInstance) {
+			pyodideWorkerInstance = new PyodideWorker(); // Your worker constructor
+		}
+		return pyodideWorkerInstance;
+	};
+
+	// Generate unique IDs for requests
+	let _formatReqId = 0;
+
+	const formatPythonCodePyodide = (code) => {
+		return new Promise((resolve, reject) => {
+			const id = `format-${++_formatReqId}`;
+			let timeout;
+			const worker = getPyodideWorker();
+
+			const script = `
+import black
+print(black.format_str("""${code.replace(/\\/g, '\\\\').replace(/`/g, '\\`').replace(/"/g, '\\"')}""", mode=black.Mode()))
+`;
+
+			const packages = ['black'];
+
+			function handleMessage(event) {
+				const { id: eventId, stdout, stderr } = event.data;
+				if (eventId !== id) return; // Only handle our message
+				clearTimeout(timeout);
+				worker.removeEventListener('message', handleMessage);
+				worker.removeEventListener('error', handleError);
+
+				if (stderr) {
+					reject(stderr);
+				} else {
+					const formatted = stdout && typeof stdout === 'string' ? stdout.trim() : '';
+					resolve({ code: formatted });
+				}
+			}
+
+			function handleError(event) {
+				clearTimeout(timeout);
+				worker.removeEventListener('message', handleMessage);
+				worker.removeEventListener('error', handleError);
+				reject(event.message || 'Pyodide worker error');
+			}
+
+			worker.addEventListener('message', handleMessage);
+			worker.addEventListener('error', handleError);
+
+			// Send to worker
+			worker.postMessage({ id, code: script, packages });
+
+			// Timeout
+			timeout = setTimeout(() => {
+				worker.removeEventListener('message', handleMessage);
+				worker.removeEventListener('error', handleError);
+				try {
+					worker.terminate();
+				} catch {}
+				pyodideWorkerInstance = null;
+				reject('Execution Time Limit Exceeded');
+			}, 60000);
+		});
+	};
+
 	export const formatPythonCodeHandler = async () => {
 		if (codeEditor) {
-			const res = await formatPythonCode(localStorage.token, _value).catch((error) => {
+			const res = await (
+				$user?.role === 'admin'
+					? formatPythonCode(localStorage.token, _value)
+					: formatPythonCodePyodide(_value)
+			).catch((error) => {
 				toast.error(`${error}`);
 				return null;
 			});
-
 			if (res && res.code) {
 				const formattedCode = res.code;
 				codeEditor.dispatch({
@@ -240,6 +312,12 @@
 			document.removeEventListener('keydown', keydownHandler);
 		};
 	});
+
+	onDestroy(() => {
+		if (pyodideWorkerInstance) {
+			pyodideWorkerInstance.terminate();
+		}
+	});
 </script>
 
 <div id="code-textarea-{id}" class="h-full w-full text-sm" />

src/lib/components/common/Collapsible.svelte

@@ -13,6 +13,9 @@
 	dayjs.extend(relativeTime);
 
 	async function loadLocale(locales) {
+		if (!locales || !Array.isArray(locales)) {
+			return;
+		}
 		for (const locale of locales) {
 			try {
 				dayjs.locale(locale);

src/lib/components/common/ImagePreview.svelte

@@ -2,6 +2,9 @@
 	import { onDestroy, onMount } from 'svelte';
 	import panzoom, { type PanZoom } from 'panzoom';
 
+	import fileSaver from 'file-saver';
+	const { saveAs } = fileSaver;
+
 	export let show = false;
 	export let src = '';
 	export let alt = '';
@@ -29,22 +32,6 @@
 		console.log(instance.getTransform());
 	};
 
-	const downloadImage = (url, filename, prefixName = '') => {
-		fetch(url)
-			.then((response) => response.blob())
-			.then((blob) => {
-				const objectUrl = window.URL.createObjectURL(blob);
-				const link = document.createElement('a');
-				link.href = objectUrl;
-				link.download = `${prefixName}${filename}`;
-				document.body.appendChild(link);
-				link.click();
-				document.body.removeChild(link);
-				window.URL.revokeObjectURL(objectUrl);
-			})
-			.catch((error) => console.error('Error downloading image:', error));
-	};
-
 	const handleKeyDown = (event: KeyboardEvent) => {
 		if (event.key === 'Escape') {
 			console.log('Escape');
@@ -82,7 +69,7 @@
 		bind:this={previewElement}
 		class="modal fixed top-0 right-0 left-0 bottom-0 bg-black text-white w-full min-h-screen h-screen flex justify-center z-9999 overflow-hidden overscroll-contain"
 	>
-		<div class=" absolute left-0 w-full flex justify-between select-none z-10">
+		<div class=" absolute left-0 w-full flex justify-between select-none z-20">
 			<div>
 				<button
 					class=" p-5"
@@ -110,14 +97,74 @@
 
 			<div>
 				<button
-					class=" p-5"
-					on:pointerdown={(e) => {
-						e.stopImmediatePropagation();
-						e.preventDefault();
-						downloadImage(src, src.substring(src.lastIndexOf('/') + 1), alt);
-					}}
-					on:click={(e) => {
-						downloadImage(src, src.substring(src.lastIndexOf('/') + 1), alt);
+					class=" p-5 z-999"
+					on:click={() => {
+						if (src.startsWith('data:image/')) {
+							const base64Data = src.split(',')[1];
+							const blob = new Blob([Uint8Array.from(atob(base64Data), (c) => c.charCodeAt(0))], {
+								type: 'image/png'
+							});
+
+							const mimeType = blob.type || 'image/png';
+							// create file name based on the MIME type, alt should be a valid file name with extension
+							const fileName = alt
+								? `${alt.replaceAll('.', '')}.${mimeType.split('/')[1]}`
+								: 'download.png';
+
+							// Use FileSaver to save the blob
+							saveAs(blob, fileName);
+							return;
+						} else if (src.startsWith('blob:')) {
+							// Handle blob URLs
+							fetch(src)
+								.then((response) => response.blob())
+								.then((blob) => {
+									// detect the MIME type from the blob
+									const mimeType = blob.type || 'image/png';
+
+									// Create a new Blob with the correct MIME type
+									const blobWithType = new Blob([blob], { type: mimeType });
+
+									// create file name based on the MIME type, alt should be a valid file name with extension
+									const fileName = alt
+										? `${alt.replaceAll('.', '')}.${mimeType.split('/')[1]}`
+										: 'download.png';
+
+									// Use FileSaver to save the blob
+									saveAs(blobWithType, fileName);
+								})
+								.catch((error) => {
+									console.error('Error downloading blob:', error);
+								});
+							return;
+						} else if (
+							src.startsWith('/') ||
+							src.startsWith('http://') ||
+							src.startsWith('https://')
+						) {
+							// Handle remote URLs
+							fetch(src)
+								.then((response) => response.blob())
+								.then((blob) => {
+									// detect the MIME type from the blob
+									const mimeType = blob.type || 'image/png';
+
+									// Create a new Blob with the correct MIME type
+									const blobWithType = new Blob([blob], { type: mimeType });
+
+									// create file name based on the MIME type, alt should be a valid file name with extension
+									const fileName = alt
+										? `${alt.replaceAll('.', '')}.${mimeType.split('/')[1]}`
+										: 'download.png';
+
+									// Use FileSaver to save the blob
+									saveAs(blobWithType, fileName);
+								})
+								.catch((error) => {
+									console.error('Error downloading remote image:', error);
+								});
+							return;
+						}
 					}}
 				>
 					<svg
@@ -136,8 +183,14 @@
 				</button>
 			</div>
 		</div>
-		<div bind:this={sceneElement} class="flex h-full max-h-full justify-center items-center">
-			<img {src} {alt} class=" mx-auto h-full object-scale-down select-none" draggable="false" />
+		<div class="flex h-full max-h-full justify-center items-center z-0">
+			<img
+				bind:this={sceneElement}
+				{src}
+				{alt}
+				class=" mx-auto h-full object-scale-down select-none"
+				draggable="false"
+			/>
 		</div>
 	</div>
 {/if}

src/lib/components/common/Loader.svelte

@@ -33,7 +33,9 @@
 	});
 
 	onDestroy(() => {
-		observer.disconnect();
+		if (observer) {
+			observer.disconnect();
+		}
 
 		if (intervalId) {
 			clearInterval(intervalId);

src/lib/components/common/RichTextInput.svelte

@@ -301,30 +301,38 @@
 							}
 
 							if (event.key === 'Enter') {
-								// Check if the current selection is inside a structured block (like codeBlock or list)
-								const { state } = view;
-								const { $head } = state.selection;
+								const isCtrlPressed = event.ctrlKey || event.metaKey; // metaKey is for Cmd key on Mac
+								if (event.shiftKey && !isCtrlPressed) {
+									editor.commands.setHardBreak(); // Insert a hard break
+									view.dispatch(view.state.tr.scrollIntoView()); // Move viewport to the cursor
+									event.preventDefault();
+									return true;
+								} else {
+									// Check if the current selection is inside a structured block (like codeBlock or list)
+									const { state } = view;
+									const { $head } = state.selection;
 
-								// Recursive function to check ancestors for specific node types
-								function isInside(nodeTypes: string[]): boolean {
-									let currentNode = $head;
-									while (currentNode) {
-										if (nodeTypes.includes(currentNode.parent.type.name)) {
-											return true;
+									// Recursive function to check ancestors for specific node types
+									function isInside(nodeTypes: string[]): boolean {
+										let currentNode = $head;
+										while (currentNode) {
+											if (nodeTypes.includes(currentNode.parent.type.name)) {
+												return true;
+											}
+											if (!currentNode.depth) break; // Stop if we reach the top
+											currentNode = state.doc.resolve(currentNode.before()); // Move to the parent node
 										}
-										if (!currentNode.depth) break; // Stop if we reach the top
-										currentNode = state.doc.resolve(currentNode.before()); // Move to the parent node
+										return false;
 									}
-									return false;
-								}
 
-								const isInCodeBlock = isInside(['codeBlock']);
-								const isInList = isInside(['listItem', 'bulletList', 'orderedList']);
-								const isInHeading = isInside(['heading']);
+									const isInCodeBlock = isInside(['codeBlock']);
+									const isInList = isInside(['listItem', 'bulletList', 'orderedList']);
+									const isInHeading = isInside(['heading']);
 
-								if (isInCodeBlock || isInList || isInHeading) {
-									// Let ProseMirror handle the normal Enter behavior
-									return false;
+									if (isInCodeBlock || isInList || isInHeading) {
+										// Let ProseMirror handle the normal Enter behavior
+										return false;
+									}
 								}
 							}
 

src/lib/components/common/Tooltip.svelte

@@ -2,10 +2,8 @@
 	import DOMPurify from 'dompurify';
 
 	import { onDestroy } from 'svelte';
-	import { marked } from 'marked';
 
 	import tippy from 'tippy.js';
-	import { roundArrow } from 'tippy.js';
 
 	export let placement = 'top';
 	export let content = `I'm a tooltip!`;
@@ -47,6 +45,6 @@
 	});
 </script>
 
-<div bind:this={tooltipElement} aria-label={DOMPurify.sanitize(content)} class={className}>
+<div bind:this={tooltipElement} class={className}>
 	<slot />
 </div>

src/lib/components/icons/ArrowTurnDownRight.svelte

@@ -0,0 +1,12 @@
+<script lang="ts">
+	export let className = 'size-4';
+	export let strokeWidth = '1.5';
+</script>
+
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16" fill="currentColor" class={className}>
+	<path
+		fill-rule="evenodd"
+		d="M2.75 2a.75.75 0 0 1 .75.75v6.5h7.94l-.97-.97a.75.75 0 0 1 1.06-1.06l2.25 2.25a.75.75 0 0 1 0 1.06l-2.25 2.25a.75.75 0 1 1-1.06-1.06l.97-.97H2.75A.75.75 0 0 1 2 10V2.75A.75.75 0 0 1 2.75 2Z"
+		clip-rule="evenodd"
+	/>
+</svg>