Merge pull request #12715 from open-webui/dev

0.6.3

.github/ISSUE_TEMPLATE/bug_report.yaml

@@ -27,6 +27,8 @@ body:
       options:
         - label: I have searched the existing issues and discussions.
           required: true
+        - label: I am using the latest version of Open WebUI.
+          required: true
 
   - type: dropdown
     id: installation-method
@@ -83,9 +85,9 @@ body:
           required: true
         - label: I am using the latest version of **both** Open WebUI and Ollama.
           required: true
-        - label: I have checked the browser console logs.
+        - label: I have included the browser console logs.
           required: true
-        - label: I have checked the Docker container logs.
+        - label: I have included the Docker container logs.
           required: true
         - label: I have listed steps to reproduce the bug in detail.
           required: true
@@ -110,7 +112,7 @@ body:
     id: reproduction-steps
     attributes:
       label: Steps to Reproduce
-      description: Provide step-by-step instructions to reproduce the issue.
+      description: Providing clear, step-by-step instructions helps us reproduce and fix the issue faster. If we can't reproduce it, we can't fix it.
       placeholder: |
         1. Go to '...'
         2. Click on '...'

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1 @@
+blank_issues_enabled: false

.github/dependabot.yml

@@ -1,12 +1,26 @@
 version: 2
 updates:
+  - package-ecosystem: uv
+    directory: '/'
+    schedule:
+      interval: monthly
+    target-branch: 'dev'
+
   - package-ecosystem: pip
     directory: '/backend'
     schedule:
       interval: monthly
     target-branch: 'dev'
+
+  - package-ecosystem: npm
+    directory: '/'
+    schedule:
+      interval: monthly
+    target-branch: 'dev'
+
   - package-ecosystem: 'github-actions'
     directory: '/'
     schedule:
       # Check for updates to GitHub Actions every week
       interval: monthly
+    target-branch: 'dev'

.github/pull_request_template.md

@@ -9,9 +9,9 @@
 - [ ] **Changelog:** Ensure a changelog entry following the format of [Keep a Changelog](https://keepachangelog.com/) is added at the bottom of the PR description.
 - [ ] **Documentation:** Have you updated relevant documentation [Open WebUI Docs](https://github.com/open-webui/docs), or other documentation sources?
 - [ ] **Dependencies:** Are there any new dependencies? Have you updated the dependency versions in the documentation?
-- [ ] **Testing:** Have you written and run sufficient tests for validating the changes?
+- [ ] **Testing:** Have you written and run sufficient tests to validate the changes?
 - [ ] **Code review:** Have you performed a self-review of your code, addressing any coding standard issues and ensuring adherence to the project's coding standards?
-- [ ] **Prefix:** To cleary categorize this pull request, prefix the pull request title, using one of the following:
+- [ ] **Prefix:** To clearly categorize this pull request, prefix the pull request title using one of the following:
   - **BREAKING CHANGE**: Significant changes that may affect compatibility
   - **build**: Changes that affect the build system or external dependencies
   - **ci**: Changes to our continuous integration processes or workflows
@@ -22,7 +22,7 @@
   - **i18n**: Internationalization or localization changes
   - **perf**: Performance improvement
   - **refactor**: Code restructuring for better maintainability, readability, or scalability
-  - **style**: Changes that do not affect the meaning of the code (white-space, formatting, missing semi-colons, etc.)
+  - **style**: Changes that do not affect the meaning of the code (white space, formatting, missing semi-colons, etc.)
   - **test**: Adding missing tests or correcting existing tests
   - **WIP**: Work in progress, a temporary label for incomplete or ongoing work
 

.github/workflows/format-backend.yaml

@@ -5,10 +5,18 @@ on:
     branches:
       - main
       - dev
+    paths:
+      - 'backend/**'
+      - 'pyproject.toml'
+      - 'uv.lock'
   pull_request:
     branches:
       - main
       - dev
+    paths:
+      - 'backend/**'
+      - 'pyproject.toml'
+      - 'uv.lock'
 
 jobs:
   build:
@@ -17,7 +25,9 @@ jobs:
 
     strategy:
       matrix:
-        python-version: [3.11]
+        python-version:
+          - 3.11.x
+          - 3.12.x
 
     steps:
       - uses: actions/checkout@v4
@@ -25,7 +35,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: ${{ matrix.python-version }}
+          python-version: '${{ matrix.python-version }}'
 
       - name: Install dependencies
         run: |

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

@@ -5,10 +5,18 @@ on:
     branches:
       - main
       - dev
+    paths-ignore:
+      - 'backend/**'
+      - 'pyproject.toml'
+      - 'uv.lock'
   pull_request:
     branches:
       - main
       - dev
+    paths-ignore:
+      - 'backend/**'
+      - 'pyproject.toml'
+      - 'uv.lock'
 
 jobs:
   build:
@@ -21,7 +29,7 @@ jobs:
       - name: Setup Node.js
         uses: actions/setup-node@v4
         with:
-          node-version: '22' # Or specify any other version you want to use
+          node-version: '22'
 
       - name: Install Dependencies
         run: npm install

CHANGELOG.md

@@ -5,6 +5,135 @@ 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).
 
+## [0.6.3] - 2025-04-12
+
+### Added
+
+- 🧪 **Auto-Artifact Detection Toggle**: Automatically detects artifacts in results—but now you can disable this behavior under advanced settings for full control.
+- 🖼️ **Widescreen Mode for Shared Chats**: Shared link conversations now support widescreen layouts—perfect for presentations or easier review across wider displays.
+- 🔁 **Reindex Knowledge Files on Demand**: Admins can now trigger reindexing of all knowledge files after changing embeddings—ensuring immediate alignment with new models for optimal RAG performance.
+- 📄 **OpenAPI YAML Format Support**: External tools can now use YAML-format OpenAPI specs—making integration simpler for developers familiar with YAML-based configurations.
+- 💬 **Message Content Copy Behavior**: Copy action now excludes 'details' tags—streamlining clipboard content when sharing or pasting summaries elsewhere.
+- 🧭 **Sougou Web Search Integration**: New search engine option added—enhancing global relevance and diversity of search sources for multilingual users.
+- 🧰 **Frontend Web Loader Engine Configuration**: Admins can now set preferred web loader engine for RAG workflows directly from the frontend—offering more control across setups.
+- 👥 **Multi-Model Chat Permission Control**: Admins can manage access to multi-model chats per user group—allowing tighter governance in team environments.
+- 🧱 **Persistent Configuration Can Be Disabled**: New environment variable lets advanced users and hosts turn off persistent configs—ideal for volatile or stateless deployments.
+- 🧠 **Elixir Code Highlighting Support**: Elixir syntax is now beautifully rendered in code blocks—perfect for developers using this language in AI or automation projects.
+- 🌐 **PWA External Manifest URL Support**: You can now define an external manifest.json—integrate Open WebUI seamlessly in managed or proxy-based PWA environments like Cloudflare Zero Trust.
+- 🧪 **Azure AI Speech-to-Text Provider Integration**: Easily transcribe large audio files (up to 200MB) with high accuracy using Microsoft's Azure STT—fully configurable in Audio Settings.
+- 🔏 **PKCE (Code Challenge Method) Support for OIDC**: Enhance your OIDC login security with Proof Key for Code Exchange—ideal for zero-trust and native client apps.
+- ✨ **General UI/UX Enhancements**: Numerous refinements across layout, styling, and tool interactions—reducing visual noise and improving overall usability across key workflows.
+- 🌍 **Translation Updates Across Multiple Languages**: Refined Catalan, Russian, Chinese (Simplified & Traditional), Hungarian, and Spanish translations for clearer navigation and instructions globally.
+
+### Fixed
+
+- 💥 **Chat Completion Error with Missing Models Resolved**: Fixed internal server error when referencing a model that doesn’t exist—ensuring graceful fallback and clear error guidance.
+- 🔧 **Correct Knowledge Base Citations Restored**: Citations generated by RAG workflows now show accurate references—ensuring verifiability in outputs from sourced content.
+- 🎙️ **Broken OGG/WebM Audio Upload Handling for OpenAI Fixed**: Uploading OGG or WebM files now converts properly to WAV before transcription—restoring accurate AI speech recognition workflows.
+- 🔐 **Tool Server 'Session' Authentication Restored**: Previously broken session auth on external tool servers is now fully functional—ensuring secure and seamless access to connected tools.
+- 🌐 **Folder-Based Chat Rename Now Updates Correctly**: Renaming chats in folders now reflects instantly everywhere—improving chat organization and clarity.
+- 📜 **KaTeX Overflow Displays Fixed**: Math expressions now stay neatly within message bounds—preserving layout consistency even with long formulas.
+- 🚫 **Stopping Ongoing Chat Fixed**: You can now return to an active (ongoing) chat and stop generation at any time—ensuring full control over sessions.
+- 🔧 **TOOL_SERVERS / TOOL_SERVER_CONNECTIONS Indexing Issue Fixed**: Fixed a mismatch between tool lists and their access paths—restoring full function and preventing confusion in tool management.
+- 🔐 **LDAP Login Handles Multiple Emails**: When LDAP returns multiple email attributes, the first valid one is now used—ensuring login success and account consistency.
+- 🧩 **Model Visibility Toggle Fix**: Toggling model visibility now works even for untouched models—letting admins smoothly manage user access across base models.
+- ⚙️ **Cross-Origin manifest.json Now Loads Properly**: Compatibility issues with Cloudflare Zero Trust (and others) resolved, allowing manifest.json to load behind authenticated proxies.
+
+### Changed
+
+- 🔒 **Default Access Scopes Set to Private for All Resources**: Models, tools, and knowledge are now private by default when created—ensuring better baseline security and visibility controls.
+- 🧱 **General Backend Refactoring for Stability**: Numerous invisible improvements enhance backend scalability, security, and maintainability—powering upcoming features with a stronger foundation.
+- 🧩 **Stable Dependency Upgrades**: Updated key platform libraries—Chromadb (0.6.3), pgvector (0.4.0), Azure Identity (1.21.0), and Youtube Transcript API (1.0.3)—for improved compatibility, functionality, and security.
+
+## [0.6.2] - 2025-04-06
+
+### Added
+
+- 🌍 **Improved Global Language Support**: Expanded and refined translations across multiple languages to enhance clarity and consistency for international users.
+
+### Fixed
+
+- 🛠️ **Accurate Tool Descriptions from OpenAPI Servers**: External tools now use full endpoint descriptions instead of summaries when generating tool specifications—helping AI models understand tool purpose more precisely and choose the right tool more accurately in tool workflows.
+- 🔧 **Precise Web Results Source Attribution**: Fixed a key issue where all web search results showed the same source ID—now each result gets its correct and distinct source, ensuring accurate citations and traceability.
+- 🔍 **Clean Web Search Retrieval**: Web search now retains only results from URLs where real content was successfully fetched—improving accuracy and removing empty or broken links from citations.
+- 🎵 **Audio File Upload Response Restored**: Resolved an issue where uploading audio files did not return valid responses, restoring smooth file handling for transcription and audio-based workflows.
+
+### Changed
+
+- 🧰 **General Backend Refactoring**: Multiple behind-the-scenes improvements streamline backend performance, reduce complexity, and ensure a more stable, maintainable system overall—making everything smoother without changing your workflow.
+
+## [0.6.1] - 2025-04-05
+
+### Added
+
+- 🛠️ **Global Tool Servers Configuration**: Admins can now centrally configure global external tool servers from Admin Settings > Tools, allowing seamless sharing of tool integrations across all users without manual setup per user.
+- 🔐 **Direct Tool Usage Permission for Users**: Introduced a new user-level permission toggle that grants non-admin users access to direct external tools, empowering broader team collaboration while maintaining control.
+- 🧠 **Mistral OCR Content Extraction Support**: Added native support for Mistral OCR as a high-accuracy document loader, drastically improving text extraction from scanned documents in RAG workflows.
+- 🖼️ **Tools Indicator UI Redesign**: Enhanced message input now smartly displays both built-in and external tools via a unified dropdown, making it simpler and more intuitive to activate tools during conversations.
+- 📄 **RAG Prompt Improved and More Coherent**: Default RAG system prompt has been revised to be more clear and citation-focused—admins can leave the template field empty to use this new gold-standard prompt.
+- 🧰 **Performance & Developer Improvements**: Major internal restructuring of several tool-related components, simplifying styling and merging external/internal handling logic, resulting in better maintainability and performance.
+- 🌍 **Improved Translations**: Updated translations for Tibetan, Polish, Chinese (Simplified & Traditional), Arabic, Russian, Ukrainian, Dutch, Finnish, and French to improve clarity and consistency across the interface.
+
+### Fixed
+
+- 🔑 **External Tool Server API Key Bug Resolved**: Fixed a critical issue where authentication headers were not being sent when calling tools from external OpenAPI tool servers, ensuring full security and smooth tool operations.
+- 🚫 **Conditional Export Button Visibility**: UI now gracefully hides export buttons when there's nothing to export in models, prompts, tools, or functions, improving visual clarity and reducing confusion.
+- 🧪 **Hybrid Search Failure Recovery**: Resolved edge case in parallel hybrid search where empty or unindexed collections caused backend crashes—these are now cleanly skipped to ensure system stability.
+- 📂 **Admin Folder Deletion Fix**: Addressed an issue where folders created in the admin workspace couldn't be deleted, restoring full organizational flexibility for admins.
+- 🔐 **Improved Generic Error Feedback on Login**: Authentication errors now show simplified, non-revealing messages for privacy and improved UX, especially with federated logins.
+- 📝 **Tool Message with Images Improved**: Enhanced how tool-generated messages with image outputs are shown in chat, making them more readable and consistent with the overall UI design.
+- ⚙️ **Auto-Exclusion for Broken RAG Collections**: Auto-skips document collections that fail to fetch data or return "None", preventing silent errors and streamlining retrieval workflows.
+- 📝 **Docling Text File Handling Fix**: Fixed file parsing inconsistency that broke docling-based RAG functionality for certain plain text files, ensuring wider file compatibility.
+
+## [0.6.0] - 2025-03-31
+
+### Added
+
+- 🧩 **External Tool Server Support via OpenAPI**: Connect Open WebUI to any OpenAPI-compatible REST server instantly—offering immediate integration with thousands of developer tools, SDKs, and SaaS systems for powerful extensibility. Learn more: https://github.com/open-webui/openapi-servers
+- 🛠️ **MCP Server Support via MCPO**: You can now convert and expose your internal MCP tools as interoperable OpenAPI HTTP servers within Open WebUI for seamless, plug-n-play AI toolchain creation. Learn more: https://github.com/open-webui/mcpo
+- 📨 **/messages Chat API Endpoint Support**: For power users building external AI systems, new endpoints allow precise control of messages asynchronously—feed long-running external responses into Open WebUI chats without coupling with the frontend.
+- 📝 **Client-Side PDF Generation**: PDF exports are now generated fully client-side for drastically improved output quality—perfect for saving conversations or documents.
+- 💼 **Enforced Temporary Chats Mode**: Admins can now enforce temporary chat sessions by default to align with stringent data retention and compliance requirements.
+- 🌍 **Public Resource Sharing Permission Controls**: Fine-grained user group permissions now allow enabling/disabling public sharing for models, knowledge, prompts, and tools—ideal for privacy, team control, and internal deployments.
+- 📦 **Custom pip Options for Tools/Functions**: You can now specify custom pip installation options with "PIP_OPTIONS", "PIP_PACKAGE_INDEX_OPTIONS" environment variables—improving compatibility, support for private indexes, and better control over Python environments.
+- 🔢 **Editable Message Counter**: You can now double-click the message count number and jump straight to editing the index—quickly navigate complex chats or regenerate specific messages precisely.
+- 🧠 **Embedding Prefix Support Added**: Add custom prefixes to your embeddings for instruct-style tokens, enabling stronger model alignment and more consistent RAG performance.
+- 🙈 **Ability to Hide Base Models**: Optionally hide base models from the UI, helping users streamline model visibility and limit access to only usable endpoints..
+- 📚 **Docling Content Extraction Support**: Open WebUI now supports Docling as a content extraction engine, enabling smarter and more accurate parsing of complex file formats—ideal for advanced document understanding and Retrieval-Augmented Generation (RAG) workflows.
+- 🗃️ **Redis Sentinel Support Added**: Enhance deployment redundancy with support for Redis Sentinel for highly available, failover-safe Redis-based caching or pub/sub.
+- 📚 **JSON Schema Format for Ollama**: Added support for defining the format using JSON schema in Ollama-compatible models, improving flexibility and validation of model outputs.
+- 🔍 **Chat Sidebar Search "Clear” Button**: Quickly clear search filters in chat sidebar using the new ✖️ button—streamline your chat navigation with one click.
+- 🗂️ **Auto-Focus + Enter Submit for Folder Name**: When creating a new folder, the system automatically enters rename mode with name preselected—simplifying your org workflow.
+- 🧱 **Markdown Alerts Rendering**: Blockquotes with syntax hinting (e.g. ⚠️, ℹ️, ✅) now render styled Markdown alert banners, making messages and documentation more visually structured.
+- 🔁 **Hybrid Search Runs in Parallel Now**: Hybrid (BM25 + embedding) search components now run in parallel—dramatically reducing response times and speeding up document retrieval.
+- 📋 **Cleaner UI for Tool Call Display**: Optimized the visual layout of called tools inside chat messages for better clarity and reduced visual clutter.
+- 🧪 **Playwright Timeout Now Configurable**: Default timeout for Playwright processes is now shorter and adjustable via environment variables—making web scraping more robust and tunable to environments.
+- 📈 **OpenTelemetry Support for Observability**: Open WebUI now integrates with OpenTelemetry, allowing you to connect with tools like Grafana, Jaeger, or Prometheus for detailed performance insights and real-time visibility—entirely opt-in and fully self-hosted. Even if enabled, no data is ever sent to us, ensuring your privacy and ownership over all telemetry data.
+- 🛠 **General UI Enhancements & UX Polish**: Numerous refinements across sidebar, code blocks, modal interactions, button alignment, scrollbar visibility, and folder behavior improve overall fluidity and usability of the interface.
+- 🧱 **General Backend Refactoring**: Numerous backend components have been refactored to improve stability, maintainability, and performance—ensuring a more consistent and reliable system across all features.
+- 🌍 **Internationalization Language Support Updates**: Added Estonian and Galician languages, improved Spanish (fully revised), Traditional Chinese, Simplified Chinese, Turkish, Catalan, Ukrainian, and German for a more localized and inclusive interface.
+
+### Fixed
+
+- 🧑‍💻 **Firefox Input Height Bug**: Text input in Firefox now maintains proper height, ensuring message boxes look consistent and behave predictably.
+- 🧾 **Tika Blank Line Bug**: PDFs processed with Apache Tika 3.1.0.0 no longer introduce excessive blank lines—improving RAG output quality and visual cleanliness.
+- 🧪 **CSV Loader Encoding Issues**: CSV files with unknown encodings now automatically detect character sets, resolving import errors in non-UTF-8 datasets.
+- ✅ **LDAP Auth Config Fix**: Path to certificate file is now optional for LDAP setups, fixing authentication trouble for users without preconfigured cert paths.
+- 📥 **File Deletion in Bypass Mode**: Resolved issue where files couldn’t be deleted from knowledge when “bypass embedding” mode was enabled.
+- 🧩 **Hybrid Search Result Sorting & Deduplication Fixed**: Fixed citation and sorting issues in RAG hybrid and reranker modes, ensuring retrieved documents are shown in correct order per score.
+- 🧷 **Model Export/Import Broken for a Single Model**: Fixed bug where individual models couldn’t be exported or re-imported, restoring full portability.
+- 📫 **Auth Redirect Fix**: Logged-in users are now routed properly without unnecessary login prompts when already authenticated.
+
+### Changed
+
+- 🧠 **Prompt Autocompletion Disabled By Default**: Autocomplete suggestions while typing are now disabled unless explicitly re-enabled in user preferences—reduces distractions while composing prompts for advanced users.
+- 🧾 **Normalize Citation Numbering**: Source citations now properly begin from "1" instead of "0"—improving consistency and professional presentation in AI outputs.
+- 📚 **Improved Error Handling from Pipelines**: Pipelines now show the actual returned error message from failed tasks rather than generic "Connection closed"—making debugging far more user-friendly.
+
+### Removed
+
+- 🧾 **ENABLE_AUDIT_LOGS Setting Removed**: Deprecated setting “ENABLE_AUDIT_LOGS” has been fully removed—now controlled via “AUDIT_LOG_LEVEL” instead.
+
 ## [0.5.20] - 2025-03-05
 
 ### Added

Dockerfile

@@ -132,7 +132,7 @@ RUN if [ "$USE_OLLAMA" = "true" ]; then \
 # install python dependencies
 COPY --chown=$UID:$GID ./backend/requirements.txt ./requirements.txt
 
-RUN pip3 install uv && \
+RUN pip3 install --no-cache-dir uv && \
     if [ "$USE_CUDA" = "true" ]; then \
     # If you use CUDA the whisper and embedding model will be downloaded on first use
     pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/$USE_CUDA_DOCKER_VER --no-cache-dir && \

backend/open_webui/__init__.py

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

backend/open_webui/config.py

@@ -3,6 +3,7 @@ import logging
 import os
 import shutil
 import base64
+import redis
 
 from datetime import datetime
 from pathlib import Path
@@ -17,6 +18,9 @@ from open_webui.env import (
     DATA_DIR,
     DATABASE_URL,
     ENV,
+    REDIS_URL,
+    REDIS_SENTINEL_HOSTS,
+    REDIS_SENTINEL_PORT,
     FRONTEND_BUILD_DIR,
     OFFLINE_MODE,
     OPEN_WEBUI_DIR,
@@ -26,6 +30,7 @@ from open_webui.env import (
     log,
 )
 from open_webui.internal.db import Base, get_db
+from open_webui.utils.redis import get_redis_connection
 
 
 class EndpointFilter(logging.Filter):
@@ -196,6 +201,10 @@ def save_config(config):
 
 T = TypeVar("T")
 
+ENABLE_PERSISTENT_CONFIG = (
+    os.environ.get("ENABLE_PERSISTENT_CONFIG", "True").lower() == "true"
+)
+
 
 class PersistentConfig(Generic[T]):
     def __init__(self, env_name: str, config_path: str, env_value: T):
@@ -203,7 +212,7 @@ class PersistentConfig(Generic[T]):
         self.config_path = config_path
         self.env_value = env_value
         self.config_value = get_config_value(config_path)
-        if self.config_value is not None:
+        if self.config_value is not None and ENABLE_PERSISTENT_CONFIG:
             log.info(f"'{env_name}' loaded from the latest database entry")
             self.value = self.config_value
         else:
@@ -248,9 +257,17 @@ class PersistentConfig(Generic[T]):
 
 class AppConfig:
     _state: dict[str, PersistentConfig]
+    _redis: Optional[redis.Redis] = None
 
-    def __init__(self):
+    def __init__(
+        self, redis_url: Optional[str] = None, redis_sentinels: Optional[list] = []
+    ):
         super().__setattr__("_state", {})
+        if redis_url:
+            super().__setattr__(
+                "_redis",
+                get_redis_connection(redis_url, redis_sentinels, decode_responses=True),
+            )
 
     def __setattr__(self, key, value):
         if isinstance(value, PersistentConfig):
@@ -259,7 +276,31 @@ class AppConfig:
             self._state[key].value = value
             self._state[key].save()
 
+            if self._redis:
+                redis_key = f"open-webui:config:{key}"
+                self._redis.set(redis_key, json.dumps(self._state[key].value))
+
     def __getattr__(self, key):
+        if key not in self._state:
+            raise AttributeError(f"Config key '{key}' not found")
+
+        # If Redis is available, check for an updated value
+        if self._redis:
+            redis_key = f"open-webui:config:{key}"
+            redis_value = self._redis.get(redis_key)
+
+            if redis_value is not None:
+                try:
+                    decoded_value = json.loads(redis_value)
+
+                    # Update the in-memory value if different
+                    if self._state[key].value != decoded_value:
+                        self._state[key].value = decoded_value
+                        log.info(f"Updated {key} from Redis: {decoded_value}")
+
+                except json.JSONDecodeError:
+                    log.error(f"Invalid JSON format in Redis for {key}: {redis_value}")
+
         return self._state[key].value
 
 
@@ -294,12 +335,14 @@ JWT_EXPIRES_IN = PersistentConfig(
 # OAuth config
 ####################################
 
+
 ENABLE_OAUTH_SIGNUP = PersistentConfig(
     "ENABLE_OAUTH_SIGNUP",
     "oauth.enable_signup",
     os.environ.get("ENABLE_OAUTH_SIGNUP", "False").lower() == "true",
 )
 
+
 OAUTH_MERGE_ACCOUNTS_BY_EMAIL = PersistentConfig(
     "OAUTH_MERGE_ACCOUNTS_BY_EMAIL",
     "oauth.merge_accounts_by_email",
@@ -417,6 +460,12 @@ OAUTH_SCOPES = PersistentConfig(
     os.environ.get("OAUTH_SCOPES", "openid email profile"),
 )
 
+OAUTH_CODE_CHALLENGE_METHOD = PersistentConfig(
+    "OAUTH_CODE_CHALLENGE_METHOD",
+    "oauth.oidc.code_challenge_method",
+    os.environ.get("OAUTH_CODE_CHALLENGE_METHOD", None),
+)
+
 OAUTH_PROVIDER_NAME = PersistentConfig(
     "OAUTH_PROVIDER_NAME",
     "oauth.oidc.provider_name",
@@ -429,6 +478,7 @@ OAUTH_USERNAME_CLAIM = PersistentConfig(
     os.environ.get("OAUTH_USERNAME_CLAIM", "name"),
 )
 
+
 OAUTH_PICTURE_CLAIM = PersistentConfig(
     "OAUTH_PICTURE_CLAIM",
     "oauth.oidc.avatar_claim",
@@ -520,7 +570,7 @@ 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",
+                server_metadata_url=f"https://login.microsoftonline.com/{MICROSOFT_CLIENT_TENANT_ID.value}/v2.0/.well-known/openid-configuration?appid={MICROSOFT_CLIENT_ID.value}",
                 client_kwargs={
                     "scope": MICROSOFT_OAUTH_SCOPE.value,
                 },
@@ -561,14 +611,27 @@ def load_oauth_providers():
     ):
 
         def oidc_oauth_register(client):
+            client_kwargs = {
+                "scope": OAUTH_SCOPES.value,
+            }
+
+            if (
+                OAUTH_CODE_CHALLENGE_METHOD.value
+                and OAUTH_CODE_CHALLENGE_METHOD.value == "S256"
+            ):
+                client_kwargs["code_challenge_method"] = "S256"
+            elif OAUTH_CODE_CHALLENGE_METHOD.value:
+                raise Exception(
+                    'Code challenge methods other than "%s" not supported. Given: "%s"'
+                    % ("S256", OAUTH_CODE_CHALLENGE_METHOD.value)
+                )
+
             client.register(
                 name="oidc",
                 client_id=OAUTH_CLIENT_ID.value,
                 client_secret=OAUTH_CLIENT_SECRET.value,
                 server_metadata_url=OPENID_PROVIDER_URL.value,
-                client_kwargs={
-                    "scope": OAUTH_SCOPES.value,
-                },
+                client_kwargs=client_kwargs,
                 redirect_uri=OPENID_REDIRECT_URI.value,
             )
 
@@ -841,6 +904,17 @@ except Exception:
     pass
 OPENAI_API_BASE_URL = "https://api.openai.com/v1"
 
+####################################
+# TOOL_SERVERS
+####################################
+
+
+TOOL_SERVER_CONNECTIONS = PersistentConfig(
+    "TOOL_SERVER_CONNECTIONS",
+    "tool_server.connections",
+    [],
+)
+
 ####################################
 # WEBUI
 ####################################
@@ -943,6 +1017,35 @@ USER_PERMISSIONS_WORKSPACE_TOOLS_ACCESS = (
     os.environ.get("USER_PERMISSIONS_WORKSPACE_TOOLS_ACCESS", "False").lower() == "true"
 )
 
+USER_PERMISSIONS_WORKSPACE_MODELS_ALLOW_PUBLIC_SHARING = (
+    os.environ.get(
+        "USER_PERMISSIONS_WORKSPACE_MODELS_ALLOW_PUBLIC_SHARING", "False"
+    ).lower()
+    == "true"
+)
+
+USER_PERMISSIONS_WORKSPACE_KNOWLEDGE_ALLOW_PUBLIC_SHARING = (
+    os.environ.get(
+        "USER_PERMISSIONS_WORKSPACE_KNOWLEDGE_ALLOW_PUBLIC_SHARING", "False"
+    ).lower()
+    == "true"
+)
+
+USER_PERMISSIONS_WORKSPACE_PROMPTS_ALLOW_PUBLIC_SHARING = (
+    os.environ.get(
+        "USER_PERMISSIONS_WORKSPACE_PROMPTS_ALLOW_PUBLIC_SHARING", "False"
+    ).lower()
+    == "true"
+)
+
+USER_PERMISSIONS_WORKSPACE_TOOLS_ALLOW_PUBLIC_SHARING = (
+    os.environ.get(
+        "USER_PERMISSIONS_WORKSPACE_TOOLS_ALLOW_PUBLIC_SHARING", "False"
+    ).lower()
+    == "true"
+)
+
+
 USER_PERMISSIONS_CHAT_CONTROLS = (
     os.environ.get("USER_PERMISSIONS_CHAT_CONTROLS", "True").lower() == "true"
 )
@@ -959,10 +1062,25 @@ USER_PERMISSIONS_CHAT_EDIT = (
     os.environ.get("USER_PERMISSIONS_CHAT_EDIT", "True").lower() == "true"
 )
 
+USER_PERMISSIONS_CHAT_MULTIPLE_MODELS = (
+    os.environ.get("USER_PERMISSIONS_CHAT_MULTIPLE_MODELS", "True").lower() == "true"
+)
+
 USER_PERMISSIONS_CHAT_TEMPORARY = (
     os.environ.get("USER_PERMISSIONS_CHAT_TEMPORARY", "True").lower() == "true"
 )
 
+USER_PERMISSIONS_CHAT_TEMPORARY_ENFORCED = (
+    os.environ.get("USER_PERMISSIONS_CHAT_TEMPORARY_ENFORCED", "False").lower()
+    == "true"
+)
+
+
+USER_PERMISSIONS_FEATURES_DIRECT_TOOL_SERVERS = (
+    os.environ.get("USER_PERMISSIONS_FEATURES_DIRECT_TOOL_SERVERS", "False").lower()
+    == "true"
+)
+
 USER_PERMISSIONS_FEATURES_WEB_SEARCH = (
     os.environ.get("USER_PERMISSIONS_FEATURES_WEB_SEARCH", "True").lower() == "true"
 )
@@ -985,14 +1103,23 @@ DEFAULT_USER_PERMISSIONS = {
         "prompts": USER_PERMISSIONS_WORKSPACE_PROMPTS_ACCESS,
         "tools": USER_PERMISSIONS_WORKSPACE_TOOLS_ACCESS,
     },
+    "sharing": {
+        "public_models": USER_PERMISSIONS_WORKSPACE_MODELS_ALLOW_PUBLIC_SHARING,
+        "public_knowledge": USER_PERMISSIONS_WORKSPACE_KNOWLEDGE_ALLOW_PUBLIC_SHARING,
+        "public_prompts": USER_PERMISSIONS_WORKSPACE_PROMPTS_ALLOW_PUBLIC_SHARING,
+        "public_tools": USER_PERMISSIONS_WORKSPACE_TOOLS_ALLOW_PUBLIC_SHARING,
+    },
     "chat": {
         "controls": USER_PERMISSIONS_CHAT_CONTROLS,
         "file_upload": USER_PERMISSIONS_CHAT_FILE_UPLOAD,
         "delete": USER_PERMISSIONS_CHAT_DELETE,
         "edit": USER_PERMISSIONS_CHAT_EDIT,
+        "multiple_models": USER_PERMISSIONS_CHAT_MULTIPLE_MODELS,
         "temporary": USER_PERMISSIONS_CHAT_TEMPORARY,
+        "temporary_enforced": USER_PERMISSIONS_CHAT_TEMPORARY_ENFORCED,
     },
     "features": {
+        "direct_tool_servers": USER_PERMISSIONS_FEATURES_DIRECT_TOOL_SERVERS,
         "web_search": USER_PERMISSIONS_FEATURES_WEB_SEARCH,
         "image_generation": USER_PERMISSIONS_FEATURES_IMAGE_GENERATION,
         "code_interpreter": USER_PERMISSIONS_FEATURES_CODE_INTERPRETER,
@@ -1055,6 +1182,12 @@ ENABLE_MESSAGE_RATING = PersistentConfig(
     os.environ.get("ENABLE_MESSAGE_RATING", "True").lower() == "true",
 )
 
+ENABLE_USER_WEBHOOKS = PersistentConfig(
+    "ENABLE_USER_WEBHOOKS",
+    "ui.enable_user_webhooks",
+    os.environ.get("ENABLE_USER_WEBHOOKS", "True").lower() == "true",
+)
+
 
 def validate_cors_origins(origins):
     for origin in origins:
@@ -1276,7 +1409,7 @@ Strictly return in JSON format:
 ENABLE_AUTOCOMPLETE_GENERATION = PersistentConfig(
     "ENABLE_AUTOCOMPLETE_GENERATION",
     "task.autocomplete.enable",
-    os.environ.get("ENABLE_AUTOCOMPLETE_GENERATION", "True").lower() == "true",
+    os.environ.get("ENABLE_AUTOCOMPLETE_GENERATION", "False").lower() == "true",
 )
 
 AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH = PersistentConfig(
@@ -1548,8 +1681,10 @@ QDRANT_API_KEY = os.environ.get("QDRANT_API_KEY", None)
 
 # OpenSearch
 OPENSEARCH_URI = os.environ.get("OPENSEARCH_URI", "https://localhost:9200")
-OPENSEARCH_SSL = os.environ.get("OPENSEARCH_SSL", True)
-OPENSEARCH_CERT_VERIFY = os.environ.get("OPENSEARCH_CERT_VERIFY", False)
+OPENSEARCH_SSL = os.environ.get("OPENSEARCH_SSL", "true").lower() == "true"
+OPENSEARCH_CERT_VERIFY = (
+    os.environ.get("OPENSEARCH_CERT_VERIFY", "false").lower() == "true"
+)
 OPENSEARCH_USERNAME = os.environ.get("OPENSEARCH_USERNAME", None)
 OPENSEARCH_PASSWORD = os.environ.get("OPENSEARCH_PASSWORD", None)
 
@@ -1623,6 +1758,12 @@ TIKA_SERVER_URL = PersistentConfig(
     os.getenv("TIKA_SERVER_URL", "http://tika:9998"),  # Default for sidecar deployment
 )
 
+DOCLING_SERVER_URL = PersistentConfig(
+    "DOCLING_SERVER_URL",
+    "rag.docling_server_url",
+    os.getenv("DOCLING_SERVER_URL", "http://docling:5001"),
+)
+
 DOCUMENT_INTELLIGENCE_ENDPOINT = PersistentConfig(
     "DOCUMENT_INTELLIGENCE_ENDPOINT",
     "rag.document_intelligence_endpoint",
@@ -1635,6 +1776,11 @@ DOCUMENT_INTELLIGENCE_KEY = PersistentConfig(
     os.getenv("DOCUMENT_INTELLIGENCE_KEY", ""),
 )
 
+MISTRAL_OCR_API_KEY = PersistentConfig(
+    "MISTRAL_OCR_API_KEY",
+    "rag.mistral_ocr_api_key",
+    os.getenv("MISTRAL_OCR_API_KEY", ""),
+)
 
 BYPASS_EMBEDDING_AND_RETRIEVAL = PersistentConfig(
     "BYPASS_EMBEDDING_AND_RETRIEVAL",
@@ -1646,6 +1792,11 @@ BYPASS_EMBEDDING_AND_RETRIEVAL = PersistentConfig(
 RAG_TOP_K = PersistentConfig(
     "RAG_TOP_K", "rag.top_k", int(os.environ.get("RAG_TOP_K", "3"))
 )
+RAG_TOP_K_RERANKER = PersistentConfig(
+    "RAG_TOP_K_RERANKER",
+    "rag.top_k_reranker",
+    int(os.environ.get("RAG_TOP_K_RERANKER", "3")),
+)
 RAG_RELEVANCE_THRESHOLD = PersistentConfig(
     "RAG_RELEVANCE_THRESHOLD",
     "rag.relevance_threshold",
@@ -1684,12 +1835,6 @@ RAG_FILE_MAX_SIZE = PersistentConfig(
     ),
 )
 
-ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION = PersistentConfig(
-    "ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION",
-    "rag.enable_web_loader_ssl_verification",
-    os.environ.get("ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION", "True").lower() == "true",
-)
-
 RAG_EMBEDDING_ENGINE = PersistentConfig(
     "RAG_EMBEDDING_ENGINE",
     "rag.embedding_engine",
@@ -1727,6 +1872,14 @@ RAG_EMBEDDING_BATCH_SIZE = PersistentConfig(
     ),
 )
 
+RAG_EMBEDDING_QUERY_PREFIX = os.environ.get("RAG_EMBEDDING_QUERY_PREFIX", None)
+
+RAG_EMBEDDING_CONTENT_PREFIX = os.environ.get("RAG_EMBEDDING_CONTENT_PREFIX", None)
+
+RAG_EMBEDDING_PREFIX_FIELD_NAME = os.environ.get(
+    "RAG_EMBEDDING_PREFIX_FIELD_NAME", None
+)
+
 RAG_RERANKING_MODEL = PersistentConfig(
     "RAG_RERANKING_MODEL",
     "rag.reranking_model",
@@ -1770,7 +1923,7 @@ CHUNK_OVERLAP = PersistentConfig(
 )
 
 DEFAULT_RAG_TEMPLATE = """### Task:
-Respond to the user query using the provided context, incorporating inline citations in the format [source_id] **only when the <source_id> tag is explicitly provided** in the context.
+Respond to the user query using the provided context, incorporating inline citations in the format [id] **only when the <source> tag includes an explicit id attribute** (e.g., <source id="1">).
 
 ### Guidelines:
 - If you don't know the answer, clearly state that.
@@ -1778,18 +1931,17 @@ Respond to the user query using the provided context, incorporating inline citat
 - Respond in the same language as the user's query.
 - If the context is unreadable or of poor quality, inform the user and provide the best possible answer.
 - If the answer isn't present in the context but you possess the knowledge, explain this to the user and provide the answer using your own understanding.
-- **Only include inline citations using [source_id] (e.g., [1], [2]) when a `<source_id>` tag is explicitly provided in the context.**
-- Do not cite if the <source_id> tag is not provided in the context.  
+- **Only include inline citations using [id] (e.g., [1], [2]) when the <source> tag includes an id attribute.**
+- Do not cite if the <source> tag does not contain an id attribute.
 - Do not use XML tags in your response.
 - Ensure citations are concise and directly related to the information provided.
 
 ### Example of Citation:
-If the user asks about a specific topic and the information is found in "whitepaper.pdf" with a provided <source_id>, the response should include the citation like so:  
-* "According to the study, the proposed method increases efficiency by 20% [whitepaper.pdf]."
-If no <source_id> is present, the response should omit the citation.
+If the user asks about a specific topic and the information is found in a source with a provided id attribute, the response should include the citation like in the following example:
+* "According to the study, the proposed method increases efficiency by 20% [1]."
 
 ### Output:
-Provide a clear and direct response to the user's query, including inline citations in the format [source_id] only when the <source_id> tag is present in the context.
+Provide a clear and direct response to the user's query, including inline citations in the format [id] only when the <source> tag with id attribute is present in the context.
 
 <context>
 {{CONTEXT}}
@@ -1847,16 +1999,20 @@ YOUTUBE_LOADER_PROXY_URL = PersistentConfig(
 )
 
 
-ENABLE_RAG_WEB_SEARCH = PersistentConfig(
-    "ENABLE_RAG_WEB_SEARCH",
+####################################
+# Web Search (RAG)
+####################################
+
+ENABLE_WEB_SEARCH = PersistentConfig(
+    "ENABLE_WEB_SEARCH",
     "rag.web.search.enable",
-    os.getenv("ENABLE_RAG_WEB_SEARCH", "False").lower() == "true",
+    os.getenv("ENABLE_WEB_SEARCH", "False").lower() == "true",
 )
 
-RAG_WEB_SEARCH_ENGINE = PersistentConfig(
-    "RAG_WEB_SEARCH_ENGINE",
+WEB_SEARCH_ENGINE = PersistentConfig(
+    "WEB_SEARCH_ENGINE",
     "rag.web.search.engine",
-    os.getenv("RAG_WEB_SEARCH_ENGINE", ""),
+    os.getenv("WEB_SEARCH_ENGINE", ""),
 )
 
 BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL = PersistentConfig(
@@ -1865,10 +2021,18 @@ BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL = PersistentConfig(
     os.getenv("BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL", "False").lower() == "true",
 )
 
+
+WEB_SEARCH_RESULT_COUNT = PersistentConfig(
+    "WEB_SEARCH_RESULT_COUNT",
+    "rag.web.search.result_count",
+    int(os.getenv("WEB_SEARCH_RESULT_COUNT", "3")),
+)
+
+
 # You can provide a list of your own websites to filter after performing a web search.
 # This ensures the highest level of safety and reliability of the information sources.
-RAG_WEB_SEARCH_DOMAIN_FILTER_LIST = PersistentConfig(
-    "RAG_WEB_SEARCH_DOMAIN_FILTER_LIST",
+WEB_SEARCH_DOMAIN_FILTER_LIST = PersistentConfig(
+    "WEB_SEARCH_DOMAIN_FILTER_LIST",
     "rag.web.search.domain.filter_list",
     [
         # "wikipedia.com",
@@ -1877,6 +2041,30 @@ RAG_WEB_SEARCH_DOMAIN_FILTER_LIST = PersistentConfig(
     ],
 )
 
+WEB_SEARCH_CONCURRENT_REQUESTS = PersistentConfig(
+    "WEB_SEARCH_CONCURRENT_REQUESTS",
+    "rag.web.search.concurrent_requests",
+    int(os.getenv("WEB_SEARCH_CONCURRENT_REQUESTS", "10")),
+)
+
+WEB_LOADER_ENGINE = PersistentConfig(
+    "WEB_LOADER_ENGINE",
+    "rag.web.loader.engine",
+    os.environ.get("WEB_LOADER_ENGINE", ""),
+)
+
+ENABLE_WEB_LOADER_SSL_VERIFICATION = PersistentConfig(
+    "ENABLE_WEB_LOADER_SSL_VERIFICATION",
+    "rag.web.loader.ssl_verification",
+    os.environ.get("ENABLE_WEB_LOADER_SSL_VERIFICATION", "True").lower() == "true",
+)
+
+WEB_SEARCH_TRUST_ENV = PersistentConfig(
+    "WEB_SEARCH_TRUST_ENV",
+    "rag.web.search.trust_env",
+    os.getenv("WEB_SEARCH_TRUST_ENV", "False").lower() == "true",
+)
+
 
 SEARXNG_QUERY_URL = PersistentConfig(
     "SEARXNG_QUERY_URL",
@@ -1944,12 +2132,6 @@ SERPLY_API_KEY = PersistentConfig(
     os.getenv("SERPLY_API_KEY", ""),
 )
 
-TAVILY_API_KEY = PersistentConfig(
-    "TAVILY_API_KEY",
-    "rag.web.search.tavily_api_key",
-    os.getenv("TAVILY_API_KEY", ""),
-)
-
 JINA_API_KEY = PersistentConfig(
     "JINA_API_KEY",
     "rag.web.search.jina_api_key",
@@ -2006,48 +2188,55 @@ PERPLEXITY_API_KEY = PersistentConfig(
     os.getenv("PERPLEXITY_API_KEY", ""),
 )
 
-RAG_WEB_SEARCH_RESULT_COUNT = PersistentConfig(
-    "RAG_WEB_SEARCH_RESULT_COUNT",
-    "rag.web.search.result_count",
-    int(os.getenv("RAG_WEB_SEARCH_RESULT_COUNT", "3")),
+SOUGOU_API_SID = PersistentConfig(
+    "SOUGOU_API_SID",
+    "rag.web.search.sougou_api_sid",
+    os.getenv("SOUGOU_API_SID", ""),
 )
 
-RAG_WEB_SEARCH_CONCURRENT_REQUESTS = PersistentConfig(
-    "RAG_WEB_SEARCH_CONCURRENT_REQUESTS",
-    "rag.web.search.concurrent_requests",
-    int(os.getenv("RAG_WEB_SEARCH_CONCURRENT_REQUESTS", "10")),
+SOUGOU_API_SK = PersistentConfig(
+    "SOUGOU_API_SK",
+    "rag.web.search.sougou_api_sk",
+    os.getenv("SOUGOU_API_SK", ""),
 )
 
-RAG_WEB_LOADER_ENGINE = PersistentConfig(
-    "RAG_WEB_LOADER_ENGINE",
-    "rag.web.loader.engine",
-    os.environ.get("RAG_WEB_LOADER_ENGINE", "safe_web"),
+TAVILY_API_KEY = PersistentConfig(
+    "TAVILY_API_KEY",
+    "rag.web.search.tavily_api_key",
+    os.getenv("TAVILY_API_KEY", ""),
 )
 
-RAG_WEB_SEARCH_TRUST_ENV = PersistentConfig(
-    "RAG_WEB_SEARCH_TRUST_ENV",
-    "rag.web.search.trust_env",
-    os.getenv("RAG_WEB_SEARCH_TRUST_ENV", "False").lower() == "true",
+TAVILY_EXTRACT_DEPTH = PersistentConfig(
+    "TAVILY_EXTRACT_DEPTH",
+    "rag.web.search.tavily_extract_depth",
+    os.getenv("TAVILY_EXTRACT_DEPTH", "basic"),
 )
 
-PLAYWRIGHT_WS_URI = PersistentConfig(
-    "PLAYWRIGHT_WS_URI",
-    "rag.web.loader.engine.playwright.ws.uri",
-    os.environ.get("PLAYWRIGHT_WS_URI", None),
+PLAYWRIGHT_WS_URL = PersistentConfig(
+    "PLAYWRIGHT_WS_URL",
+    "rag.web.loader.playwright_ws_url",
+    os.environ.get("PLAYWRIGHT_WS_URL", ""),
+)
+
+PLAYWRIGHT_TIMEOUT = PersistentConfig(
+    "PLAYWRIGHT_TIMEOUT",
+    "rag.web.loader.playwright_timeout",
+    int(os.environ.get("PLAYWRIGHT_TIMEOUT", "10000")),
 )
 
 FIRECRAWL_API_KEY = PersistentConfig(
     "FIRECRAWL_API_KEY",
-    "firecrawl.api_key",
+    "rag.web.loader.firecrawl_api_key",
     os.environ.get("FIRECRAWL_API_KEY", ""),
 )
 
 FIRECRAWL_API_BASE_URL = PersistentConfig(
     "FIRECRAWL_API_BASE_URL",
-    "firecrawl.api_url",
+    "rag.web.loader.firecrawl_api_url",
     os.environ.get("FIRECRAWL_API_BASE_URL", "https://api.firecrawl.dev"),
 )
 
+
 ####################################
 # Images
 ####################################
@@ -2331,6 +2520,24 @@ AUDIO_STT_MODEL = PersistentConfig(
     os.getenv("AUDIO_STT_MODEL", ""),
 )
 
+AUDIO_STT_AZURE_API_KEY = PersistentConfig(
+    "AUDIO_STT_AZURE_API_KEY",
+    "audio.stt.azure.api_key",
+    os.getenv("AUDIO_STT_AZURE_API_KEY", ""),
+)
+
+AUDIO_STT_AZURE_REGION = PersistentConfig(
+    "AUDIO_STT_AZURE_REGION",
+    "audio.stt.azure.region",
+    os.getenv("AUDIO_STT_AZURE_REGION", ""),
+)
+
+AUDIO_STT_AZURE_LOCALES = PersistentConfig(
+    "AUDIO_STT_AZURE_LOCALES",
+    "audio.stt.azure.locales",
+    os.getenv("AUDIO_STT_AZURE_LOCALES", ""),
+)
+
 AUDIO_TTS_OPENAI_API_BASE_URL = PersistentConfig(
     "AUDIO_TTS_OPENAI_API_BASE_URL",
     "audio.tts.openai.api_base_url",

backend/open_webui/constants.py

@@ -31,6 +31,7 @@ class ERROR_MESSAGES(str, Enum):
     USERNAME_TAKEN = (
         "Uh-oh! This username is already registered. Please choose another username."
     )
+    PASSWORD_TOO_LONG = "Uh-oh! The password you entered is too long. Please make sure your password is less than 72 bytes long."
     COMMAND_TAKEN = "Uh-oh! This command is already registered. Please choose another command string."
     FILE_EXISTS = "Uh-oh! This file is already registered. Please choose another file."
 

backend/open_webui/env.py

@@ -105,7 +105,6 @@ for source in log_sources:
 
 log.setLevel(SRC_LOG_LEVELS["CONFIG"])
 
-
 WEBUI_NAME = os.environ.get("WEBUI_NAME", "Open WebUI")
 if WEBUI_NAME != "Open WebUI":
     WEBUI_NAME += " (Open WebUI)"
@@ -130,7 +129,6 @@ else:
     except Exception:
         PACKAGE_DATA = {"version": "0.0.0"}
 
-
 VERSION = PACKAGE_DATA["version"]
 
 
@@ -161,7 +159,6 @@ try:
 except Exception:
     changelog_content = (pkgutil.get_data("open_webui", "CHANGELOG.md") or b"").decode()
 
-
 # Convert markdown content to HTML
 html_content = markdown.markdown(changelog_content)
 
@@ -192,7 +189,6 @@ for version in soup.find_all("h2"):
 
     changelog_json[version_number] = version_data
 
-
 CHANGELOG = changelog_json
 
 ####################################
@@ -209,7 +205,6 @@ ENABLE_FORWARD_USER_INFO_HEADERS = (
     os.environ.get("ENABLE_FORWARD_USER_INFO_HEADERS", "False").lower() == "true"
 )
 
-
 ####################################
 # WEBUI_BUILD_HASH
 ####################################
@@ -244,7 +239,6 @@ if FROM_INIT_PY:
 
     DATA_DIR = Path(os.getenv("DATA_DIR", OPEN_WEBUI_DIR / "data"))
 
-
 STATIC_DIR = Path(os.getenv("STATIC_DIR", OPEN_WEBUI_DIR / "static"))
 
 FONTS_DIR = Path(os.getenv("FONTS_DIR", OPEN_WEBUI_DIR / "static" / "fonts"))
@@ -256,7 +250,6 @@ if FROM_INIT_PY:
         os.getenv("FRONTEND_BUILD_DIR", OPEN_WEBUI_DIR / "frontend")
     ).resolve()
 
-
 ####################################
 # Database
 ####################################
@@ -321,7 +314,6 @@ RESET_CONFIG_ON_START = (
     os.environ.get("RESET_CONFIG_ON_START", "False").lower() == "true"
 )
 
-
 ENABLE_REALTIME_CHAT_SAVE = (
     os.environ.get("ENABLE_REALTIME_CHAT_SAVE", "False").lower() == "true"
 )
@@ -330,7 +322,23 @@ ENABLE_REALTIME_CHAT_SAVE = (
 # REDIS
 ####################################
 
-REDIS_URL = os.environ.get("REDIS_URL", "redis://localhost:6379/0")
+REDIS_URL = os.environ.get("REDIS_URL", "")
+REDIS_SENTINEL_HOSTS = os.environ.get("REDIS_SENTINEL_HOSTS", "")
+REDIS_SENTINEL_PORT = os.environ.get("REDIS_SENTINEL_PORT", "26379")
+
+####################################
+# UVICORN WORKERS
+####################################
+
+# Number of uvicorn worker processes for handling requests
+UVICORN_WORKERS = os.environ.get("UVICORN_WORKERS", "1")
+try:
+    UVICORN_WORKERS = int(UVICORN_WORKERS)
+    if UVICORN_WORKERS < 1:
+        UVICORN_WORKERS = 1
+except ValueError:
+    UVICORN_WORKERS = 1
+    log.info(f"Invalid UVICORN_WORKERS value, defaulting to {UVICORN_WORKERS}")
 
 ####################################
 # WEBUI_AUTH (Required for security)
@@ -387,6 +395,10 @@ WEBSOCKET_MANAGER = os.environ.get("WEBSOCKET_MANAGER", "")
 WEBSOCKET_REDIS_URL = os.environ.get("WEBSOCKET_REDIS_URL", REDIS_URL)
 WEBSOCKET_REDIS_LOCK_TIMEOUT = os.environ.get("WEBSOCKET_REDIS_LOCK_TIMEOUT", 60)
 
+WEBSOCKET_SENTINEL_HOSTS = os.environ.get("WEBSOCKET_SENTINEL_HOSTS", "")
+
+WEBSOCKET_SENTINEL_PORT = os.environ.get("WEBSOCKET_SENTINEL_PORT", "26379")
+
 AIOHTTP_CLIENT_TIMEOUT = os.environ.get("AIOHTTP_CLIENT_TIMEOUT", "")
 
 if AIOHTTP_CLIENT_TIMEOUT == "":
@@ -399,19 +411,32 @@ else:
 
 AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = os.environ.get(
     "AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST",
-    os.environ.get("AIOHTTP_CLIENT_TIMEOUT_OPENAI_MODEL_LIST", ""),
+    os.environ.get("AIOHTTP_CLIENT_TIMEOUT_OPENAI_MODEL_LIST", "10"),
 )
 
-
 if AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST == "":
     AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = None
 else:
     try:
         AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = int(AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST)
     except Exception:
-        AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = 5
+        AIOHTTP_CLIENT_TIMEOUT_MODEL_LIST = 10
 
 
+AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = os.environ.get(
+    "AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA", "10"
+)
+
+if AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA == "":
+    AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = None
+else:
+    try:
+        AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = int(
+            AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA
+        )
+    except Exception:
+        AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA = 10
+
 ####################################
 # OFFLINE_MODE
 ####################################
@@ -424,13 +449,12 @@ if OFFLINE_MODE:
 ####################################
 # AUDIT LOGGING
 ####################################
-ENABLE_AUDIT_LOGS = os.getenv("ENABLE_AUDIT_LOGS", "false").lower() == "true"
 # Where to store log file
 AUDIT_LOGS_FILE_PATH = f"{DATA_DIR}/audit.log"
 # Maximum size of a file before rotating into a new log file
 AUDIT_LOG_FILE_ROTATION_SIZE = os.getenv("AUDIT_LOG_FILE_ROTATION_SIZE", "10MB")
 # METADATA | REQUEST | REQUEST_RESPONSE
-AUDIT_LOG_LEVEL = os.getenv("AUDIT_LOG_LEVEL", "REQUEST_RESPONSE").upper()
+AUDIT_LOG_LEVEL = os.getenv("AUDIT_LOG_LEVEL", "NONE").upper()
 try:
     MAX_BODY_LOG_SIZE = int(os.environ.get("MAX_BODY_LOG_SIZE") or 2048)
 except ValueError:
@@ -442,3 +466,33 @@ AUDIT_EXCLUDED_PATHS = os.getenv("AUDIT_EXCLUDED_PATHS", "/chats,/chat,/folders"
 )
 AUDIT_EXCLUDED_PATHS = [path.strip() for path in AUDIT_EXCLUDED_PATHS]
 AUDIT_EXCLUDED_PATHS = [path.lstrip("/") for path in AUDIT_EXCLUDED_PATHS]
+
+####################################
+# OPENTELEMETRY
+####################################
+
+ENABLE_OTEL = os.environ.get("ENABLE_OTEL", "False").lower() == "true"
+OTEL_EXPORTER_OTLP_ENDPOINT = os.environ.get(
+    "OTEL_EXPORTER_OTLP_ENDPOINT", "http://localhost:4317"
+)
+OTEL_SERVICE_NAME = os.environ.get("OTEL_SERVICE_NAME", "open-webui")
+OTEL_RESOURCE_ATTRIBUTES = os.environ.get(
+    "OTEL_RESOURCE_ATTRIBUTES", ""
+)  # e.g. key1=val1,key2=val2
+OTEL_TRACES_SAMPLER = os.environ.get(
+    "OTEL_TRACES_SAMPLER", "parentbased_always_on"
+).lower()
+
+####################################
+# TOOLS/FUNCTIONS PIP OPTIONS
+####################################
+
+PIP_OPTIONS = os.getenv("PIP_OPTIONS", "").split()
+PIP_PACKAGE_INDEX_OPTIONS = os.getenv("PIP_PACKAGE_INDEX_OPTIONS", "").split()
+
+
+####################################
+# PROGRESSIVE WEB APP OPTIONS
+####################################
+
+EXTERNAL_PWA_MANIFEST_URL = os.environ.get("EXTERNAL_PWA_MANIFEST_URL")

backend/open_webui/functions.py

@@ -223,6 +223,9 @@ async def generate_function_chat_completion(
     extra_params = {
         "__event_emitter__": __event_emitter__,
         "__event_call__": __event_call__,
+        "__chat_id__": metadata.get("chat_id", None),
+        "__session_id__": metadata.get("session_id", None),
+        "__message_id__": metadata.get("message_id", None),
         "__task__": __task__,
         "__task_body__": __task_body__,
         "__files__": files,

backend/open_webui/main.py

@@ -84,11 +84,12 @@ from open_webui.routers.retrieval import (
     get_rf,
 )
 
-from open_webui.internal.db import Session
+from open_webui.internal.db import Session, engine
 
 from open_webui.models.functions import Functions
 from open_webui.models.models import Models
 from open_webui.models.users import UserModel, Users
+from open_webui.models.chats import Chats
 
 from open_webui.config import (
     LICENSE_KEY,
@@ -104,6 +105,8 @@ from open_webui.config import (
     OPENAI_API_CONFIGS,
     # Direct Connections
     ENABLE_DIRECT_CONNECTIONS,
+    # Tool Server Configs
+    TOOL_SERVER_CONNECTIONS,
     # Code Execution
     ENABLE_CODE_EXECUTION,
     CODE_EXECUTION_ENGINE,
@@ -145,6 +148,9 @@ from open_webui.config import (
     AUDIO_STT_MODEL,
     AUDIO_STT_OPENAI_API_BASE_URL,
     AUDIO_STT_OPENAI_API_KEY,
+    AUDIO_STT_AZURE_API_KEY,
+    AUDIO_STT_AZURE_REGION,
+    AUDIO_STT_AZURE_LOCALES,
     AUDIO_TTS_API_KEY,
     AUDIO_TTS_ENGINE,
     AUDIO_TTS_MODEL,
@@ -154,10 +160,11 @@ from open_webui.config import (
     AUDIO_TTS_VOICE,
     AUDIO_TTS_AZURE_SPEECH_REGION,
     AUDIO_TTS_AZURE_SPEECH_OUTPUT_FORMAT,
-    PLAYWRIGHT_WS_URI,
+    PLAYWRIGHT_WS_URL,
+    PLAYWRIGHT_TIMEOUT,
     FIRECRAWL_API_BASE_URL,
     FIRECRAWL_API_KEY,
-    RAG_WEB_LOADER_ENGINE,
+    WEB_LOADER_ENGINE,
     WHISPER_MODEL,
     DEEPGRAM_API_KEY,
     WHISPER_MODEL_AUTO_UPDATE,
@@ -186,21 +193,25 @@ from open_webui.config import (
     CHUNK_SIZE,
     CONTENT_EXTRACTION_ENGINE,
     TIKA_SERVER_URL,
+    DOCLING_SERVER_URL,
     DOCUMENT_INTELLIGENCE_ENDPOINT,
     DOCUMENT_INTELLIGENCE_KEY,
+    MISTRAL_OCR_API_KEY,
     RAG_TOP_K,
+    RAG_TOP_K_RERANKER,
     RAG_TEXT_SPLITTER,
     TIKTOKEN_ENCODING_NAME,
     PDF_EXTRACT_IMAGES,
     YOUTUBE_LOADER_LANGUAGE,
     YOUTUBE_LOADER_PROXY_URL,
     # Retrieval (Web Search)
-    RAG_WEB_SEARCH_ENGINE,
+    ENABLE_WEB_SEARCH,
+    WEB_SEARCH_ENGINE,
     BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL,
-    RAG_WEB_SEARCH_RESULT_COUNT,
-    RAG_WEB_SEARCH_CONCURRENT_REQUESTS,
-    RAG_WEB_SEARCH_TRUST_ENV,
-    RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
+    WEB_SEARCH_RESULT_COUNT,
+    WEB_SEARCH_CONCURRENT_REQUESTS,
+    WEB_SEARCH_TRUST_ENV,
+    WEB_SEARCH_DOMAIN_FILTER_LIST,
     JINA_API_KEY,
     SEARCHAPI_API_KEY,
     SEARCHAPI_ENGINE,
@@ -212,11 +223,14 @@ from open_webui.config import (
     SERPSTACK_API_KEY,
     SERPSTACK_HTTPS,
     TAVILY_API_KEY,
+    TAVILY_EXTRACT_DEPTH,
     BING_SEARCH_V7_ENDPOINT,
     BING_SEARCH_V7_SUBSCRIPTION_KEY,
     BRAVE_SEARCH_API_KEY,
     EXA_API_KEY,
     PERPLEXITY_API_KEY,
+    SOUGOU_API_SID,
+    SOUGOU_API_SK,
     KAGI_SEARCH_API_KEY,
     MOJEEK_SEARCH_API_KEY,
     BOCHA_SEARCH_API_KEY,
@@ -227,8 +241,7 @@ from open_webui.config import (
     ONEDRIVE_CLIENT_ID,
     ENABLE_RAG_HYBRID_SEARCH,
     ENABLE_RAG_LOCAL_WEB_FETCH,
-    ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
-    ENABLE_RAG_WEB_SEARCH,
+    ENABLE_WEB_LOADER_SSL_VERIFICATION,
     ENABLE_GOOGLE_DRIVE_INTEGRATION,
     ENABLE_ONEDRIVE_INTEGRATION,
     UPLOAD_DIR,
@@ -248,6 +261,7 @@ from open_webui.config import (
     ENABLE_CHANNELS,
     ENABLE_COMMUNITY_SHARING,
     ENABLE_MESSAGE_RATING,
+    ENABLE_USER_WEBHOOKS,
     ENABLE_EVALUATION_ARENA_MODELS,
     USER_PERMISSIONS,
     DEFAULT_USER_ROLE,
@@ -312,6 +326,9 @@ from open_webui.env import (
     AUDIT_EXCLUDED_PATHS,
     AUDIT_LOG_LEVEL,
     CHANGELOG,
+    REDIS_URL,
+    REDIS_SENTINEL_HOSTS,
+    REDIS_SENTINEL_PORT,
     GLOBAL_LOG_LEVEL,
     MAX_BODY_LOG_SIZE,
     SAFE_MODE,
@@ -327,6 +344,8 @@ from open_webui.env import (
     BYPASS_MODEL_ACCESS_CONTROL,
     RESET_CONFIG_ON_START,
     OFFLINE_MODE,
+    ENABLE_OTEL,
+    EXTERNAL_PWA_MANIFEST_URL,
 )
 
 
@@ -345,6 +364,7 @@ from open_webui.utils.access_control import has_access
 
 from open_webui.utils.auth import (
     get_license_data,
+    get_http_authorization_cred,
     decode_token,
     get_admin_user,
     get_verified_user,
@@ -352,7 +372,13 @@ from open_webui.utils.auth import (
 from open_webui.utils.oauth import OAuthManager
 from open_webui.utils.security_headers import SecurityHeadersMiddleware
 
-from open_webui.tasks import stop_task, list_tasks  # Import from tasks.py
+from open_webui.tasks import (
+    list_task_ids_by_chat_id,
+    stop_task,
+    list_tasks,
+)  # Import from tasks.py
+
+from open_webui.utils.redis import get_sentinels_from_env
 
 
 if SAFE_MODE:
@@ -410,6 +436,7 @@ async def lifespan(app: FastAPI):
 
 
 app = FastAPI(
+    title="Open WebUI",
     docs_url="/docs" if ENV == "dev" else None,
     openapi_url="/openapi.json" if ENV == "dev" else None,
     redoc_url=None,
@@ -418,11 +445,27 @@ app = FastAPI(
 
 oauth_manager = OAuthManager(app)
 
-app.state.config = AppConfig()
+app.state.config = AppConfig(
+    redis_url=REDIS_URL,
+    redis_sentinels=get_sentinels_from_env(REDIS_SENTINEL_HOSTS, REDIS_SENTINEL_PORT),
+)
 
 app.state.WEBUI_NAME = WEBUI_NAME
 app.state.LICENSE_METADATA = None
 
+
+########################################
+#
+# OPENTELEMETRY
+#
+########################################
+
+if ENABLE_OTEL:
+    from open_webui.utils.telemetry.setup import setup as setup_opentelemetry
+
+    setup_opentelemetry(app=app, db_engine=engine)
+
+
 ########################################
 #
 # OLLAMA
@@ -449,6 +492,15 @@ app.state.config.OPENAI_API_CONFIGS = OPENAI_API_CONFIGS
 
 app.state.OPENAI_MODELS = {}
 
+########################################
+#
+# TOOL SERVERS
+#
+########################################
+
+app.state.config.TOOL_SERVER_CONNECTIONS = TOOL_SERVER_CONNECTIONS
+app.state.TOOL_SERVERS = []
+
 ########################################
 #
 # DIRECT CONNECTIONS
@@ -492,6 +544,7 @@ app.state.config.MODEL_ORDER_LIST = MODEL_ORDER_LIST
 app.state.config.ENABLE_CHANNELS = ENABLE_CHANNELS
 app.state.config.ENABLE_COMMUNITY_SHARING = ENABLE_COMMUNITY_SHARING
 app.state.config.ENABLE_MESSAGE_RATING = ENABLE_MESSAGE_RATING
+app.state.config.ENABLE_USER_WEBHOOKS = ENABLE_USER_WEBHOOKS
 
 app.state.config.ENABLE_EVALUATION_ARENA_MODELS = ENABLE_EVALUATION_ARENA_MODELS
 app.state.config.EVALUATION_ARENA_MODELS = EVALUATION_ARENA_MODELS
@@ -522,6 +575,7 @@ app.state.config.LDAP_CIPHERS = LDAP_CIPHERS
 
 app.state.AUTH_TRUSTED_EMAIL_HEADER = WEBUI_AUTH_TRUSTED_EMAIL_HEADER
 app.state.AUTH_TRUSTED_NAME_HEADER = WEBUI_AUTH_TRUSTED_NAME_HEADER
+app.state.EXTERNAL_PWA_MANIFEST_URL = EXTERNAL_PWA_MANIFEST_URL
 
 app.state.USER_COUNT = None
 app.state.TOOLS = {}
@@ -535,6 +589,7 @@ app.state.FUNCTIONS = {}
 
 
 app.state.config.TOP_K = RAG_TOP_K
+app.state.config.TOP_K_RERANKER = RAG_TOP_K_RERANKER
 app.state.config.RELEVANCE_THRESHOLD = RAG_RELEVANCE_THRESHOLD
 app.state.config.FILE_MAX_SIZE = RAG_FILE_MAX_SIZE
 app.state.config.FILE_MAX_COUNT = RAG_FILE_MAX_COUNT
@@ -543,14 +598,14 @@ app.state.config.FILE_MAX_COUNT = RAG_FILE_MAX_COUNT
 app.state.config.RAG_FULL_CONTEXT = RAG_FULL_CONTEXT
 app.state.config.BYPASS_EMBEDDING_AND_RETRIEVAL = BYPASS_EMBEDDING_AND_RETRIEVAL
 app.state.config.ENABLE_RAG_HYBRID_SEARCH = ENABLE_RAG_HYBRID_SEARCH
-app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION = (
-    ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION
-)
+app.state.config.ENABLE_WEB_LOADER_SSL_VERIFICATION = ENABLE_WEB_LOADER_SSL_VERIFICATION
 
 app.state.config.CONTENT_EXTRACTION_ENGINE = CONTENT_EXTRACTION_ENGINE
 app.state.config.TIKA_SERVER_URL = TIKA_SERVER_URL
+app.state.config.DOCLING_SERVER_URL = DOCLING_SERVER_URL
 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
 
 app.state.config.TEXT_SPLITTER = RAG_TEXT_SPLITTER
 app.state.config.TIKTOKEN_ENCODING_NAME = TIKTOKEN_ENCODING_NAME
@@ -576,12 +631,16 @@ app.state.config.YOUTUBE_LOADER_LANGUAGE = YOUTUBE_LOADER_LANGUAGE
 app.state.config.YOUTUBE_LOADER_PROXY_URL = YOUTUBE_LOADER_PROXY_URL
 
 
-app.state.config.ENABLE_RAG_WEB_SEARCH = ENABLE_RAG_WEB_SEARCH
-app.state.config.RAG_WEB_SEARCH_ENGINE = RAG_WEB_SEARCH_ENGINE
+app.state.config.ENABLE_WEB_SEARCH = ENABLE_WEB_SEARCH
+app.state.config.WEB_SEARCH_ENGINE = WEB_SEARCH_ENGINE
+app.state.config.WEB_SEARCH_DOMAIN_FILTER_LIST = WEB_SEARCH_DOMAIN_FILTER_LIST
+app.state.config.WEB_SEARCH_RESULT_COUNT = WEB_SEARCH_RESULT_COUNT
+app.state.config.WEB_SEARCH_CONCURRENT_REQUESTS = WEB_SEARCH_CONCURRENT_REQUESTS
+app.state.config.WEB_LOADER_ENGINE = WEB_LOADER_ENGINE
+app.state.config.WEB_SEARCH_TRUST_ENV = WEB_SEARCH_TRUST_ENV
 app.state.config.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL = (
     BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL
 )
-app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST = RAG_WEB_SEARCH_DOMAIN_FILTER_LIST
 
 app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION = ENABLE_GOOGLE_DRIVE_INTEGRATION
 app.state.config.ENABLE_ONEDRIVE_INTEGRATION = ENABLE_ONEDRIVE_INTEGRATION
@@ -606,14 +665,15 @@ 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.SOUGOU_API_SID = SOUGOU_API_SID
+app.state.config.SOUGOU_API_SK = SOUGOU_API_SK
 
-app.state.config.RAG_WEB_SEARCH_RESULT_COUNT = RAG_WEB_SEARCH_RESULT_COUNT
-app.state.config.RAG_WEB_SEARCH_CONCURRENT_REQUESTS = RAG_WEB_SEARCH_CONCURRENT_REQUESTS
-app.state.config.RAG_WEB_LOADER_ENGINE = RAG_WEB_LOADER_ENGINE
-app.state.config.RAG_WEB_SEARCH_TRUST_ENV = RAG_WEB_SEARCH_TRUST_ENV
-app.state.config.PLAYWRIGHT_WS_URI = PLAYWRIGHT_WS_URI
+
+app.state.config.PLAYWRIGHT_WS_URL = PLAYWRIGHT_WS_URL
+app.state.config.PLAYWRIGHT_TIMEOUT = PLAYWRIGHT_TIMEOUT
 app.state.config.FIRECRAWL_API_BASE_URL = FIRECRAWL_API_BASE_URL
 app.state.config.FIRECRAWL_API_KEY = FIRECRAWL_API_KEY
+app.state.config.TAVILY_EXTRACT_DEPTH = TAVILY_EXTRACT_DEPTH
 
 app.state.EMBEDDING_FUNCTION = None
 app.state.ef = None
@@ -731,6 +791,10 @@ app.state.config.STT_MODEL = AUDIO_STT_MODEL
 app.state.config.WHISPER_MODEL = WHISPER_MODEL
 app.state.config.DEEPGRAM_API_KEY = DEEPGRAM_API_KEY
 
+app.state.config.AUDIO_STT_AZURE_API_KEY = AUDIO_STT_AZURE_API_KEY
+app.state.config.AUDIO_STT_AZURE_REGION = AUDIO_STT_AZURE_REGION
+app.state.config.AUDIO_STT_AZURE_LOCALES = AUDIO_STT_AZURE_LOCALES
+
 app.state.config.TTS_OPENAI_API_BASE_URL = AUDIO_TTS_OPENAI_API_BASE_URL
 app.state.config.TTS_OPENAI_API_KEY = AUDIO_TTS_OPENAI_API_KEY
 app.state.config.TTS_ENGINE = AUDIO_TTS_ENGINE
@@ -829,6 +893,10 @@ async def commit_session_after_request(request: Request, call_next):
 @app.middleware("http")
 async def check_url(request: Request, call_next):
     start_time = int(time.time())
+    request.state.token = get_http_authorization_cred(
+        request.headers.get("Authorization")
+    )
+
     request.state.enable_api_key = app.state.config.ENABLE_API_KEY
     response = await call_next(request)
     process_time = int(time.time()) - start_time
@@ -947,14 +1015,24 @@ async def get_models(request: Request, user=Depends(get_verified_user)):
 
         return filtered_models
 
-    models = await get_all_models(request, user=user)
+    all_models = await get_all_models(request, user=user)
 
-    # Filter out filter pipelines
-    models = [
-        model
-        for model in models
-        if "pipeline" not in model or model["pipeline"].get("type", None) != "filter"
-    ]
+    models = []
+    for model in all_models:
+        # Filter out filter pipelines
+        if "pipeline" in model and model["pipeline"].get("type", None) == "filter":
+            continue
+
+        model_tags = [
+            tag.get("name")
+            for tag in model.get("info", {}).get("meta", {}).get("tags", [])
+        ]
+        tags = [tag.get("name") for tag in model.get("tags", [])]
+
+        tags = list(set(model_tags + tags))
+        model["tags"] = [{"name": tag} for tag in tags]
+
+        models.append(model)
 
     model_order_list = request.app.state.config.MODEL_ORDER_LIST
     if model_order_list:
@@ -992,6 +1070,7 @@ async def chat_completion(
     model_item = form_data.pop("model_item", {})
     tasks = form_data.pop("background_tasks", None)
 
+    metadata = {}
     try:
         if not model_item.get("direct", False):
             model_id = form_data.get("model", None)
@@ -1020,6 +1099,7 @@ async def chat_completion(
             "message_id": form_data.pop("id", None),
             "session_id": form_data.pop("session_id", None),
             "tool_ids": form_data.get("tool_ids", None),
+            "tool_servers": form_data.pop("tool_servers", None),
             "files": form_data.get("files", None),
             "features": form_data.get("features", None),
             "variables": form_data.get("variables", None),
@@ -1046,6 +1126,16 @@ async def chat_completion(
 
     except Exception as e:
         log.debug(f"Error processing chat payload: {e}")
+        if metadata.get("chat_id") and metadata.get("message_id"):
+            # Update the chat message with the error
+            Chats.upsert_message_to_chat_by_id_and_message_id(
+                metadata["chat_id"],
+                metadata["message_id"],
+                {
+                    "error": {"content": str(e)},
+                },
+            )
+
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,
             detail=str(e),
@@ -1110,7 +1200,7 @@ async def chat_action(
 @app.post("/api/tasks/stop/{task_id}")
 async def stop_task_endpoint(task_id: str, user=Depends(get_verified_user)):
     try:
-        result = await stop_task(task_id)  # Use the function from tasks.py
+        result = await stop_task(task_id)
         return result
     except ValueError as e:
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(e))
@@ -1118,7 +1208,19 @@ async def stop_task_endpoint(task_id: str, user=Depends(get_verified_user)):
 
 @app.get("/api/tasks")
 async def list_tasks_endpoint(user=Depends(get_verified_user)):
-    return {"tasks": list_tasks()}  # Use the function from tasks.py
+    return {"tasks": list_tasks()}
+
+
+@app.get("/api/tasks/chat/{chat_id}")
+async def list_tasks_by_chat_id_endpoint(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)
+
+    print(f"Task IDs for chat {chat_id}: {task_ids}")
+    return {"task_ids": task_ids}
 
 
 ##################################
@@ -1174,13 +1276,14 @@ async def get_app_config(request: Request):
                 {
                     "enable_direct_connections": app.state.config.ENABLE_DIRECT_CONNECTIONS,
                     "enable_channels": app.state.config.ENABLE_CHANNELS,
-                    "enable_web_search": app.state.config.ENABLE_RAG_WEB_SEARCH,
+                    "enable_web_search": app.state.config.ENABLE_WEB_SEARCH,
                     "enable_code_execution": app.state.config.ENABLE_CODE_EXECUTION,
                     "enable_code_interpreter": app.state.config.ENABLE_CODE_INTERPRETER,
                     "enable_image_generation": app.state.config.ENABLE_IMAGE_GENERATION,
                     "enable_autocomplete_generation": app.state.config.ENABLE_AUTOCOMPLETE_GENERATION,
                     "enable_community_sharing": app.state.config.ENABLE_COMMUNITY_SHARING,
                     "enable_message_rating": app.state.config.ENABLE_MESSAGE_RATING,
+                    "enable_user_webhooks": app.state.config.ENABLE_USER_WEBHOOKS,
                     "enable_admin_export": ENABLE_ADMIN_EXPORT,
                     "enable_admin_chat_access": ENABLE_ADMIN_CHAT_ACCESS,
                     "enable_google_drive_integration": app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
@@ -1319,29 +1422,32 @@ async def oauth_callback(provider: str, request: Request, response: Response):
 
 @app.get("/manifest.json")
 async def get_manifest_json():
-    return {
-        "name": app.state.WEBUI_NAME,
-        "short_name": app.state.WEBUI_NAME,
-        "description": "Open WebUI is an open, extensible, user-friendly interface for AI that adapts to your workflow.",
-        "start_url": "/",
-        "display": "standalone",
-        "background_color": "#343541",
-        "orientation": "natural",
-        "icons": [
-            {
-                "src": "/static/logo.png",
-                "type": "image/png",
-                "sizes": "500x500",
-                "purpose": "any",
-            },
-            {
-                "src": "/static/logo.png",
-                "type": "image/png",
-                "sizes": "500x500",
-                "purpose": "maskable",
-            },
-        ],
-    }
+    if app.state.EXTERNAL_PWA_MANIFEST_URL:
+        return requests.get(app.state.EXTERNAL_PWA_MANIFEST_URL).json()
+    else:
+        return {
+            "name": app.state.WEBUI_NAME,
+            "short_name": app.state.WEBUI_NAME,
+            "description": "Open WebUI is an open, extensible, user-friendly interface for AI that adapts to your workflow.",
+            "start_url": "/",
+            "display": "standalone",
+            "background_color": "#343541",
+            "orientation": "natural",
+            "icons": [
+                {
+                    "src": "/static/logo.png",
+                    "type": "image/png",
+                    "sizes": "500x500",
+                    "purpose": "any",
+                },
+                {
+                    "src": "/static/logo.png",
+                    "type": "image/png",
+                    "sizes": "500x500",
+                    "purpose": "maskable",
+                },
+            ],
+        }
 
 
 @app.get("/opensearch.xml")

backend/open_webui/models/folders.py

@@ -9,6 +9,8 @@ from open_webui.models.chats import Chats
 from open_webui.env import SRC_LOG_LEVELS
 from pydantic import BaseModel, ConfigDict
 from sqlalchemy import BigInteger, Column, Text, JSON, Boolean
+from open_webui.utils.access_control import get_permissions
+
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MODELS"])
@@ -234,15 +236,18 @@ class FolderTable:
             log.error(f"update_folder: {e}")
             return
 
-    def delete_folder_by_id_and_user_id(self, id: str, user_id: str) -> bool:
+    def delete_folder_by_id_and_user_id(
+        self, id: str, user_id: str, delete_chats=True
+    ) -> bool:
         try:
             with get_db() as db:
                 folder = db.query(Folder).filter_by(id=id, user_id=user_id).first()
                 if not folder:
                     return False
 
-                # Delete all chats in the folder
-                Chats.delete_chats_by_user_id_and_folder_id(user_id, folder.id)
+                if delete_chats:
+                    # Delete all chats in the folder
+                    Chats.delete_chats_by_user_id_and_folder_id(user_id, folder.id)
 
                 # Delete all children folders
                 def delete_children(folder):
@@ -250,9 +255,11 @@ class FolderTable:
                         folder.id, user_id
                     )
                     for folder_child in folder_children:
-                        Chats.delete_chats_by_user_id_and_folder_id(
-                            user_id, folder_child.id
-                        )
+                        if delete_chats:
+                            Chats.delete_chats_by_user_id_and_folder_id(
+                                user_id, folder_child.id
+                            )
+
                         delete_children(folder_child)
 
                         folder = db.query(Folder).filter_by(id=folder_child.id).first()

backend/open_webui/models/memories.py

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

backend/open_webui/retrieval/loaders/main.py

@@ -20,6 +20,9 @@ from langchain_community.document_loaders import (
     YoutubeLoader,
 )
 from langchain_core.documents import Document
+
+from open_webui.retrieval.loaders.mistral import MistralLoader
+
 from open_webui.env import SRC_LOG_LEVELS, GLOBAL_LOG_LEVEL
 
 logging.basicConfig(stream=sys.stdout, level=GLOBAL_LOG_LEVEL)
@@ -105,7 +108,7 @@ class TikaLoader:
 
         if r.ok:
             raw_metadata = r.json()
-            text = raw_metadata.get("X-TIKA:content", "<No text content found>")
+            text = raw_metadata.get("X-TIKA:content", "<No text content found>").strip()
 
             if "Content-Type" in raw_metadata:
                 headers["Content-Type"] = raw_metadata["Content-Type"]
@@ -117,6 +120,52 @@ class TikaLoader:
             raise Exception(f"Error calling Tika: {r.reason}")
 
 
+class DoclingLoader:
+    def __init__(self, url, file_path=None, mime_type=None):
+        self.url = url.rstrip("/")
+        self.file_path = file_path
+        self.mime_type = mime_type
+
+    def load(self) -> list[Document]:
+        with open(self.file_path, "rb") as f:
+            files = {
+                "files": (
+                    self.file_path,
+                    f,
+                    self.mime_type or "application/octet-stream",
+                )
+            }
+
+            params = {
+                "image_export_mode": "placeholder",
+                "table_mode": "accurate",
+            }
+
+            endpoint = f"{self.url}/v1alpha/convert/file"
+            r = requests.post(endpoint, files=files, data=params)
+
+        if r.ok:
+            result = r.json()
+            document_data = result.get("document", {})
+            text = document_data.get("md_content", "<No text content found>")
+
+            metadata = {"Content-Type": self.mime_type} if self.mime_type else {}
+
+            log.debug("Docling extracted text: %s", text)
+
+            return [Document(page_content=text, metadata=metadata)]
+        else:
+            error_msg = f"Error calling Docling API: {r.reason}"
+            if r.text:
+                try:
+                    error_data = r.json()
+                    if "detail" in error_data:
+                        error_msg += f" - {error_data['detail']}"
+                except Exception:
+                    error_msg += f" - {r.text}"
+            raise Exception(f"Error calling Docling: {error_msg}")
+
+
 class Loader:
     def __init__(self, engine: str = "", **kwargs):
         self.engine = engine
@@ -135,13 +184,16 @@ class Loader:
             for doc in docs
         ]
 
+    def _is_text_file(self, file_ext: str, file_content_type: str) -> bool:
+        return file_ext in known_source_ext or (
+            file_content_type and file_content_type.find("text/") >= 0
+        )
+
     def _get_loader(self, filename: str, file_content_type: str, file_path: str):
         file_ext = filename.split(".")[-1].lower()
 
         if self.engine == "tika" and self.kwargs.get("TIKA_SERVER_URL"):
-            if file_ext in known_source_ext or (
-                file_content_type and file_content_type.find("text/") >= 0
-            ):
+            if self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:
                 loader = TikaLoader(
@@ -149,6 +201,15 @@ class Loader:
                     file_path=file_path,
                     mime_type=file_content_type,
                 )
+        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:
+                loader = DoclingLoader(
+                    url=self.kwargs.get("DOCLING_SERVER_URL"),
+                    file_path=file_path,
+                    mime_type=file_content_type,
+                )
         elif (
             self.engine == "document_intelligence"
             and self.kwargs.get("DOCUMENT_INTELLIGENCE_ENDPOINT") != ""
@@ -170,13 +231,22 @@ class Loader:
                 api_endpoint=self.kwargs.get("DOCUMENT_INTELLIGENCE_ENDPOINT"),
                 api_key=self.kwargs.get("DOCUMENT_INTELLIGENCE_KEY"),
             )
+        elif (
+            self.engine == "mistral_ocr"
+            and self.kwargs.get("MISTRAL_OCR_API_KEY") != ""
+            and file_ext
+            in ["pdf"]  # Mistral OCR currently only supports PDF and images
+        ):
+            loader = MistralLoader(
+                api_key=self.kwargs.get("MISTRAL_OCR_API_KEY"), file_path=file_path
+            )
         else:
             if file_ext == "pdf":
                 loader = PyPDFLoader(
                     file_path, extract_images=self.kwargs.get("PDF_EXTRACT_IMAGES")
                 )
             elif file_ext == "csv":
-                loader = CSVLoader(file_path)
+                loader = CSVLoader(file_path, autodetect_encoding=True)
             elif file_ext == "rst":
                 loader = UnstructuredRSTLoader(file_path, mode="elements")
             elif file_ext == "xml":
@@ -205,9 +275,7 @@ class Loader:
                 loader = UnstructuredPowerPointLoader(file_path)
             elif file_ext == "msg":
                 loader = OutlookMessageLoader(file_path)
-            elif file_ext in known_source_ext or (
-                file_content_type and file_content_type.find("text/") >= 0
-            ):
+            elif self._is_text_file(file_ext, file_content_type):
                 loader = TextLoader(file_path, autodetect_encoding=True)
             else:
                 loader = TextLoader(file_path, autodetect_encoding=True)

backend/open_webui/retrieval/loaders/mistral.py

@@ -0,0 +1,225 @@
+import requests
+import logging
+import os
+import sys
+from typing import List, Dict, Any
+
+from langchain_core.documents import Document
+from open_webui.env import SRC_LOG_LEVELS, GLOBAL_LOG_LEVEL
+
+logging.basicConfig(stream=sys.stdout, level=GLOBAL_LOG_LEVEL)
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class MistralLoader:
+    """
+    Loads documents by processing them through the Mistral OCR API.
+    """
+
+    BASE_API_URL = "https://api.mistral.ai/v1"
+
+    def __init__(self, api_key: str, file_path: str):
+        """
+        Initializes the loader.
+
+        Args:
+            api_key: Your Mistral API key.
+            file_path: The local path to the PDF file to process.
+        """
+        if not api_key:
+            raise ValueError("API key cannot be empty.")
+        if not os.path.exists(file_path):
+            raise FileNotFoundError(f"File not found at {file_path}")
+
+        self.api_key = api_key
+        self.file_path = file_path
+        self.headers = {"Authorization": f"Bearer {self.api_key}"}
+
+    def _handle_response(self, response: requests.Response) -> Dict[str, Any]:
+        """Checks response status and returns JSON content."""
+        try:
+            response.raise_for_status()  # Raises HTTPError for bad responses (4xx or 5xx)
+            # Handle potential empty responses for certain successful requests (e.g., DELETE)
+            if response.status_code == 204 or not response.content:
+                return {}  # Return empty dict if no content
+            return response.json()
+        except requests.exceptions.HTTPError as http_err:
+            log.error(f"HTTP error occurred: {http_err} - Response: {response.text}")
+            raise
+        except requests.exceptions.RequestException as req_err:
+            log.error(f"Request exception occurred: {req_err}")
+            raise
+        except ValueError as json_err:  # Includes JSONDecodeError
+            log.error(f"JSON decode error: {json_err} - Response: {response.text}")
+            raise  # Re-raise after logging
+
+    def _upload_file(self) -> str:
+        """Uploads the file to Mistral for OCR processing."""
+        log.info("Uploading file to Mistral API")
+        url = f"{self.BASE_API_URL}/files"
+        file_name = os.path.basename(self.file_path)
+
+        try:
+            with open(self.file_path, "rb") as f:
+                files = {"file": (file_name, f, "application/pdf")}
+                data = {"purpose": "ocr"}
+
+                upload_headers = self.headers.copy()  # Avoid modifying self.headers
+
+                response = requests.post(
+                    url, headers=upload_headers, files=files, data=data
+                )
+
+            response_data = self._handle_response(response)
+            file_id = response_data.get("id")
+            if not file_id:
+                raise ValueError("File ID not found in upload response.")
+            log.info(f"File uploaded successfully. File ID: {file_id}")
+            return file_id
+        except Exception as e:
+            log.error(f"Failed to upload file: {e}")
+            raise
+
+    def _get_signed_url(self, file_id: str) -> str:
+        """Retrieves a temporary signed URL for the uploaded file."""
+        log.info(f"Getting signed URL for file ID: {file_id}")
+        url = f"{self.BASE_API_URL}/files/{file_id}/url"
+        params = {"expiry": 1}
+        signed_url_headers = {**self.headers, "Accept": "application/json"}
+
+        try:
+            response = requests.get(url, headers=signed_url_headers, params=params)
+            response_data = self._handle_response(response)
+            signed_url = response_data.get("url")
+            if not signed_url:
+                raise ValueError("Signed URL not found in response.")
+            log.info("Signed URL received.")
+            return signed_url
+        except Exception as e:
+            log.error(f"Failed to get signed URL: {e}")
+            raise
+
+    def _process_ocr(self, signed_url: str) -> Dict[str, Any]:
+        """Sends the signed URL to the OCR endpoint for processing."""
+        log.info("Processing OCR via Mistral API")
+        url = f"{self.BASE_API_URL}/ocr"
+        ocr_headers = {
+            **self.headers,
+            "Content-Type": "application/json",
+            "Accept": "application/json",
+        }
+        payload = {
+            "model": "mistral-ocr-latest",
+            "document": {
+                "type": "document_url",
+                "document_url": signed_url,
+            },
+            "include_image_base64": False,
+        }
+
+        try:
+            response = requests.post(url, headers=ocr_headers, json=payload)
+            ocr_response = self._handle_response(response)
+            log.info("OCR processing done.")
+            log.debug("OCR response: %s", ocr_response)
+            return ocr_response
+        except Exception as e:
+            log.error(f"Failed during OCR processing: {e}")
+            raise
+
+    def _delete_file(self, file_id: str) -> None:
+        """Deletes the file from Mistral storage."""
+        log.info(f"Deleting uploaded file ID: {file_id}")
+        url = f"{self.BASE_API_URL}/files/{file_id}"
+        # No specific Accept header needed, default or Authorization is usually sufficient
+
+        try:
+            response = requests.delete(url, headers=self.headers)
+            delete_response = self._handle_response(
+                response
+            )  # Check status, ignore response body unless needed
+            log.info(
+                f"File deleted successfully: {delete_response}"
+            )  # Log the response if available
+        except Exception as e:
+            # Log error but don't necessarily halt execution if deletion fails
+            log.error(f"Failed to delete file ID {file_id}: {e}")
+            # Depending on requirements, you might choose to raise the error here
+
+    def load(self) -> List[Document]:
+        """
+        Executes the full OCR workflow: upload, get URL, process OCR, delete file.
+
+        Returns:
+            A list of Document objects, one for each page processed.
+        """
+        file_id = None
+        try:
+            # 1. Upload file
+            file_id = self._upload_file()
+
+            # 2. Get Signed URL
+            signed_url = self._get_signed_url(file_id)
+
+            # 3. Process OCR
+            ocr_response = self._process_ocr(signed_url)
+
+            # 4. Process results
+            pages_data = ocr_response.get("pages")
+            if not pages_data:
+                log.warning("No pages found in OCR response.")
+                return [Document(page_content="No text content found", metadata={})]
+
+            documents = []
+            total_pages = len(pages_data)
+            for page_data in pages_data:
+                page_content = page_data.get("markdown")
+                page_index = page_data.get("index")  # API uses 0-based index
+
+                if page_content is not None and page_index is not None:
+                    documents.append(
+                        Document(
+                            page_content=page_content,
+                            metadata={
+                                "page": page_index,  # 0-based index from API
+                                "page_label": page_index
+                                + 1,  # 1-based label for convenience
+                                "total_pages": total_pages,
+                                # Add other relevant metadata from page_data if available/needed
+                                # e.g., page_data.get('width'), page_data.get('height')
+                            },
+                        )
+                    )
+                else:
+                    log.warning(
+                        f"Skipping page due to missing 'markdown' or 'index'. Data: {page_data}"
+                    )
+
+            if not documents:
+                # Case where pages existed but none had valid markdown/index
+                log.warning(
+                    "OCR response contained pages, but none had valid content/index."
+                )
+                return [
+                    Document(
+                        page_content="No text content found in valid pages", metadata={}
+                    )
+                ]
+
+            return documents
+
+        except Exception as e:
+            log.error(f"An error occurred during the loading process: {e}")
+            # Return an empty list or a specific error document on failure
+            return [Document(page_content=f"Error during processing: {e}", metadata={})]
+        finally:
+            # 5. Delete file (attempt even if prior steps failed after upload)
+            if file_id:
+                try:
+                    self._delete_file(file_id)
+                except Exception as del_e:
+                    # Log deletion error, but don't overwrite original error if one occurred
+                    log.error(
+                        f"Cleanup error: Could not delete file ID {file_id}. Reason: {del_e}"
+                    )

backend/open_webui/retrieval/loaders/tavily.py

@@ -0,0 +1,93 @@
+import requests
+import logging
+from typing import Iterator, List, Literal, Union
+
+from langchain_core.document_loaders import BaseLoader
+from langchain_core.documents import Document
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+class TavilyLoader(BaseLoader):
+    """Extract web page content from URLs using Tavily Extract API.
+
+    This is a LangChain document loader that uses Tavily's Extract API to
+    retrieve content from web pages and return it as Document objects.
+
+    Args:
+        urls: URL or list of URLs to extract content from.
+        api_key: The Tavily API key.
+        extract_depth: Depth of extraction, either "basic" or "advanced".
+        continue_on_failure: Whether to continue if extraction of a URL fails.
+    """
+
+    def __init__(
+        self,
+        urls: Union[str, List[str]],
+        api_key: str,
+        extract_depth: Literal["basic", "advanced"] = "basic",
+        continue_on_failure: bool = True,
+    ) -> None:
+        """Initialize Tavily Extract client.
+
+        Args:
+            urls: URL or list of URLs to extract content from.
+            api_key: The Tavily API key.
+            include_images: Whether to include images in the extraction.
+            extract_depth: Depth of extraction, either "basic" or "advanced".
+                advanced extraction retrieves more data, including tables and
+                embedded content, with higher success but may increase latency.
+                basic costs 1 credit per 5 successful URL extractions,
+                advanced costs 2 credits per 5 successful URL extractions.
+            continue_on_failure: Whether to continue if extraction of a URL fails.
+        """
+        if not urls:
+            raise ValueError("At least one URL must be provided.")
+
+        self.api_key = api_key
+        self.urls = urls if isinstance(urls, list) else [urls]
+        self.extract_depth = extract_depth
+        self.continue_on_failure = continue_on_failure
+        self.api_url = "https://api.tavily.com/extract"
+
+    def lazy_load(self) -> Iterator[Document]:
+        """Extract and yield documents from the URLs using Tavily Extract API."""
+        batch_size = 20
+        for i in range(0, len(self.urls), batch_size):
+            batch_urls = self.urls[i : i + batch_size]
+            try:
+                headers = {
+                    "Content-Type": "application/json",
+                    "Authorization": f"Bearer {self.api_key}",
+                }
+                # Use string for single URL, array for multiple URLs
+                urls_param = batch_urls[0] if len(batch_urls) == 1 else batch_urls
+                payload = {"urls": urls_param, "extract_depth": self.extract_depth}
+                # Make the API call
+                response = requests.post(self.api_url, headers=headers, json=payload)
+                response.raise_for_status()
+                response_data = response.json()
+                # Process successful results
+                for result in response_data.get("results", []):
+                    url = result.get("url", "")
+                    content = result.get("raw_content", "")
+                    if not content:
+                        log.warning(f"No content extracted from {url}")
+                        continue
+                    # Add URLs as metadata
+                    metadata = {"source": url}
+                    yield Document(
+                        page_content=content,
+                        metadata=metadata,
+                    )
+                for failed in response_data.get("failed_results", []):
+                    url = failed.get("url", "")
+                    error = failed.get("error", "Unknown error")
+                    log.error(f"Failed to extract content from {url}: {error}")
+            except Exception as e:
+                if self.continue_on_failure:
+                    log.error(f"Error extracting content from batch {batch_urls}: {e}")
+                else:
+                    raise e

backend/open_webui/retrieval/loaders/youtube.py

@@ -110,7 +110,7 @@ class YoutubeLoader:
 
         transcript = " ".join(
             map(
-                lambda transcript_piece: transcript_piece["text"].strip(" "),
+                lambda transcript_piece: transcript_piece.text.strip(" "),
                 transcript_pieces,
             )
         )

backend/open_webui/retrieval/utils.py

@@ -1,30 +1,35 @@
 import logging
 import os
-import uuid
 from typing import Optional, Union
 
-import asyncio
 import requests
 import hashlib
+from concurrent.futures import ThreadPoolExecutor
 
 from huggingface_hub import snapshot_download
 from langchain.retrievers import ContextualCompressionRetriever, EnsembleRetriever
 from langchain_community.retrievers import BM25Retriever
 from langchain_core.documents import Document
 
-
 from open_webui.config import VECTOR_DB
 from open_webui.retrieval.vector.connector import VECTOR_DB_CLIENT
-from open_webui.utils.misc import get_last_user_message, calculate_sha256_string
 
 from open_webui.models.users import UserModel
 from open_webui.models.files import Files
 
+from open_webui.retrieval.vector.main import GetResult
+
+
 from open_webui.env import (
     SRC_LOG_LEVELS,
     OFFLINE_MODE,
     ENABLE_FORWARD_USER_INFO_HEADERS,
 )
+from open_webui.config import (
+    RAG_EMBEDDING_QUERY_PREFIX,
+    RAG_EMBEDDING_CONTENT_PREFIX,
+    RAG_EMBEDDING_PREFIX_FIELD_NAME,
+)
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
@@ -49,7 +54,7 @@ class VectorSearchRetriever(BaseRetriever):
     ) -> list[Document]:
         result = VECTOR_DB_CLIENT.search(
             collection_name=self.collection_name,
-            vectors=[self.embedding_function(query)],
+            vectors=[self.embedding_function(query, RAG_EMBEDDING_QUERY_PREFIX)],
             limit=self.top_k,
         )
 
@@ -72,6 +77,7 @@ def query_doc(
     collection_name: str, query_embedding: list[float], k: int, user: UserModel = None
 ):
     try:
+        log.debug(f"query_doc:doc {collection_name}")
         result = VECTOR_DB_CLIENT.search(
             collection_name=collection_name,
             vectors=[query_embedding],
@@ -89,6 +95,7 @@ def query_doc(
 
 def get_doc(collection_name: str, user: UserModel = None):
     try:
+        log.debug(f"get_doc:doc {collection_name}")
         result = VECTOR_DB_CLIENT.get(collection_name=collection_name)
 
         if result:
@@ -102,18 +109,19 @@ def get_doc(collection_name: str, user: UserModel = None):
 
 def query_doc_with_hybrid_search(
     collection_name: str,
+    collection_result: GetResult,
     query: str,
     embedding_function,
     k: int,
     reranking_function,
+    k_reranker: int,
     r: float,
 ) -> dict:
     try:
-        result = VECTOR_DB_CLIENT.get(collection_name=collection_name)
-
+        log.debug(f"query_doc_with_hybrid_search:doc {collection_name}")
         bm25_retriever = BM25Retriever.from_texts(
-            texts=result.documents[0],
-            metadatas=result.metadatas[0],
+            texts=collection_result.documents[0],
+            metadatas=collection_result.metadatas[0],
         )
         bm25_retriever.k = k
 
@@ -128,7 +136,7 @@ def query_doc_with_hybrid_search(
         )
         compressor = RerankCompressor(
             embedding_function=embedding_function,
-            top_n=k,
+            top_n=k_reranker,
             reranking_function=reranking_function,
             r_score=r,
         )
@@ -138,10 +146,23 @@ def query_doc_with_hybrid_search(
         )
 
         result = compression_retriever.invoke(query)
+
+        distances = [d.metadata.get("score") for d in result]
+        documents = [d.page_content for d in result]
+        metadatas = [d.metadata for d in result]
+
+        # retrieve only min(k, k_reranker) items, sort and cut by distance if k < k_reranker
+        if k < k_reranker:
+            sorted_items = sorted(
+                zip(distances, metadatas, documents), key=lambda x: x[0], reverse=True
+            )
+            sorted_items = sorted_items[:k]
+            distances, documents, metadatas = map(list, zip(*sorted_items))
+
         result = {
-            "distances": [[d.metadata.get("score") for d in result]],
-            "documents": [[d.page_content for d in result]],
-            "metadatas": [[d.metadata for d in result]],
+            "distances": [distances],
+            "documents": [documents],
+            "metadatas": [metadatas],
         }
 
         log.info(
@@ -150,6 +171,7 @@ def query_doc_with_hybrid_search(
         )
         return result
     except Exception as e:
+        log.exception(f"Error querying doc {collection_name} with hybrid search: {e}")
         raise e
 
 
@@ -174,12 +196,9 @@ def merge_get_results(get_results: list[dict]) -> dict:
     return result
 
 
-def merge_and_sort_query_results(
-    query_results: list[dict], k: int, reverse: bool = False
-) -> dict:
+def merge_and_sort_query_results(query_results: list[dict], k: int) -> dict:
     # Initialize lists to store combined data
-    combined = []
-    seen_hashes = set()  # To store unique document hashes
+    combined = dict()  # To store documents with unique document hashes
 
     for data in query_results:
         distances = data["distances"][0]
@@ -192,12 +211,17 @@ def merge_and_sort_query_results(
                     document.encode()
                 ).hexdigest()  # Compute a hash for uniqueness
 
-                if doc_hash not in seen_hashes:
-                    seen_hashes.add(doc_hash)
-                    combined.append((distance, document, metadata))
+                if doc_hash not in combined.keys():
+                    combined[doc_hash] = (distance, document, metadata)
+                    continue  # if doc is new, no further comparison is needed
 
+                # if doc is alredy in, but new distance is better, update
+                if distance > combined[doc_hash][0]:
+                    combined[doc_hash] = (distance, document, metadata)
+
+    combined = list(combined.values())
     # Sort the list based on distances
-    combined.sort(key=lambda x: x[0], reverse=reverse)
+    combined.sort(key=lambda x: x[0], reverse=True)
 
     # Slice to keep only the top k elements
     sorted_distances, sorted_documents, sorted_metadatas = (
@@ -237,7 +261,8 @@ def query_collection(
 ) -> dict:
     results = []
     for query in queries:
-        query_embedding = embedding_function(query)
+        log.debug(f"query_collection:query {query}")
+        query_embedding = embedding_function(query, prefix=RAG_EMBEDDING_QUERY_PREFIX)
         for collection_name in collection_names:
             if collection_name:
                 try:
@@ -253,12 +278,7 @@ def query_collection(
             else:
                 pass
 
-    if VECTOR_DB == "chroma":
-        # Chroma uses unconventional cosine similarity, so we don't need to reverse the results
-        # https://docs.trychroma.com/docs/collections/configure#configuring-chroma-collections
-        return merge_and_sort_query_results(results, k=k, reverse=False)
-    else:
-        return merge_and_sort_query_results(results, k=k, reverse=True)
+    return merge_and_sort_query_results(results, k=k)
 
 
 def query_collection_with_hybrid_search(
@@ -267,39 +287,72 @@ def query_collection_with_hybrid_search(
     embedding_function,
     k: int,
     reranking_function,
+    k_reranker: int,
     r: float,
 ) -> dict:
     results = []
     error = False
+    # Fetch collection data once per collection sequentially
+    # Avoid fetching the same data multiple times later
+    collection_results = {}
     for collection_name in collection_names:
         try:
-            for query in queries:
-                result = query_doc_with_hybrid_search(
-                    collection_name=collection_name,
-                    query=query,
-                    embedding_function=embedding_function,
-                    k=k,
-                    reranking_function=reranking_function,
-                    r=r,
-                )
-                results.append(result)
-        except Exception as e:
-            log.exception(
-                "Error when querying the collection with " f"hybrid_search: {e}"
+            log.debug(
+                f"query_collection_with_hybrid_search:VECTOR_DB_CLIENT.get:collection {collection_name}"
             )
-            error = True
+            collection_results[collection_name] = VECTOR_DB_CLIENT.get(
+                collection_name=collection_name
+            )
+        except Exception as e:
+            log.exception(f"Failed to fetch collection {collection_name}: {e}")
+            collection_results[collection_name] = None
 
-    if error:
+    log.info(
+        f"Starting hybrid search for {len(queries)} queries in {len(collection_names)} collections..."
+    )
+
+    def process_query(collection_name, query):
+        try:
+            result = query_doc_with_hybrid_search(
+                collection_name=collection_name,
+                collection_result=collection_results[collection_name],
+                query=query,
+                embedding_function=embedding_function,
+                k=k,
+                reranking_function=reranking_function,
+                k_reranker=k_reranker,
+                r=r,
+            )
+            return result, None
+        except Exception as e:
+            log.exception(f"Error when querying the collection with hybrid_search: {e}")
+            return None, e
+
+    # Prepare tasks for all collections and queries
+    # Avoid running any tasks for collections that failed to fetch data (have assigned None)
+    tasks = [
+        (cn, q)
+        for cn in collection_names
+        if collection_results[cn] is not None
+        for q in queries
+    ]
+
+    with ThreadPoolExecutor() as executor:
+        future_results = [executor.submit(process_query, cn, q) for cn, q in tasks]
+        task_results = [future.result() for future in future_results]
+
+    for result, err in task_results:
+        if err is not None:
+            error = True
+        elif result is not None:
+            results.append(result)
+
+    if error and not results:
         raise Exception(
-            "Hybrid search failed for all collections. Using Non hybrid search as fallback."
+            "Hybrid search failed for all collections. Using Non-hybrid search as fallback."
         )
 
-    if VECTOR_DB == "chroma":
-        # Chroma uses unconventional cosine similarity, so we don't need to reverse the results
-        # https://docs.trychroma.com/docs/collections/configure#configuring-chroma-collections
-        return merge_and_sort_query_results(results, k=k, reverse=False)
-    else:
-        return merge_and_sort_query_results(results, k=k, reverse=True)
+    return merge_and_sort_query_results(results, k=k)
 
 
 def get_embedding_function(
@@ -311,29 +364,38 @@ def get_embedding_function(
     embedding_batch_size,
 ):
     if embedding_engine == "":
-        return lambda query, user=None: embedding_function.encode(query).tolist()
+        return lambda query, prefix=None, user=None: embedding_function.encode(
+            query, **({"prompt": prefix} if prefix else {})
+        ).tolist()
     elif embedding_engine in ["ollama", "openai"]:
-        func = lambda query, user=None: generate_embeddings(
+        func = lambda query, prefix=None, user=None: generate_embeddings(
             engine=embedding_engine,
             model=embedding_model,
             text=query,
+            prefix=prefix,
             url=url,
             key=key,
             user=user,
         )
 
-        def generate_multiple(query, user, func):
+        def generate_multiple(query, prefix, user, func):
             if isinstance(query, list):
                 embeddings = []
                 for i in range(0, len(query), embedding_batch_size):
                     embeddings.extend(
-                        func(query[i : i + embedding_batch_size], user=user)
+                        func(
+                            query[i : i + embedding_batch_size],
+                            prefix=prefix,
+                            user=user,
+                        )
                     )
                 return embeddings
             else:
-                return func(query, user)
+                return func(query, prefix, user)
 
-        return lambda query, user=None: generate_multiple(query, user, func)
+        return lambda query, prefix=None, user=None: generate_multiple(
+            query, prefix, user, func
+        )
     else:
         raise ValueError(f"Unknown embedding engine: {embedding_engine}")
 
@@ -345,6 +407,7 @@ def get_sources_from_files(
     embedding_function,
     k,
     reranking_function,
+    k_reranker,
     r,
     hybrid_search,
     full_context=False,
@@ -461,6 +524,7 @@ def get_sources_from_files(
                                     embedding_function=embedding_function,
                                     k=k,
                                     reranking_function=reranking_function,
+                                    k_reranker=k_reranker,
                                     r=r,
                                 )
                             except Exception as e:
@@ -553,9 +617,17 @@ def generate_openai_batch_embeddings(
     texts: list[str],
     url: str = "https://api.openai.com/v1",
     key: str = "",
+    prefix: str = None,
     user: UserModel = None,
 ) -> Optional[list[list[float]]]:
     try:
+        log.debug(
+            f"generate_openai_batch_embeddings:model {model} batch size: {len(texts)}"
+        )
+        json_data = {"input": texts, "model": model}
+        if isinstance(RAG_EMBEDDING_PREFIX_FIELD_NAME, str) and isinstance(prefix, str):
+            json_data[RAG_EMBEDDING_PREFIX_FIELD_NAME] = prefix
+
         r = requests.post(
             f"{url}/embeddings",
             headers={
@@ -572,7 +644,7 @@ def generate_openai_batch_embeddings(
                     else {}
                 ),
             },
-            json={"input": texts, "model": model},
+            json=json_data,
         )
         r.raise_for_status()
         data = r.json()
@@ -586,9 +658,21 @@ def generate_openai_batch_embeddings(
 
 
 def generate_ollama_batch_embeddings(
-    model: str, texts: list[str], url: str, key: str = "", user: UserModel = None
+    model: str,
+    texts: list[str],
+    url: str,
+    key: str = "",
+    prefix: str = None,
+    user: UserModel = None,
 ) -> Optional[list[list[float]]]:
     try:
+        log.debug(
+            f"generate_ollama_batch_embeddings:model {model} batch size: {len(texts)}"
+        )
+        json_data = {"input": texts, "model": model}
+        if isinstance(RAG_EMBEDDING_PREFIX_FIELD_NAME, str) and isinstance(prefix, str):
+            json_data[RAG_EMBEDDING_PREFIX_FIELD_NAME] = prefix
+
         r = requests.post(
             f"{url}/api/embed",
             headers={
@@ -605,7 +689,7 @@ def generate_ollama_batch_embeddings(
                     else {}
                 ),
             },
-            json={"input": texts, "model": model},
+            json=json_data,
         )
         r.raise_for_status()
         data = r.json()
@@ -619,15 +703,34 @@ def generate_ollama_batch_embeddings(
         return None
 
 
-def generate_embeddings(engine: str, model: str, text: Union[str, list[str]], **kwargs):
+def generate_embeddings(
+    engine: str,
+    model: str,
+    text: Union[str, list[str]],
+    prefix: Union[str, None] = None,
+    **kwargs,
+):
     url = kwargs.get("url", "")
     key = kwargs.get("key", "")
     user = kwargs.get("user")
 
+    if prefix is not None and RAG_EMBEDDING_PREFIX_FIELD_NAME is None:
+        if isinstance(text, list):
+            text = [f"{prefix}{text_element}" for text_element in text]
+        else:
+            text = f"{prefix}{text}"
+
     if engine == "ollama":
         if isinstance(text, list):
             embeddings = generate_ollama_batch_embeddings(
-                **{"model": model, "texts": text, "url": url, "key": key, "user": user}
+                **{
+                    "model": model,
+                    "texts": text,
+                    "url": url,
+                    "key": key,
+                    "prefix": prefix,
+                    "user": user,
+                }
             )
         else:
             embeddings = generate_ollama_batch_embeddings(
@@ -636,16 +739,20 @@ def generate_embeddings(engine: str, model: str, text: Union[str, list[str]], **
                     "texts": [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, user)
+            embeddings = generate_openai_batch_embeddings(
+                model, text, url, key, prefix, user
+            )
         else:
-            embeddings = generate_openai_batch_embeddings(model, [text], url, key, user)
-
+            embeddings = generate_openai_batch_embeddings(
+                model, [text], url, key, prefix, user
+            )
         return embeddings[0] if isinstance(text, str) else embeddings
 
 
@@ -681,9 +788,9 @@ class RerankCompressor(BaseDocumentCompressor):
         else:
             from sentence_transformers import util
 
-            query_embedding = self.embedding_function(query)
+            query_embedding = self.embedding_function(query, RAG_EMBEDDING_QUERY_PREFIX)
             document_embedding = self.embedding_function(
-                [doc.page_content for doc in documents]
+                [doc.page_content for doc in documents], RAG_EMBEDDING_CONTENT_PREFIX
             )
             scores = util.cos_sim(query_embedding, document_embedding)[0]
 

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

@@ -75,10 +75,16 @@ class ChromaClient:
                     n_results=limit,
                 )
 
+                # chromadb has cosine distance, 2 (worst) -> 0 (best). Re-odering to 0 -> 1
+                # https://docs.trychroma.com/docs/collections/configure cosine equation
+                distances: list = result["distances"][0]
+                distances = [2 - dist for dist in distances]
+                distances = [[dist / 2 for dist in distances]]
+
                 return SearchResult(
                     **{
                         "ids": result["ids"],
-                        "distances": result["distances"],
+                        "distances": distances,
                         "documents": result["documents"],
                         "metadatas": result["metadatas"],
                     }
@@ -166,12 +172,19 @@ class ChromaClient:
         filter: Optional[dict] = None,
     ):
         # Delete the items from the collection based on the ids.
-        collection = self.client.get_collection(name=collection_name)
-        if collection:
-            if ids:
-                collection.delete(ids=ids)
-            elif filter:
-                collection.delete(where=filter)
+        try:
+            collection = self.client.get_collection(name=collection_name)
+            if collection:
+                if ids:
+                    collection.delete(ids=ids)
+                elif filter:
+                    collection.delete(where=filter)
+        except Exception as e:
+            # If collection doesn't exist, that's fine - nothing to delete
+            log.debug(
+                f"Attempted to delete from non-existent collection {collection_name}. Ignoring."
+            )
+            pass
 
     def reset(self):
         # Resets the database. This will delete all collections and item entries.

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

@@ -64,7 +64,10 @@ class MilvusClient:
 
             for item in match:
                 _ids.append(item.get("id"))
-                _distances.append(item.get("distance"))
+                # normalize milvus score from [-1, 1] to [0, 1] range
+                # https://milvus.io/docs/de/metric.md
+                _dist = (item.get("distance") + 1.0) / 2.0
+                _distances.append(_dist)
                 _documents.append(item.get("entity", {}).get("data", {}).get("text"))
                 _metadatas.append(item.get("entity", {}).get("metadata"))
 

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

@@ -1,4 +1,5 @@
 from opensearchpy import OpenSearch
+from opensearchpy.helpers import bulk
 from typing import Optional
 
 from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
@@ -21,7 +22,13 @@ class OpenSearchClient:
             http_auth=(OPENSEARCH_USERNAME, OPENSEARCH_PASSWORD),
         )
 
+    def _get_index_name(self, collection_name: str) -> str:
+        return f"{self.index_prefix}_{collection_name}"
+
     def _result_to_get_result(self, result) -> GetResult:
+        if not result["hits"]["hits"]:
+            return None
+
         ids = []
         documents = []
         metadatas = []
@@ -31,9 +38,12 @@ class OpenSearchClient:
             documents.append(hit["_source"].get("text"))
             metadatas.append(hit["_source"].get("metadata"))
 
-        return GetResult(ids=ids, documents=documents, metadatas=metadatas)
+        return GetResult(ids=[ids], documents=[documents], metadatas=[metadatas])
 
     def _result_to_search_result(self, result) -> SearchResult:
+        if not result["hits"]["hits"]:
+            return None
+
         ids = []
         distances = []
         documents = []
@@ -46,34 +56,40 @@ class OpenSearchClient:
             metadatas.append(hit["_source"].get("metadata"))
 
         return SearchResult(
-            ids=ids, distances=distances, documents=documents, metadatas=metadatas
+            ids=[ids],
+            distances=[distances],
+            documents=[documents],
+            metadatas=[metadatas],
         )
 
     def _create_index(self, collection_name: str, dimension: int):
         body = {
+            "settings": {"index": {"knn": True}},
             "mappings": {
                 "properties": {
                     "id": {"type": "keyword"},
                     "vector": {
-                        "type": "dense_vector",
-                        "dims": dimension,  # Adjust based on your vector dimensions
-                        "index": true,
+                        "type": "knn_vector",
+                        "dimension": dimension,  # Adjust based on your vector dimensions
+                        "index": True,
                         "similarity": "faiss",
                         "method": {
                             "name": "hnsw",
-                            "space_type": "ip",  # Use inner product to approximate cosine similarity
+                            "space_type": "innerproduct",  # Use inner product to approximate cosine similarity
                             "engine": "faiss",
-                            "ef_construction": 128,
-                            "m": 16,
+                            "parameters": {
+                                "ef_construction": 128,
+                                "m": 16,
+                            },
                         },
                     },
                     "text": {"type": "text"},
                     "metadata": {"type": "object"},
                 }
-            }
+            },
         }
         self.client.indices.create(
-            index=f"{self.index_prefix}_{collection_name}", body=body
+            index=self._get_index_name(collection_name), body=body
         )
 
     def _create_batches(self, items: list[VectorItem], batch_size=100):
@@ -83,39 +99,45 @@ class OpenSearchClient:
     def has_collection(self, collection_name: str) -> bool:
         # has_collection here means has index.
         # We are simply adapting to the norms of the other DBs.
-        return self.client.indices.exists(
-            index=f"{self.index_prefix}_{collection_name}"
-        )
+        return self.client.indices.exists(index=self._get_index_name(collection_name))
 
-    def delete_colleciton(self, collection_name: str):
+    def delete_collection(self, collection_name: str):
         # delete_collection here means delete index.
         # We are simply adapting to the norms of the other DBs.
-        self.client.indices.delete(index=f"{self.index_prefix}_{collection_name}")
+        self.client.indices.delete(index=self._get_index_name(collection_name))
 
     def search(
-        self, collection_name: str, vectors: list[list[float]], limit: int
+        self, collection_name: str, vectors: list[list[float | int]], limit: int
     ) -> Optional[SearchResult]:
-        query = {
-            "size": limit,
-            "_source": ["text", "metadata"],
-            "query": {
-                "script_score": {
-                    "query": {"match_all": {}},
-                    "script": {
-                        "source": "cosineSimilarity(params.vector, 'vector') + 1.0",
-                        "params": {
-                            "vector": vectors[0]
-                        },  # Assuming single query vector
-                    },
-                }
-            },
-        }
+        try:
+            if not self.has_collection(collection_name):
+                return None
 
-        result = self.client.search(
-            index=f"{self.index_prefix}_{collection_name}", body=query
-        )
+            query = {
+                "size": limit,
+                "_source": ["text", "metadata"],
+                "query": {
+                    "script_score": {
+                        "query": {"match_all": {}},
+                        "script": {
+                            "source": "(cosineSimilarity(params.query_value, doc[params.field]) + 1.0) / 2.0",
+                            "params": {
+                                "field": "vector",
+                                "query_value": vectors[0],
+                            },  # Assuming single query vector
+                        },
+                    }
+                },
+            }
 
-        return self._result_to_search_result(result)
+            result = self.client.search(
+                index=self._get_index_name(collection_name), body=query
+            )
+
+            return self._result_to_search_result(result)
+
+        except Exception as e:
+            return None
 
     def query(
         self, collection_name: str, filter: dict, limit: Optional[int] = None
@@ -129,13 +151,15 @@ class OpenSearchClient:
         }
 
         for field, value in filter.items():
-            query_body["query"]["bool"]["filter"].append({"term": {field: value}})
+            query_body["query"]["bool"]["filter"].append(
+                {"match": {"metadata." + str(field): value}}
+            )
 
         size = limit if limit else 10
 
         try:
             result = self.client.search(
-                index=f"{self.index_prefix}_{collection_name}",
+                index=self._get_index_name(collection_name),
                 body=query_body,
                 size=size,
             )
@@ -146,14 +170,14 @@ class OpenSearchClient:
             return None
 
     def _create_index_if_not_exists(self, collection_name: str, dimension: int):
-        if not self.has_index(collection_name):
+        if not self.has_collection(collection_name):
             self._create_index(collection_name, dimension)
 
     def get(self, collection_name: str) -> Optional[GetResult]:
         query = {"query": {"match_all": {}}, "_source": ["text", "metadata"]}
 
         result = self.client.search(
-            index=f"{self.index_prefix}_{collection_name}", body=query
+            index=self._get_index_name(collection_name), body=query
         )
         return self._result_to_get_result(result)
 
@@ -165,18 +189,18 @@ class OpenSearchClient:
         for batch in self._create_batches(items):
             actions = [
                 {
-                    "index": {
-                        "_id": item["id"],
-                        "_source": {
-                            "vector": item["vector"],
-                            "text": item["text"],
-                            "metadata": item["metadata"],
-                        },
-                    }
+                    "_op_type": "index",
+                    "_index": self._get_index_name(collection_name),
+                    "_id": item["id"],
+                    "_source": {
+                        "vector": item["vector"],
+                        "text": item["text"],
+                        "metadata": item["metadata"],
+                    },
                 }
                 for item in batch
             ]
-            self.client.bulk(actions)
+            bulk(self.client, actions)
 
     def upsert(self, collection_name: str, items: list[VectorItem]):
         self._create_index_if_not_exists(
@@ -186,26 +210,47 @@ class OpenSearchClient:
         for batch in self._create_batches(items):
             actions = [
                 {
-                    "index": {
-                        "_id": item["id"],
-                        "_index": f"{self.index_prefix}_{collection_name}",
-                        "_source": {
-                            "vector": item["vector"],
-                            "text": item["text"],
-                            "metadata": item["metadata"],
-                        },
-                    }
+                    "_op_type": "update",
+                    "_index": self._get_index_name(collection_name),
+                    "_id": item["id"],
+                    "doc": {
+                        "vector": item["vector"],
+                        "text": item["text"],
+                        "metadata": item["metadata"],
+                    },
+                    "doc_as_upsert": True,
                 }
                 for item in batch
             ]
-            self.client.bulk(actions)
+            bulk(self.client, actions)
 
-    def delete(self, collection_name: str, ids: list[str]):
-        actions = [
-            {"delete": {"_index": f"{self.index_prefix}_{collection_name}", "_id": id}}
-            for id in ids
-        ]
-        self.client.bulk(body=actions)
+    def delete(
+        self,
+        collection_name: str,
+        ids: Optional[list[str]] = None,
+        filter: Optional[dict] = None,
+    ):
+        if ids:
+            actions = [
+                {
+                    "_op_type": "delete",
+                    "_index": self._get_index_name(collection_name),
+                    "_id": id,
+                }
+                for id in ids
+            ]
+            bulk(self.client, actions)
+        elif filter:
+            query_body = {
+                "query": {"bool": {"filter": []}},
+            }
+            for field, value in filter.items():
+                query_body["query"]["bool"]["filter"].append(
+                    {"match": {"metadata." + str(field): value}}
+                )
+            self.client.delete_by_query(
+                index=self._get_index_name(collection_name), body=query_body
+            )
 
     def reset(self):
         indices = self.client.indices.get(index=f"{self.index_prefix}_*")

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

@@ -278,7 +278,9 @@ class PgvectorClient:
             for row in results:
                 qid = int(row.qid)
                 ids[qid].append(row.id)
-                distances[qid].append(row.distance)
+                # normalize and re-orders pgvec distance from [2, 0] to [0, 1] score range
+                # https://github.com/pgvector/pgvector?tab=readme-ov-file#querying
+                distances[qid].append((2.0 - row.distance) / 2.0)
                 documents[qid].append(row.text)
                 metadatas[qid].append(row.vmetadata)
 

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

@@ -99,7 +99,8 @@ class QdrantClient:
             ids=get_result.ids,
             documents=get_result.documents,
             metadatas=get_result.metadatas,
-            distances=[[point.score for point in query_response.points]],
+            # qdrant distance is [-1, 1], normalize to [0, 1]
+            distances=[[(point.score + 1.0) / 2.0 for point in query_response.points]],
         )
 
     def query(self, collection_name: str, filter: dict, limit: Optional[int] = None):

backend/open_webui/retrieval/web/duckduckgo.py

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

backend/open_webui/retrieval/web/sougou.py

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

backend/open_webui/retrieval/web/utils.py

@@ -24,13 +24,17 @@ from langchain_community.document_loaders import PlaywrightURLLoader, WebBaseLoa
 from langchain_community.document_loaders.firecrawl import FireCrawlLoader
 from langchain_community.document_loaders.base import BaseLoader
 from langchain_core.documents import Document
+from open_webui.retrieval.loaders.tavily import TavilyLoader
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.config import (
     ENABLE_RAG_LOCAL_WEB_FETCH,
-    PLAYWRIGHT_WS_URI,
-    RAG_WEB_LOADER_ENGINE,
+    PLAYWRIGHT_WS_URL,
+    PLAYWRIGHT_TIMEOUT,
+    WEB_LOADER_ENGINE,
     FIRECRAWL_API_BASE_URL,
     FIRECRAWL_API_KEY,
+    TAVILY_API_KEY,
+    TAVILY_EXTRACT_DEPTH,
 )
 from open_webui.env import SRC_LOG_LEVELS
 
@@ -113,7 +117,47 @@ def verify_ssl_cert(url: str) -> bool:
         return False
 
 
-class SafeFireCrawlLoader(BaseLoader):
+class RateLimitMixin:
+    async def _wait_for_rate_limit(self):
+        """Wait to respect the rate limit if specified."""
+        if self.requests_per_second and self.last_request_time:
+            min_interval = timedelta(seconds=1.0 / self.requests_per_second)
+            time_since_last = datetime.now() - self.last_request_time
+            if time_since_last < min_interval:
+                await asyncio.sleep((min_interval - time_since_last).total_seconds())
+        self.last_request_time = datetime.now()
+
+    def _sync_wait_for_rate_limit(self):
+        """Synchronous version of rate limit wait."""
+        if self.requests_per_second and self.last_request_time:
+            min_interval = timedelta(seconds=1.0 / self.requests_per_second)
+            time_since_last = datetime.now() - self.last_request_time
+            if time_since_last < min_interval:
+                time.sleep((min_interval - time_since_last).total_seconds())
+        self.last_request_time = datetime.now()
+
+
+class URLProcessingMixin:
+    def _verify_ssl_cert(self, url: str) -> bool:
+        """Verify SSL certificate for a URL."""
+        return verify_ssl_cert(url)
+
+    async def _safe_process_url(self, url: str) -> bool:
+        """Perform safety checks before processing a URL."""
+        if self.verify_ssl and not self._verify_ssl_cert(url):
+            raise ValueError(f"SSL certificate verification failed for {url}")
+        await self._wait_for_rate_limit()
+        return True
+
+    def _safe_process_url_sync(self, url: str) -> bool:
+        """Synchronous version of safety checks."""
+        if self.verify_ssl and not self._verify_ssl_cert(url):
+            raise ValueError(f"SSL certificate verification failed for {url}")
+        self._sync_wait_for_rate_limit()
+        return True
+
+
+class SafeFireCrawlLoader(BaseLoader, RateLimitMixin, URLProcessingMixin):
     def __init__(
         self,
         web_paths,
@@ -184,7 +228,7 @@ class SafeFireCrawlLoader(BaseLoader):
                 yield from loader.lazy_load()
             except Exception as e:
                 if self.continue_on_failure:
-                    log.exception(e, "Error loading %s", url)
+                    log.exception(f"Error loading {url}: {e}")
                     continue
                 raise e
 
@@ -204,47 +248,124 @@ class SafeFireCrawlLoader(BaseLoader):
                     yield document
             except Exception as e:
                 if self.continue_on_failure:
-                    log.exception(e, "Error loading %s", url)
+                    log.exception(f"Error loading {url}: {e}")
                     continue
                 raise e
 
-    def _verify_ssl_cert(self, url: str) -> bool:
-        return verify_ssl_cert(url)
 
-    async def _wait_for_rate_limit(self):
-        """Wait to respect the rate limit if specified."""
-        if self.requests_per_second and self.last_request_time:
-            min_interval = timedelta(seconds=1.0 / self.requests_per_second)
-            time_since_last = datetime.now() - self.last_request_time
-            if time_since_last < min_interval:
-                await asyncio.sleep((min_interval - time_since_last).total_seconds())
-        self.last_request_time = datetime.now()
+class SafeTavilyLoader(BaseLoader, RateLimitMixin, URLProcessingMixin):
+    def __init__(
+        self,
+        web_paths: Union[str, List[str]],
+        api_key: str,
+        extract_depth: Literal["basic", "advanced"] = "basic",
+        continue_on_failure: bool = True,
+        requests_per_second: Optional[float] = None,
+        verify_ssl: bool = True,
+        trust_env: bool = False,
+        proxy: Optional[Dict[str, str]] = None,
+    ):
+        """Initialize SafeTavilyLoader with rate limiting and SSL verification support.
 
-    def _sync_wait_for_rate_limit(self):
-        """Synchronous version of rate limit wait."""
-        if self.requests_per_second and self.last_request_time:
-            min_interval = timedelta(seconds=1.0 / self.requests_per_second)
-            time_since_last = datetime.now() - self.last_request_time
-            if time_since_last < min_interval:
-                time.sleep((min_interval - time_since_last).total_seconds())
-        self.last_request_time = datetime.now()
+        Args:
+            web_paths: List of URLs/paths to process.
+            api_key: The Tavily API key.
+            extract_depth: Depth of extraction ("basic" or "advanced").
+            continue_on_failure: Whether to continue if extraction of a URL fails.
+            requests_per_second: Number of requests per second to limit to.
+            verify_ssl: If True, verify SSL certificates.
+            trust_env: If True, use proxy settings from environment variables.
+            proxy: Optional proxy configuration.
+        """
+        # Initialize proxy configuration if using environment variables
+        proxy_server = proxy.get("server") if proxy else None
+        if trust_env and not proxy_server:
+            env_proxies = urllib.request.getproxies()
+            env_proxy_server = env_proxies.get("https") or env_proxies.get("http")
+            if env_proxy_server:
+                if proxy:
+                    proxy["server"] = env_proxy_server
+                else:
+                    proxy = {"server": env_proxy_server}
 
-    async def _safe_process_url(self, url: str) -> bool:
-        """Perform safety checks before processing a URL."""
-        if self.verify_ssl and not self._verify_ssl_cert(url):
-            raise ValueError(f"SSL certificate verification failed for {url}")
-        await self._wait_for_rate_limit()
-        return True
+        # Store parameters for creating TavilyLoader instances
+        self.web_paths = web_paths if isinstance(web_paths, list) else [web_paths]
+        self.api_key = api_key
+        self.extract_depth = extract_depth
+        self.continue_on_failure = continue_on_failure
+        self.verify_ssl = verify_ssl
+        self.trust_env = trust_env
+        self.proxy = proxy
 
-    def _safe_process_url_sync(self, url: str) -> bool:
-        """Synchronous version of safety checks."""
-        if self.verify_ssl and not self._verify_ssl_cert(url):
-            raise ValueError(f"SSL certificate verification failed for {url}")
-        self._sync_wait_for_rate_limit()
-        return True
+        # Add rate limiting
+        self.requests_per_second = requests_per_second
+        self.last_request_time = None
+
+    def lazy_load(self) -> Iterator[Document]:
+        """Load documents with rate limiting support, delegating to TavilyLoader."""
+        valid_urls = []
+        for url in self.web_paths:
+            try:
+                self._safe_process_url_sync(url)
+                valid_urls.append(url)
+            except Exception as e:
+                log.warning(f"SSL verification failed for {url}: {str(e)}")
+                if not self.continue_on_failure:
+                    raise e
+        if not valid_urls:
+            if self.continue_on_failure:
+                log.warning("No valid URLs to process after SSL verification")
+                return
+            raise ValueError("No valid URLs to process after SSL verification")
+        try:
+            loader = TavilyLoader(
+                urls=valid_urls,
+                api_key=self.api_key,
+                extract_depth=self.extract_depth,
+                continue_on_failure=self.continue_on_failure,
+            )
+            yield from loader.lazy_load()
+        except Exception as e:
+            if self.continue_on_failure:
+                log.exception(f"Error extracting content from URLs: {e}")
+            else:
+                raise e
+
+    async def alazy_load(self) -> AsyncIterator[Document]:
+        """Async version with rate limiting and SSL verification."""
+        valid_urls = []
+        for url in self.web_paths:
+            try:
+                await self._safe_process_url(url)
+                valid_urls.append(url)
+            except Exception as e:
+                log.warning(f"SSL verification failed for {url}: {str(e)}")
+                if not self.continue_on_failure:
+                    raise e
+
+        if not valid_urls:
+            if self.continue_on_failure:
+                log.warning("No valid URLs to process after SSL verification")
+                return
+            raise ValueError("No valid URLs to process after SSL verification")
+
+        try:
+            loader = TavilyLoader(
+                urls=valid_urls,
+                api_key=self.api_key,
+                extract_depth=self.extract_depth,
+                continue_on_failure=self.continue_on_failure,
+            )
+            async for document in loader.alazy_load():
+                yield document
+        except Exception as e:
+            if self.continue_on_failure:
+                log.exception(f"Error loading URLs: {e}")
+            else:
+                raise e
 
 
-class SafePlaywrightURLLoader(PlaywrightURLLoader):
+class SafePlaywrightURLLoader(PlaywrightURLLoader, RateLimitMixin, URLProcessingMixin):
     """Load HTML pages safely with Playwright, supporting SSL verification, rate limiting, and remote browser connection.
 
     Attributes:
@@ -256,6 +377,7 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
         headless (bool): If True, the browser will run in headless mode.
         proxy (dict): Proxy override settings for the Playwright session.
         playwright_ws_url (Optional[str]): WebSocket endpoint URI for remote browser connection.
+        playwright_timeout (Optional[int]): Maximum operation time in milliseconds.
     """
 
     def __init__(
@@ -269,6 +391,7 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
         remove_selectors: Optional[List[str]] = None,
         proxy: Optional[Dict[str, str]] = None,
         playwright_ws_url: Optional[str] = None,
+        playwright_timeout: Optional[int] = 10000,
     ):
         """Initialize with additional safety parameters and remote browser support."""
 
@@ -295,6 +418,7 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
         self.last_request_time = None
         self.playwright_ws_url = playwright_ws_url
         self.trust_env = trust_env
+        self.playwright_timeout = playwright_timeout
 
     def lazy_load(self) -> Iterator[Document]:
         """Safely load URLs synchronously with support for remote browser."""
@@ -311,7 +435,7 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
                 try:
                     self._safe_process_url_sync(url)
                     page = browser.new_page()
-                    response = page.goto(url)
+                    response = page.goto(url, timeout=self.playwright_timeout)
                     if response is None:
                         raise ValueError(f"page.goto() returned None for url {url}")
 
@@ -320,7 +444,7 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
                     yield Document(page_content=text, metadata=metadata)
                 except Exception as e:
                     if self.continue_on_failure:
-                        log.exception(e, "Error loading %s", url)
+                        log.exception(f"Error loading {url}: {e}")
                         continue
                     raise e
             browser.close()
@@ -342,7 +466,7 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
                 try:
                     await self._safe_process_url(url)
                     page = await browser.new_page()
-                    response = await page.goto(url)
+                    response = await page.goto(url, timeout=self.playwright_timeout)
                     if response is None:
                         raise ValueError(f"page.goto() returned None for url {url}")
 
@@ -351,46 +475,11 @@ class SafePlaywrightURLLoader(PlaywrightURLLoader):
                     yield Document(page_content=text, metadata=metadata)
                 except Exception as e:
                     if self.continue_on_failure:
-                        log.exception(e, "Error loading %s", url)
+                        log.exception(f"Error loading {url}: {e}")
                         continue
                     raise e
             await browser.close()
 
-    def _verify_ssl_cert(self, url: str) -> bool:
-        return verify_ssl_cert(url)
-
-    async def _wait_for_rate_limit(self):
-        """Wait to respect the rate limit if specified."""
-        if self.requests_per_second and self.last_request_time:
-            min_interval = timedelta(seconds=1.0 / self.requests_per_second)
-            time_since_last = datetime.now() - self.last_request_time
-            if time_since_last < min_interval:
-                await asyncio.sleep((min_interval - time_since_last).total_seconds())
-        self.last_request_time = datetime.now()
-
-    def _sync_wait_for_rate_limit(self):
-        """Synchronous version of rate limit wait."""
-        if self.requests_per_second and self.last_request_time:
-            min_interval = timedelta(seconds=1.0 / self.requests_per_second)
-            time_since_last = datetime.now() - self.last_request_time
-            if time_since_last < min_interval:
-                time.sleep((min_interval - time_since_last).total_seconds())
-        self.last_request_time = datetime.now()
-
-    async def _safe_process_url(self, url: str) -> bool:
-        """Perform safety checks before processing a URL."""
-        if self.verify_ssl and not self._verify_ssl_cert(url):
-            raise ValueError(f"SSL certificate verification failed for {url}")
-        await self._wait_for_rate_limit()
-        return True
-
-    def _safe_process_url_sync(self, url: str) -> bool:
-        """Synchronous version of safety checks."""
-        if self.verify_ssl and not self._verify_ssl_cert(url):
-            raise ValueError(f"SSL certificate verification failed for {url}")
-        self._sync_wait_for_rate_limit()
-        return True
-
 
 class SafeWebBaseLoader(WebBaseLoader):
     """WebBaseLoader with enhanced error handling for URLs."""
@@ -472,7 +561,7 @@ class SafeWebBaseLoader(WebBaseLoader):
                 yield Document(page_content=text, metadata=metadata)
             except Exception as e:
                 # Log the error and continue with the next URL
-                log.exception(e, "Error loading %s", path)
+                log.exception(f"Error loading {path}: {e}")
 
     async def alazy_load(self) -> AsyncIterator[Document]:
         """Async lazy load text from the url(s) in web_path."""
@@ -495,12 +584,6 @@ class SafeWebBaseLoader(WebBaseLoader):
         return [document async for document in self.alazy_load()]
 
 
-RAG_WEB_LOADER_ENGINES = defaultdict(lambda: SafeWebBaseLoader)
-RAG_WEB_LOADER_ENGINES["playwright"] = SafePlaywrightURLLoader
-RAG_WEB_LOADER_ENGINES["safe_web"] = SafeWebBaseLoader
-RAG_WEB_LOADER_ENGINES["firecrawl"] = SafeFireCrawlLoader
-
-
 def get_web_loader(
     urls: Union[str, Sequence[str]],
     verify_ssl: bool = True,
@@ -518,21 +601,36 @@ def get_web_loader(
         "trust_env": trust_env,
     }
 
-    if PLAYWRIGHT_WS_URI.value:
-        web_loader_args["playwright_ws_url"] = PLAYWRIGHT_WS_URI.value
+    if WEB_LOADER_ENGINE.value == "" or WEB_LOADER_ENGINE.value == "safe_web":
+        WebLoaderClass = SafeWebBaseLoader
+    if WEB_LOADER_ENGINE.value == "playwright":
+        WebLoaderClass = SafePlaywrightURLLoader
+        web_loader_args["playwright_timeout"] = PLAYWRIGHT_TIMEOUT.value * 1000
+        if PLAYWRIGHT_WS_URL.value:
+            web_loader_args["playwright_ws_url"] = PLAYWRIGHT_WS_URL.value
 
-    if RAG_WEB_LOADER_ENGINE.value == "firecrawl":
+    if WEB_LOADER_ENGINE.value == "firecrawl":
+        WebLoaderClass = SafeFireCrawlLoader
         web_loader_args["api_key"] = FIRECRAWL_API_KEY.value
         web_loader_args["api_url"] = FIRECRAWL_API_BASE_URL.value
 
-    # Create the appropriate WebLoader based on the configuration
-    WebLoaderClass = RAG_WEB_LOADER_ENGINES[RAG_WEB_LOADER_ENGINE.value]
-    web_loader = WebLoaderClass(**web_loader_args)
+    if WEB_LOADER_ENGINE.value == "tavily":
+        WebLoaderClass = SafeTavilyLoader
+        web_loader_args["api_key"] = TAVILY_API_KEY.value
+        web_loader_args["extract_depth"] = TAVILY_EXTRACT_DEPTH.value
 
-    log.debug(
-        "Using RAG_WEB_LOADER_ENGINE %s for %s URLs",
-        web_loader.__class__.__name__,
-        len(safe_urls),
-    )
+    if WebLoaderClass:
+        web_loader = WebLoaderClass(**web_loader_args)
 
-    return web_loader
+        log.debug(
+            "Using WEB_LOADER_ENGINE %s for %s URLs",
+            web_loader.__class__.__name__,
+            len(safe_urls),
+        )
+
+        return web_loader
+    else:
+        raise ValueError(
+            f"Invalid WEB_LOADER_ENGINE: {WEB_LOADER_ENGINE.value}. "
+            "Please set it to 'safe_web', 'playwright', 'firecrawl', or 'tavily'."
+        )

backend/open_webui/routers/audio.py

@@ -50,6 +50,8 @@ router = APIRouter()
 # Constants
 MAX_FILE_SIZE_MB = 25
 MAX_FILE_SIZE = MAX_FILE_SIZE_MB * 1024 * 1024  # Convert MB to bytes
+AZURE_MAX_FILE_SIZE_MB = 200
+AZURE_MAX_FILE_SIZE = AZURE_MAX_FILE_SIZE_MB * 1024 * 1024  # Convert MB to bytes
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["AUDIO"])
@@ -68,8 +70,8 @@ from pydub import AudioSegment
 from pydub.utils import mediainfo
 
 
-def is_mp4_audio(file_path):
-    """Check if the given file is an MP4 audio file."""
+def get_audio_format(file_path):
+    """Check if the given file needs to be converted to a different format."""
     if not os.path.isfile(file_path):
         log.error(f"File not found: {file_path}")
         return False
@@ -80,13 +82,17 @@ def is_mp4_audio(file_path):
         and info.get("codec_type") == "audio"
         and info.get("codec_tag_string") == "mp4a"
     ):
-        return True
-    return False
+        return "mp4"
+    elif info.get("format_name") == "ogg":
+        return "ogg"
+    elif info.get("format_name") == "matroska,webm":
+        return "webm"
+    return None
 
 
-def convert_mp4_to_wav(file_path, output_path):
-    """Convert MP4 audio file to WAV format."""
-    audio = AudioSegment.from_file(file_path, format="mp4")
+def convert_audio_to_wav(file_path, output_path, conversion_type):
+    """Convert MP4/OGG audio file to WAV format."""
+    audio = AudioSegment.from_file(file_path, format=conversion_type)
     audio.export(output_path, format="wav")
     log.info(f"Converted {file_path} to {output_path}")
 
@@ -141,6 +147,9 @@ class STTConfigForm(BaseModel):
     MODEL: str
     WHISPER_MODEL: str
     DEEPGRAM_API_KEY: str
+    AZURE_API_KEY: str
+    AZURE_REGION: str
+    AZURE_LOCALES: str
 
 
 class AudioConfigUpdateForm(BaseModel):
@@ -169,6 +178,9 @@ async def get_audio_config(request: Request, user=Depends(get_admin_user)):
             "MODEL": request.app.state.config.STT_MODEL,
             "WHISPER_MODEL": request.app.state.config.WHISPER_MODEL,
             "DEEPGRAM_API_KEY": request.app.state.config.DEEPGRAM_API_KEY,
+            "AZURE_API_KEY": request.app.state.config.AUDIO_STT_AZURE_API_KEY,
+            "AZURE_REGION": request.app.state.config.AUDIO_STT_AZURE_REGION,
+            "AZURE_LOCALES": request.app.state.config.AUDIO_STT_AZURE_LOCALES,
         },
     }
 
@@ -195,6 +207,9 @@ async def update_audio_config(
     request.app.state.config.STT_MODEL = form_data.stt.MODEL
     request.app.state.config.WHISPER_MODEL = form_data.stt.WHISPER_MODEL
     request.app.state.config.DEEPGRAM_API_KEY = form_data.stt.DEEPGRAM_API_KEY
+    request.app.state.config.AUDIO_STT_AZURE_API_KEY = form_data.stt.AZURE_API_KEY
+    request.app.state.config.AUDIO_STT_AZURE_REGION = form_data.stt.AZURE_REGION
+    request.app.state.config.AUDIO_STT_AZURE_LOCALES = form_data.stt.AZURE_LOCALES
 
     if request.app.state.config.STT_ENGINE == "":
         request.app.state.faster_whisper_model = set_faster_whisper_model(
@@ -220,6 +235,9 @@ async def update_audio_config(
             "MODEL": request.app.state.config.STT_MODEL,
             "WHISPER_MODEL": request.app.state.config.WHISPER_MODEL,
             "DEEPGRAM_API_KEY": request.app.state.config.DEEPGRAM_API_KEY,
+            "AZURE_API_KEY": request.app.state.config.AUDIO_STT_AZURE_API_KEY,
+            "AZURE_REGION": request.app.state.config.AUDIO_STT_AZURE_REGION,
+            "AZURE_LOCALES": request.app.state.config.AUDIO_STT_AZURE_LOCALES,
         },
     }
 
@@ -312,7 +330,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500),
+                status_code=getattr(r, "status", 500) if r else 500,
                 detail=detail if detail else "Open WebUI: Server Connection Error",
             )
 
@@ -366,7 +384,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500),
+                status_code=getattr(r, "status", 500) if r else 500,
                 detail=detail if detail else "Open WebUI: Server Connection Error",
             )
 
@@ -422,7 +440,7 @@ async def speech(request: Request, user=Depends(get_verified_user)):
                 detail = f"External: {e}"
 
             raise HTTPException(
-                status_code=getattr(r, "status", 500),
+                status_code=getattr(r, "status", 500) if r else 500,
                 detail=detail if detail else "Open WebUI: Server Connection Error",
             )
 
@@ -496,10 +514,15 @@ def transcribe(request: Request, file_path):
         log.debug(data)
         return data
     elif request.app.state.config.STT_ENGINE == "openai":
-        if is_mp4_audio(file_path):
-            os.rename(file_path, file_path.replace(".wav", ".mp4"))
-            # Convert MP4 audio file to WAV format
-            convert_mp4_to_wav(file_path.replace(".wav", ".mp4"), file_path)
+        audio_format = get_audio_format(file_path)
+        if audio_format:
+            os.rename(file_path, file_path.replace(".wav", f".{audio_format}"))
+            # Convert unsupported audio file to WAV format
+            convert_audio_to_wav(
+                file_path.replace(".wav", f".{audio_format}"),
+                file_path,
+                audio_format,
+            )
 
         r = None
         try:
@@ -598,6 +621,119 @@ def transcribe(request: Request, file_path):
                     detail = f"External: {e}"
             raise Exception(detail if detail else "Open WebUI: Server Connection Error")
 
+    elif request.app.state.config.STT_ENGINE == "azure":
+        # Check file exists and size
+        if not os.path.exists(file_path):
+            raise HTTPException(status_code=400, detail="Audio file not found")
+
+        # Check file size (Azure has a larger limit of 200MB)
+        file_size = os.path.getsize(file_path)
+        if file_size > AZURE_MAX_FILE_SIZE:
+            raise HTTPException(
+                status_code=400,
+                detail=f"File size exceeds Azure's limit of {AZURE_MAX_FILE_SIZE_MB}MB",
+            )
+
+        api_key = request.app.state.config.AUDIO_STT_AZURE_API_KEY
+        region = request.app.state.config.AUDIO_STT_AZURE_REGION
+        locales = request.app.state.config.AUDIO_STT_AZURE_LOCALES
+
+        # IF NO LOCALES, USE DEFAULTS
+        if len(locales) < 2:
+            locales = [
+                "en-US",
+                "es-ES",
+                "es-MX",
+                "fr-FR",
+                "hi-IN",
+                "it-IT",
+                "de-DE",
+                "en-GB",
+                "en-IN",
+                "ja-JP",
+                "ko-KR",
+                "pt-BR",
+                "zh-CN",
+            ]
+            locales = ",".join(locales)
+
+        if not api_key or not region:
+            raise HTTPException(
+                status_code=400,
+                detail="Azure API key and region are required for Azure STT",
+            )
+
+        r = None
+        try:
+            # Prepare the request
+            data = {
+                "definition": json.dumps(
+                    {
+                        "locales": locales.split(","),
+                        "diarization": {"maxSpeakers": 3, "enabled": True},
+                    }
+                    if locales
+                    else {}
+                )
+            }
+            url = f"https://{region}.api.cognitive.microsoft.com/speechtotext/transcriptions:transcribe?api-version=2024-11-15"
+
+            # Use context manager to ensure file is properly closed
+            with open(file_path, "rb") as audio_file:
+                r = requests.post(
+                    url=url,
+                    files={"audio": audio_file},
+                    data=data,
+                    headers={
+                        "Ocp-Apim-Subscription-Key": api_key,
+                    },
+                )
+
+            r.raise_for_status()
+            response = r.json()
+
+            # Extract transcript from response
+            if not response.get("combinedPhrases"):
+                raise ValueError("No transcription found in response")
+
+            # Get the full transcript from combinedPhrases
+            transcript = response["combinedPhrases"][0].get("text", "").strip()
+            if not transcript:
+                raise ValueError("Empty transcript in response")
+
+            data = {"text": transcript}
+
+            # Save transcript to json file (consistent with other providers)
+            transcript_file = f"{file_dir}/{id}.json"
+            with open(transcript_file, "w") as f:
+                json.dump(data, f)
+
+            log.debug(data)
+            return data
+
+        except (KeyError, IndexError, ValueError) as e:
+            log.exception("Error parsing Azure response")
+            raise HTTPException(
+                status_code=500,
+                detail=f"Failed to parse Azure response: {str(e)}",
+            )
+        except requests.exceptions.RequestException as e:
+            log.exception(e)
+            detail = None
+
+            try:
+                if r is not None and r.status_code != 200:
+                    res = r.json()
+                    if "error" in res:
+                        detail = f"External: {res['error'].get('message', '')}"
+            except Exception:
+                detail = f"External: {e}"
+
+            raise HTTPException(
+                status_code=getattr(r, "status_code", 500) if r else 500,
+                detail=detail if detail else "Open WebUI: Server Connection Error",
+            )
+
 
 def compress_audio(file_path):
     if os.path.getsize(file_path) > MAX_FILE_SIZE:
@@ -625,7 +761,9 @@ def transcription(
 ):
     log.info(f"file.content_type: {file.content_type}")
 
-    if file.content_type not in ["audio/mpeg", "audio/wav", "audio/ogg", "audio/x-m4a"]:
+    supported_filetypes = ("audio/mpeg", "audio/wav", "audio/ogg", "audio/x-m4a")
+
+    if not file.content_type.startswith(supported_filetypes):
         raise HTTPException(
             status_code=status.HTTP_400_BAD_REQUEST,
             detail=ERROR_MESSAGES.FILE_NOT_SUPPORTED,

backend/open_webui/routers/auths.py

@@ -194,8 +194,8 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
             ciphers=LDAP_CIPHERS,
         )
     except Exception as e:
-        log.error(f"An error occurred on TLS: {str(e)}")
-        raise HTTPException(400, detail=str(e))
+        log.error(f"TLS configuration error: {str(e)}")
+        raise HTTPException(400, detail="Failed to configure TLS for LDAP connection.")
 
     try:
         server = Server(
@@ -210,7 +210,7 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
             LDAP_APP_DN,
             LDAP_APP_PASSWORD,
             auto_bind="NONE",
-            authentication="SIMPLE",
+            authentication="SIMPLE" if LDAP_APP_DN else "ANONYMOUS",
         )
         if not connection_app.bind():
             raise HTTPException(400, detail="Application account bind failed")
@@ -230,11 +230,13 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
 
         entry = connection_app.entries[0]
         username = str(entry[f"{LDAP_ATTRIBUTE_FOR_USERNAME}"]).lower()
-        email = str(entry[f"{LDAP_ATTRIBUTE_FOR_MAIL}"])
-        if not email or email == "" or email == "[]":
-            raise HTTPException(400, f"User {form_data.user} does not have email.")
-        else:
+        email = entry[f"{LDAP_ATTRIBUTE_FOR_MAIL}"]
+        if not email:
+            raise HTTPException(400, "User does not have a valid email address.")
+        elif isinstance(email, str):
             email = email.lower()
+        elif isinstance(email, list):
+            email = email[0].lower()
 
         cn = str(entry["cn"])
         user_dn = entry.entry_dn
@@ -248,7 +250,7 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                 authentication="SIMPLE",
             )
             if not connection_user.bind():
-                raise HTTPException(400, f"Authentication failed for {form_data.user}")
+                raise HTTPException(400, "Authentication failed.")
 
             user = Users.get_user_by_email(email)
             if not user:
@@ -276,7 +278,10 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                 except HTTPException:
                     raise
                 except Exception as err:
-                    raise HTTPException(500, detail=ERROR_MESSAGES.DEFAULT(err))
+                    log.error(f"LDAP user creation error: {str(err)}")
+                    raise HTTPException(
+                        500, detail="Internal error occurred during LDAP user creation."
+                    )
 
             user = Auths.authenticate_user_by_trusted_header(email)
 
@@ -312,12 +317,10 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
             else:
                 raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
         else:
-            raise HTTPException(
-                400,
-                f"User {form_data.user} does not match the record. Search result: {str(entry[f'{LDAP_ATTRIBUTE_FOR_USERNAME}'])}",
-            )
+            raise HTTPException(400, "User record mismatch.")
     except Exception as e:
-        raise HTTPException(400, detail=str(e))
+        log.error(f"LDAP authentication error: {str(e)}")
+        raise HTTPException(400, detail="LDAP authentication failed.")
 
 
 ############################
@@ -453,6 +456,13 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
             # Disable signup after the first user is created
             request.app.state.config.ENABLE_SIGNUP = False
 
+        # The password passed to bcrypt must be 72 bytes or fewer. If it is longer, it will be truncated before hashing.
+        if len(form_data.password.encode("utf-8")) > 72:
+            raise HTTPException(
+                status.HTTP_400_BAD_REQUEST,
+                detail=ERROR_MESSAGES.PASSWORD_TOO_LONG,
+            )
+
         hashed = get_password_hash(form_data.password)
         user = Auths.insert_new_auth(
             form_data.email.lower(),
@@ -519,7 +529,8 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
         else:
             raise HTTPException(500, detail=ERROR_MESSAGES.CREATE_USER_ERROR)
     except Exception as err:
-        raise HTTPException(500, detail=ERROR_MESSAGES.DEFAULT(err))
+        log.error(f"Signup error: {str(err)}")
+        raise HTTPException(500, detail="An internal error occurred during signup.")
 
 
 @router.get("/signout")
@@ -547,7 +558,11 @@ async def signout(request: Request, response: Response):
                                 detail="Failed to fetch OpenID configuration",
                             )
             except Exception as e:
-                raise HTTPException(status_code=500, detail=str(e))
+                log.error(f"OpenID signout error: {str(e)}")
+                raise HTTPException(
+                    status_code=500,
+                    detail="Failed to sign out from the OpenID provider.",
+                )
 
     return {"status": True}
 
@@ -591,7 +606,10 @@ async def add_user(form_data: AddUserForm, user=Depends(get_admin_user)):
         else:
             raise HTTPException(500, detail=ERROR_MESSAGES.CREATE_USER_ERROR)
     except Exception as err:
-        raise HTTPException(500, detail=ERROR_MESSAGES.DEFAULT(err))
+        log.error(f"Add user error: {str(err)}")
+        raise HTTPException(
+            500, detail="An internal error occurred while adding the user."
+        )
 
 
 ############################
@@ -639,11 +657,12 @@ async def get_admin_config(request: Request, user=Depends(get_admin_user)):
         "ENABLE_API_KEY": request.app.state.config.ENABLE_API_KEY,
         "ENABLE_API_KEY_ENDPOINT_RESTRICTIONS": request.app.state.config.ENABLE_API_KEY_ENDPOINT_RESTRICTIONS,
         "API_KEY_ALLOWED_ENDPOINTS": request.app.state.config.API_KEY_ALLOWED_ENDPOINTS,
-        "ENABLE_CHANNELS": request.app.state.config.ENABLE_CHANNELS,
         "DEFAULT_USER_ROLE": request.app.state.config.DEFAULT_USER_ROLE,
         "JWT_EXPIRES_IN": request.app.state.config.JWT_EXPIRES_IN,
         "ENABLE_COMMUNITY_SHARING": request.app.state.config.ENABLE_COMMUNITY_SHARING,
         "ENABLE_MESSAGE_RATING": request.app.state.config.ENABLE_MESSAGE_RATING,
+        "ENABLE_CHANNELS": request.app.state.config.ENABLE_CHANNELS,
+        "ENABLE_USER_WEBHOOKS": request.app.state.config.ENABLE_USER_WEBHOOKS,
     }
 
 
@@ -654,11 +673,12 @@ class AdminConfig(BaseModel):
     ENABLE_API_KEY: bool
     ENABLE_API_KEY_ENDPOINT_RESTRICTIONS: bool
     API_KEY_ALLOWED_ENDPOINTS: str
-    ENABLE_CHANNELS: bool
     DEFAULT_USER_ROLE: str
     JWT_EXPIRES_IN: str
     ENABLE_COMMUNITY_SHARING: bool
     ENABLE_MESSAGE_RATING: bool
+    ENABLE_CHANNELS: bool
+    ENABLE_USER_WEBHOOKS: bool
 
 
 @router.post("/admin/config")
@@ -693,6 +713,8 @@ async def update_admin_config(
     )
     request.app.state.config.ENABLE_MESSAGE_RATING = form_data.ENABLE_MESSAGE_RATING
 
+    request.app.state.config.ENABLE_USER_WEBHOOKS = form_data.ENABLE_USER_WEBHOOKS
+
     return {
         "SHOW_ADMIN_DETAILS": request.app.state.config.SHOW_ADMIN_DETAILS,
         "WEBUI_URL": request.app.state.config.WEBUI_URL,
@@ -705,6 +727,7 @@ async def update_admin_config(
         "JWT_EXPIRES_IN": request.app.state.config.JWT_EXPIRES_IN,
         "ENABLE_COMMUNITY_SHARING": request.app.state.config.ENABLE_COMMUNITY_SHARING,
         "ENABLE_MESSAGE_RATING": request.app.state.config.ENABLE_MESSAGE_RATING,
+        "ENABLE_USER_WEBHOOKS": request.app.state.config.ENABLE_USER_WEBHOOKS,
     }
 
 
@@ -759,11 +782,6 @@ async def update_ldap_server(
         if not value:
             raise HTTPException(400, detail=f"Required field {key} is empty")
 
-    if form_data.use_tls and not form_data.certificate_path:
-        raise HTTPException(
-            400, detail="TLS is enabled but certificate file path is missing"
-        )
-
     request.app.state.config.LDAP_SERVER_LABEL = form_data.label
     request.app.state.config.LDAP_SERVER_HOST = form_data.host
     request.app.state.config.LDAP_SERVER_PORT = form_data.port

backend/open_webui/routers/chats.py

@@ -2,6 +2,8 @@ import json
 import logging
 from typing import Optional
 
+
+from open_webui.socket.main import get_event_emitter
 from open_webui.models.chats import (
     ChatForm,
     ChatImportForm,
@@ -372,6 +374,107 @@ async def update_chat_by_id(
         )
 
 
+############################
+# UpdateChatMessageById
+############################
+class MessageForm(BaseModel):
+    content: str
+
+
+@router.post("/{id}/messages/{message_id}", response_model=Optional[ChatResponse])
+async def update_chat_message_by_id(
+    id: str, message_id: str, form_data: MessageForm, user=Depends(get_verified_user)
+):
+    chat = Chats.get_chat_by_id(id)
+
+    if not chat:
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
+    if chat.user_id != user.id and user.role != "admin":
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
+    chat = Chats.upsert_message_to_chat_by_id_and_message_id(
+        id,
+        message_id,
+        {
+            "content": form_data.content,
+        },
+    )
+
+    event_emitter = get_event_emitter(
+        {
+            "user_id": user.id,
+            "chat_id": id,
+            "message_id": message_id,
+        },
+        False,
+    )
+
+    if event_emitter:
+        await event_emitter(
+            {
+                "type": "chat:message",
+                "data": {
+                    "chat_id": id,
+                    "message_id": message_id,
+                    "content": form_data.content,
+                },
+            }
+        )
+
+    return ChatResponse(**chat.model_dump())
+
+
+############################
+# SendChatMessageEventById
+############################
+class EventForm(BaseModel):
+    type: str
+    data: dict
+
+
+@router.post("/{id}/messages/{message_id}/event", response_model=Optional[bool])
+async def send_chat_message_event_by_id(
+    id: str, message_id: str, form_data: EventForm, user=Depends(get_verified_user)
+):
+    chat = Chats.get_chat_by_id(id)
+
+    if not chat:
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
+    if chat.user_id != user.id and user.role != "admin":
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
+    event_emitter = get_event_emitter(
+        {
+            "user_id": user.id,
+            "chat_id": id,
+            "message_id": message_id,
+        }
+    )
+
+    try:
+        if event_emitter:
+            await event_emitter(form_data.model_dump())
+        else:
+            return False
+        return True
+    except:
+        return False
+
+
 ############################
 # DeleteChatById
 ############################
@@ -476,7 +579,12 @@ async def clone_chat_by_id(
 
 @router.post("/{id}/clone/shared", response_model=Optional[ChatResponse])
 async def clone_shared_chat_by_id(id: str, user=Depends(get_verified_user)):
-    chat = Chats.get_chat_by_share_id(id)
+
+    if user.role == "admin":
+        chat = Chats.get_chat_by_id(id)
+    else:
+        chat = Chats.get_chat_by_share_id(id)
+
     if chat:
         updated_chat = {
             **chat.chat,

backend/open_webui/routers/configs.py

@@ -1,5 +1,5 @@
-from fastapi import APIRouter, Depends, Request
-from pydantic import BaseModel
+from fastapi import APIRouter, Depends, Request, HTTPException
+from pydantic import BaseModel, ConfigDict
 
 from typing import Optional
 
@@ -7,6 +7,8 @@ from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.config import get_config, save_config
 from open_webui.config import BannerModel
 
+from open_webui.utils.tools import get_tool_server_data, get_tool_servers_data
+
 
 router = APIRouter()
 
@@ -66,6 +68,75 @@ async def set_direct_connections_config(
     }
 
 
+############################
+# ToolServers Config
+############################
+
+
+class ToolServerConnection(BaseModel):
+    url: str
+    path: str
+    auth_type: Optional[str]
+    key: Optional[str]
+    config: Optional[dict]
+
+    model_config = ConfigDict(extra="allow")
+
+
+class ToolServersConfigForm(BaseModel):
+    TOOL_SERVER_CONNECTIONS: list[ToolServerConnection]
+
+
+@router.get("/tool_servers", response_model=ToolServersConfigForm)
+async def get_tool_servers_config(request: Request, user=Depends(get_admin_user)):
+    return {
+        "TOOL_SERVER_CONNECTIONS": request.app.state.config.TOOL_SERVER_CONNECTIONS,
+    }
+
+
+@router.post("/tool_servers", response_model=ToolServersConfigForm)
+async def set_tool_servers_config(
+    request: Request,
+    form_data: ToolServersConfigForm,
+    user=Depends(get_admin_user),
+):
+    request.app.state.config.TOOL_SERVER_CONNECTIONS = [
+        connection.model_dump() for connection in form_data.TOOL_SERVER_CONNECTIONS
+    ]
+
+    request.app.state.TOOL_SERVERS = await get_tool_servers_data(
+        request.app.state.config.TOOL_SERVER_CONNECTIONS
+    )
+
+    return {
+        "TOOL_SERVER_CONNECTIONS": request.app.state.config.TOOL_SERVER_CONNECTIONS,
+    }
+
+
+@router.post("/tool_servers/verify")
+async def verify_tool_servers_config(
+    request: Request, form_data: ToolServerConnection, user=Depends(get_admin_user)
+):
+    """
+    Verify the connection to the tool server.
+    """
+    try:
+
+        token = None
+        if form_data.auth_type == "bearer":
+            token = form_data.key
+        elif form_data.auth_type == "session":
+            token = request.state.token.credentials
+
+        url = f"{form_data.url}/{form_data.path}"
+        return await get_tool_server_data(token, url)
+    except Exception as e:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Failed to connect to the tool server: {str(e)}",
+        )
+
+
 ############################
 # CodeInterpreterConfig
 ############################

backend/open_webui/routers/evaluations.py

@@ -56,8 +56,19 @@ async def update_config(
     }
 
 
+class FeedbackUserReponse(BaseModel):
+    id: str
+    name: str
+    email: str
+    role: str = "pending"
+
+    last_active_at: int  # timestamp in epoch
+    updated_at: int  # timestamp in epoch
+    created_at: int  # timestamp in epoch
+
+
 class FeedbackUserResponse(FeedbackResponse):
-    user: Optional[UserModel] = None
+    user: Optional[FeedbackUserReponse] = None
 
 
 @router.get("/feedbacks/all", response_model=list[FeedbackUserResponse])
@@ -65,7 +76,10 @@ async def get_all_feedbacks(user=Depends(get_admin_user)):
     feedbacks = Feedbacks.get_all_feedbacks()
     return [
         FeedbackUserResponse(
-            **feedback.model_dump(), user=Users.get_user_by_id(feedback.user_id)
+            **feedback.model_dump(),
+            user=FeedbackUserReponse(
+                **Users.get_user_by_id(feedback.user_id).model_dump()
+            ),
         )
         for feedback in feedbacks
     ]

backend/open_webui/routers/files.py

@@ -1,11 +1,21 @@
 import logging
 import os
 import uuid
+from fnmatch import fnmatch
 from pathlib import Path
 from typing import Optional
 from urllib.parse import quote
 
-from fastapi import APIRouter, Depends, File, HTTPException, Request, UploadFile, status
+from fastapi import (
+    APIRouter,
+    Depends,
+    File,
+    HTTPException,
+    Request,
+    UploadFile,
+    status,
+    Query,
+)
 from fastapi.responses import FileResponse, StreamingResponse
 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import SRC_LOG_LEVELS
@@ -15,6 +25,9 @@ from open_webui.models.files import (
     FileModelResponse,
     Files,
 )
+from open_webui.models.knowledge import Knowledges
+
+from open_webui.routers.knowledge import get_knowledge, get_knowledge_list
 from open_webui.routers.retrieval import ProcessFileForm, process_file
 from open_webui.routers.audio import transcribe
 from open_webui.storage.provider import Storage
@@ -27,6 +40,39 @@ log.setLevel(SRC_LOG_LEVELS["MODELS"])
 
 router = APIRouter()
 
+
+############################
+# Check if the current user has access to a file through any knowledge bases the user may be in.
+############################
+
+
+def has_access_to_file(
+    file_id: Optional[str], access_type: str, user=Depends(get_verified_user)
+) -> bool:
+    file = Files.get_file_by_id(file_id)
+    log.debug(f"Checking if user has {access_type} access to file")
+
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    has_access = False
+    knowledge_base_id = file.meta.get("collection_name") if file.meta else None
+
+    if knowledge_base_id:
+        knowledge_bases = Knowledges.get_knowledge_bases_by_user_id(
+            user.id, access_type
+        )
+        for knowledge_base in knowledge_bases:
+            if knowledge_base.id == knowledge_base_id:
+                has_access = True
+                break
+
+    return has_access
+
+
 ############################
 # Upload File
 ############################
@@ -38,6 +84,7 @@ def upload_file(
     file: UploadFile = File(...),
     user=Depends(get_verified_user),
     file_metadata: dict = {},
+    process: bool = Query(True),
 ):
     log.info(f"file.content_type: {file.content_type}")
     try:
@@ -66,34 +113,35 @@ def upload_file(
                 }
             ),
         )
+        if process:
+            try:
+                if file.content_type in [
+                    "audio/mpeg",
+                    "audio/wav",
+                    "audio/ogg",
+                    "audio/x-m4a",
+                ]:
+                    file_path = Storage.get_file(file_path)
+                    result = transcribe(request, file_path)
 
-        try:
-            if file.content_type in [
-                "audio/mpeg",
-                "audio/wav",
-                "audio/ogg",
-                "audio/x-m4a",
-            ]:
-                file_path = Storage.get_file(file_path)
-                result = transcribe(request, file_path)
-                process_file(
-                    request,
-                    ProcessFileForm(file_id=id, content=result.get("text", "")),
-                    user=user,
+                    process_file(
+                        request,
+                        ProcessFileForm(file_id=id, content=result.get("text", "")),
+                        user=user,
+                    )
+                elif file.content_type not in ["image/png", "image/jpeg", "image/gif"]:
+                    process_file(request, ProcessFileForm(file_id=id), user=user)
+
+                file_item = Files.get_file_by_id(id=id)
+            except Exception as e:
+                log.exception(e)
+                log.error(f"Error processing file: {file_item.id}")
+                file_item = FileModelResponse(
+                    **{
+                        **file_item.model_dump(),
+                        "error": str(e.detail) if hasattr(e, "detail") else str(e),
+                    }
                 )
-            else:
-                process_file(request, ProcessFileForm(file_id=id), user=user)
-
-            file_item = Files.get_file_by_id(id=id)
-        except Exception as e:
-            log.exception(e)
-            log.error(f"Error processing file: {file_item.id}")
-            file_item = FileModelResponse(
-                **{
-                    **file_item.model_dump(),
-                    "error": str(e.detail) if hasattr(e, "detail") else str(e),
-                }
-            )
 
         if file_item:
             return file_item
@@ -117,14 +165,60 @@ def upload_file(
 
 
 @router.get("/", response_model=list[FileModelResponse])
-async def list_files(user=Depends(get_verified_user)):
+async def list_files(user=Depends(get_verified_user), content: bool = Query(True)):
     if user.role == "admin":
         files = Files.get_files()
     else:
         files = Files.get_files_by_user_id(user.id)
+
+    if not content:
+        for file in files:
+            del file.data["content"]
+
     return files
 
 
+############################
+# Search Files
+############################
+
+
+@router.get("/search", response_model=list[FileModelResponse])
+async def search_files(
+    filename: str = Query(
+        ...,
+        description="Filename pattern to search for. Supports wildcards such as '*.txt'",
+    ),
+    content: bool = Query(True),
+    user=Depends(get_verified_user),
+):
+    """
+    Search for files by filename with support for wildcard patterns.
+    """
+    # Get files according to user role
+    if user.role == "admin":
+        files = Files.get_files()
+    else:
+        files = Files.get_files_by_user_id(user.id)
+
+    # Get matching files
+    matching_files = [
+        file for file in files if fnmatch(file.filename.lower(), filename.lower())
+    ]
+
+    if not matching_files:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="No files found matching the pattern.",
+        )
+
+    if not content:
+        for file in matching_files:
+            del file.data["content"]
+
+    return matching_files
+
+
 ############################
 # Delete All Files
 ############################
@@ -160,7 +254,17 @@ async def delete_all_files(user=Depends(get_admin_user)):
 async def get_file_by_id(id: str, user=Depends(get_verified_user)):
     file = Files.get_file_by_id(id)
 
-    if file and (file.user_id == user.id or user.role == "admin"):
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "read", user)
+    ):
         return file
     else:
         raise HTTPException(
@@ -178,7 +282,17 @@ async def get_file_by_id(id: str, user=Depends(get_verified_user)):
 async def get_file_data_content_by_id(id: str, user=Depends(get_verified_user)):
     file = Files.get_file_by_id(id)
 
-    if file and (file.user_id == user.id or user.role == "admin"):
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "read", user)
+    ):
         return {"content": file.data.get("content", "")}
     else:
         raise HTTPException(
@@ -202,7 +316,17 @@ async def update_file_data_content_by_id(
 ):
     file = Files.get_file_by_id(id)
 
-    if file and (file.user_id == user.id or user.role == "admin"):
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "write", user)
+    ):
         try:
             process_file(
                 request,
@@ -228,9 +352,22 @@ async def update_file_data_content_by_id(
 
 
 @router.get("/{id}/content")
-async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
+async def get_file_content_by_id(
+    id: str, user=Depends(get_verified_user), attachment: bool = Query(False)
+):
     file = Files.get_file_by_id(id)
-    if file and (file.user_id == user.id or user.role == "admin"):
+
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "read", user)
+    ):
         try:
             file_path = Storage.get_file(file.path)
             file_path = Path(file_path)
@@ -246,17 +383,22 @@ async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
                 encoded_filename = quote(filename)
                 headers = {}
 
-                if content_type == "application/pdf" or filename.lower().endswith(
-                    ".pdf"
-                ):
-                    headers["Content-Disposition"] = (
-                        f"inline; filename*=UTF-8''{encoded_filename}"
-                    )
-                    content_type = "application/pdf"
-                elif content_type != "text/plain":
+                if attachment:
                     headers["Content-Disposition"] = (
                         f"attachment; filename*=UTF-8''{encoded_filename}"
                     )
+                else:
+                    if content_type == "application/pdf" or filename.lower().endswith(
+                        ".pdf"
+                    ):
+                        headers["Content-Disposition"] = (
+                            f"inline; filename*=UTF-8''{encoded_filename}"
+                        )
+                        content_type = "application/pdf"
+                    elif content_type != "text/plain":
+                        headers["Content-Disposition"] = (
+                            f"attachment; filename*=UTF-8''{encoded_filename}"
+                        )
 
                 return FileResponse(file_path, headers=headers, media_type=content_type)
 
@@ -282,7 +424,18 @@ async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
 @router.get("/{id}/content/html")
 async def get_html_file_content_by_id(id: str, user=Depends(get_verified_user)):
     file = Files.get_file_by_id(id)
-    if file and (file.user_id == user.id or user.role == "admin"):
+
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "read", user)
+    ):
         try:
             file_path = Storage.get_file(file.path)
             file_path = Path(file_path)
@@ -314,7 +467,17 @@ async def get_html_file_content_by_id(id: str, user=Depends(get_verified_user)):
 async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
     file = Files.get_file_by_id(id)
 
-    if file and (file.user_id == user.id or user.role == "admin"):
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "read", user)
+    ):
         file_path = file.path
 
         # Handle Unicode filenames
@@ -365,7 +528,18 @@ async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
 @router.delete("/{id}")
 async def delete_file_by_id(id: str, user=Depends(get_verified_user)):
     file = Files.get_file_by_id(id)
-    if file and (file.user_id == user.id or user.role == "admin"):
+
+    if not file:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=ERROR_MESSAGES.NOT_FOUND,
+        )
+
+    if (
+        file.user_id == user.id
+        or user.role == "admin"
+        or has_access_to_file(id, "write", user)
+    ):
         # We should add Chroma cleanup here
 
         result = Files.delete_file_by_id(id)

backend/open_webui/routers/folders.py

@@ -20,11 +20,13 @@ from open_webui.env import SRC_LOG_LEVELS
 from open_webui.constants import ERROR_MESSAGES
 
 
-from fastapi import APIRouter, Depends, File, HTTPException, UploadFile, status
+from fastapi import APIRouter, Depends, File, HTTPException, UploadFile, status, Request
 from fastapi.responses import FileResponse, StreamingResponse
 
 
 from open_webui.utils.auth import get_admin_user, get_verified_user
+from open_webui.utils.access_control import has_permission
+
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MODELS"])
@@ -228,7 +230,19 @@ async def update_folder_is_expanded_by_id(
 
 
 @router.delete("/{id}")
-async def delete_folder_by_id(id: str, user=Depends(get_verified_user)):
+async def delete_folder_by_id(
+    request: Request, id: str, user=Depends(get_verified_user)
+):
+    chat_delete_permission = has_permission(
+        user.id, "chat.delete", request.app.state.config.USER_PERMISSIONS
+    )
+
+    if user.role != "admin" and not chat_delete_permission:
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
     folder = Folders.get_folder_by_id_and_user_id(id, user.id)
     if folder:
         try:

backend/open_webui/routers/images.py

@@ -517,10 +517,8 @@ async def image_generations(
             images = []
 
             for image in res["data"]:
-                if "url" in image:
-                    image_data, content_type = load_url_image_data(
-                        image["url"], headers
-                    )
+                if image_url := image.get("url", None):
+                    image_data, content_type = load_url_image_data(image_url, headers)
                 else:
                     image_data, content_type = load_b64_image_data(image["b64_json"])
 

backend/open_webui/routers/knowledge.py

@@ -161,6 +161,74 @@ async def create_new_knowledge(
         )
 
 
+############################
+# ReindexKnowledgeFiles
+############################
+
+
+@router.post("/reindex", response_model=bool)
+async def reindex_knowledge_files(request: Request, user=Depends(get_verified_user)):
+    if user.role != "admin":
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail=ERROR_MESSAGES.UNAUTHORIZED,
+        )
+
+    knowledge_bases = Knowledges.get_knowledge_bases()
+
+    log.info(f"Starting reindexing for {len(knowledge_bases)} knowledge bases")
+
+    for knowledge_base in knowledge_bases:
+        try:
+            files = Files.get_files_by_ids(knowledge_base.data.get("file_ids", []))
+
+            try:
+                if VECTOR_DB_CLIENT.has_collection(collection_name=knowledge_base.id):
+                    VECTOR_DB_CLIENT.delete_collection(
+                        collection_name=knowledge_base.id
+                    )
+            except Exception as e:
+                log.error(f"Error deleting collection {knowledge_base.id}: {str(e)}")
+                raise HTTPException(
+                    status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                    detail=f"Error deleting vector DB collection",
+                )
+
+            failed_files = []
+            for file in files:
+                try:
+                    process_file(
+                        request,
+                        ProcessFileForm(
+                            file_id=file.id, collection_name=knowledge_base.id
+                        ),
+                        user=user,
+                    )
+                except Exception as e:
+                    log.error(
+                        f"Error processing file {file.filename} (ID: {file.id}): {str(e)}"
+                    )
+                    failed_files.append({"file_id": file.id, "error": str(e)})
+                    continue
+
+        except Exception as e:
+            log.error(f"Error processing knowledge base {knowledge_base.id}: {str(e)}")
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail=f"Error processing knowledge base",
+            )
+
+        if failed_files:
+            log.warning(
+                f"Failed to process {len(failed_files)} files in knowledge base {knowledge_base.id}"
+            )
+            for failed in failed_files:
+                log.warning(f"File ID: {failed['file_id']}, Error: {failed['error']}")
+
+    log.info("Reindexing completed successfully")
+    return True
+
+
 ############################
 # GetKnowledgeById
 ############################
@@ -437,14 +505,24 @@ def remove_file_from_knowledge_by_id(
         )
 
     # Remove content from the vector database
-    VECTOR_DB_CLIENT.delete(
-        collection_name=knowledge.id, filter={"file_id": form_data.file_id}
-    )
+    try:
+        VECTOR_DB_CLIENT.delete(
+            collection_name=knowledge.id, filter={"file_id": form_data.file_id}
+        )
+    except Exception as e:
+        log.debug("This was most likely caused by bypassing embedding processing")
+        log.debug(e)
+        pass
 
-    # Remove the file's collection from vector database
-    file_collection = f"file-{form_data.file_id}"
-    if VECTOR_DB_CLIENT.has_collection(collection_name=file_collection):
-        VECTOR_DB_CLIENT.delete_collection(collection_name=file_collection)
+    try:
+        # Remove the file's collection from vector database
+        file_collection = f"file-{form_data.file_id}"
+        if VECTOR_DB_CLIENT.has_collection(collection_name=file_collection):
+            VECTOR_DB_CLIENT.delete_collection(collection_name=file_collection)
+    except Exception as e:
+        log.debug("This was most likely caused by bypassing embedding processing")
+        log.debug(e)
+        pass
 
     # Delete file from database
     Files.delete_file_by_id(form_data.file_id)

backend/open_webui/routers/memories.py

@@ -57,7 +57,9 @@ async def add_memory(
             {
                 "id": memory.id,
                 "text": memory.content,
-                "vector": request.app.state.EMBEDDING_FUNCTION(memory.content, user),
+                "vector": request.app.state.EMBEDDING_FUNCTION(
+                    memory.content, user=user
+                ),
                 "metadata": {"created_at": memory.created_at},
             }
         ],
@@ -82,7 +84,7 @@ async def query_memory(
 ):
     results = VECTOR_DB_CLIENT.search(
         collection_name=f"user-memory-{user.id}",
-        vectors=[request.app.state.EMBEDDING_FUNCTION(form_data.content, user)],
+        vectors=[request.app.state.EMBEDDING_FUNCTION(form_data.content, user=user)],
         limit=form_data.k,
     )
 
@@ -105,7 +107,9 @@ async def reset_memory_from_vector_db(
             {
                 "id": memory.id,
                 "text": memory.content,
-                "vector": request.app.state.EMBEDDING_FUNCTION(memory.content, user),
+                "vector": request.app.state.EMBEDDING_FUNCTION(
+                    memory.content, user=user
+                ),
                 "metadata": {
                     "created_at": memory.created_at,
                     "updated_at": memory.updated_at,
@@ -149,7 +153,9 @@ async def update_memory_by_id(
     form_data: MemoryUpdateModel,
     user=Depends(get_verified_user),
 ):
-    memory = Memories.update_memory_by_id(memory_id, form_data.content)
+    memory = Memories.update_memory_by_id_and_user_id(
+        memory_id, user.id, form_data.content
+    )
     if memory is None:
         raise HTTPException(status_code=404, detail="Memory not found")
 
@@ -161,7 +167,7 @@ async def update_memory_by_id(
                     "id": memory.id,
                     "text": memory.content,
                     "vector": request.app.state.EMBEDDING_FUNCTION(
-                        memory.content, user
+                        memory.content, user=user
                     ),
                     "metadata": {
                         "created_at": memory.created_at,

backend/open_webui/routers/ollama.py

@@ -295,7 +295,7 @@ async def update_config(
     }
 
 
-@cached(ttl=3)
+@cached(ttl=1)
 async def get_all_models(request: Request, user: UserModel = None):
     log.info("get_all_models()")
     if request.app.state.config.ENABLE_OLLAMA_API:
@@ -336,6 +336,7 @@ async def get_all_models(request: Request, user: UserModel = None):
                 )
 
                 prefix_id = api_config.get("prefix_id", None)
+                tags = api_config.get("tags", [])
                 model_ids = api_config.get("model_ids", [])
 
                 if len(model_ids) != 0 and "models" in response:
@@ -350,6 +351,10 @@ async def get_all_models(request: Request, user: UserModel = None):
                     for model in response.get("models", []):
                         model["model"] = f"{prefix_id}.{model['model']}"
 
+                if tags:
+                    for model in response.get("models", []):
+                        model["tags"] = tags
+
         def merge_models_lists(model_lists):
             merged_models = {}
 
@@ -460,18 +465,27 @@ async def get_ollama_versions(request: Request, url_idx: Optional[int] = None):
     if request.app.state.config.ENABLE_OLLAMA_API:
         if url_idx is None:
             # returns lowest version
-            request_tasks = [
-                send_get_request(
-                    f"{url}/api/version",
+            request_tasks = []
+
+            for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS):
+                api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
+                    str(idx),
                     request.app.state.config.OLLAMA_API_CONFIGS.get(
-                        str(idx),
-                        request.app.state.config.OLLAMA_API_CONFIGS.get(
-                            url, {}
-                        ),  # Legacy support
-                    ).get("key", None),
+                        url, {}
+                    ),  # Legacy support
                 )
-                for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS)
-            ]
+
+                enable = api_config.get("enable", True)
+                key = api_config.get("key", None)
+
+                if enable:
+                    request_tasks.append(
+                        send_get_request(
+                            f"{url}/api/version",
+                            key,
+                        )
+                    )
+
             responses = await asyncio.gather(*request_tasks)
             responses = list(filter(lambda x: x is not None, responses))
 
@@ -1164,7 +1178,7 @@ 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),
@@ -1183,7 +1197,7 @@ class OpenAIChatMessageContent(BaseModel):
 
 class OpenAIChatMessage(BaseModel):
     role: str
-    content: Union[str, list[OpenAIChatMessageContent]]
+    content: Union[Optional[str], list[OpenAIChatMessageContent]]
 
     model_config = ConfigDict(extra="allow")
 

backend/open_webui/routers/openai.py

@@ -36,6 +36,9 @@ from open_webui.utils.payload import (
     apply_model_params_to_body_openai,
     apply_model_system_prompt_to_body,
 )
+from open_webui.utils.misc import (
+    convert_logit_bias_input_to_json,
+)
 
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.access_control import has_access
@@ -350,6 +353,7 @@ async def get_all_models_responses(request: Request, user: UserModel) -> list:
             )
 
             prefix_id = api_config.get("prefix_id", None)
+            tags = api_config.get("tags", [])
 
             if prefix_id:
                 for model in (
@@ -357,6 +361,12 @@ async def get_all_models_responses(request: Request, user: UserModel) -> list:
                 ):
                     model["id"] = f"{prefix_id}.{model['id']}"
 
+            if tags:
+                for model in (
+                    response if isinstance(response, list) else response.get("data", [])
+                ):
+                    model["tags"] = tags
+
     log.debug(f"get_all_models:responses() {responses}")
     return responses
 
@@ -374,7 +384,7 @@ async def get_filtered_models(models, user):
     return filtered_models
 
 
-@cached(ttl=3)
+@cached(ttl=1)
 async def get_all_models(request: Request, user: UserModel) -> dict[str, list]:
     log.info("get_all_models()")
 
@@ -396,6 +406,7 @@ async def get_all_models(request: Request, user: UserModel) -> dict[str, list]:
 
         for idx, models in enumerate(model_lists):
             if models is not None and "error" not in models:
+
                 merged_list.extend(
                     [
                         {
@@ -406,18 +417,21 @@ async def get_all_models(request: Request, user: UserModel) -> dict[str, list]:
                             "urlIdx": idx,
                         }
                         for model in models
-                        if "api.openai.com"
-                        not in request.app.state.config.OPENAI_API_BASE_URLS[idx]
-                        or not any(
-                            name in model["id"]
-                            for name in [
-                                "babbage",
-                                "dall-e",
-                                "davinci",
-                                "embedding",
-                                "tts",
-                                "whisper",
-                            ]
+                        if (model.get("id") or model.get("name"))
+                        and (
+                            "api.openai.com"
+                            not in request.app.state.config.OPENAI_API_BASE_URLS[idx]
+                            or not any(
+                                name in model["id"]
+                                for name in [
+                                    "babbage",
+                                    "dall-e",
+                                    "davinci",
+                                    "embedding",
+                                    "tts",
+                                    "whisper",
+                                ]
+                            )
                         )
                     ]
                 )
@@ -666,6 +680,11 @@ async def generate_chat_completion(
         del payload["max_tokens"]
 
     # Convert the modified body back to JSON
+    if "logit_bias" in payload:
+        payload["logit_bias"] = json.loads(
+            convert_logit_bias_input_to_json(payload["logit_bias"])
+        )
+
     payload = json.dumps(payload)
 
     r = None

backend/open_webui/routers/pipelines.py

@@ -90,8 +90,8 @@ async def process_pipeline_inlet_filter(request, payload, user, models):
                     headers=headers,
                     json=request_data,
                 ) as response:
-                    response.raise_for_status()
                     payload = await response.json()
+                    response.raise_for_status()
             except aiohttp.ClientResponseError as e:
                 res = (
                     await response.json()
@@ -139,8 +139,8 @@ async def process_pipeline_outlet_filter(request, payload, user, models):
                     headers=headers,
                     json=request_data,
                 ) as response:
-                    response.raise_for_status()
                     payload = await response.json()
+                    response.raise_for_status()
             except aiohttp.ClientResponseError as e:
                 try:
                     res = (

backend/open_webui/routers/tasks.py

@@ -653,17 +653,6 @@ async def generate_moa_response(
             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 MOA model {task_model_id} for user {user.email} ")
-
     template = DEFAULT_MOA_GENERATION_PROMPT_TEMPLATE
 
     content = moa_response_generation_template(
@@ -673,7 +662,7 @@ async def generate_moa_response(
     )
 
     payload = {
-        "model": task_model_id,
+        "model": model_id,
         "messages": [{"role": "user", "content": content}],
         "stream": form_data.get("stream", False),
         "metadata": {

backend/open_webui/routers/tools.py

@@ -1,6 +1,7 @@
 import logging
 from pathlib import Path
 from typing import Optional
+import time
 
 from open_webui.models.tools import (
     ToolForm,
@@ -9,15 +10,17 @@ from open_webui.models.tools import (
     ToolUserResponse,
     Tools,
 )
-from open_webui.utils.plugin import load_tools_module_by_id, replace_imports
+from open_webui.utils.plugin import load_tool_module_by_id, replace_imports
 from open_webui.config import CACHE_DIR
 from open_webui.constants import ERROR_MESSAGES
 from fastapi import APIRouter, Depends, HTTPException, Request, status
-from open_webui.utils.tools import get_tools_specs
+from open_webui.utils.tools import get_tool_specs
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.access_control import has_access, has_permission
 from open_webui.env import SRC_LOG_LEVELS
 
+from open_webui.utils.tools import get_tool_servers_data
+
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
@@ -30,11 +33,51 @@ router = APIRouter()
 
 
 @router.get("/", response_model=list[ToolUserResponse])
-async def get_tools(user=Depends(get_verified_user)):
-    if user.role == "admin":
-        tools = Tools.get_tools()
-    else:
-        tools = Tools.get_tools_by_user_id(user.id, "read")
+async def get_tools(request: Request, user=Depends(get_verified_user)):
+
+    if not request.app.state.TOOL_SERVERS:
+        # If the tool servers are not set, we need to set them
+        # This is done only once when the server starts
+        # This is done to avoid loading the tool servers every time
+
+        request.app.state.TOOL_SERVERS = await get_tool_servers_data(
+            request.app.state.config.TOOL_SERVER_CONNECTIONS
+        )
+
+    tools = Tools.get_tools()
+    for server in request.app.state.TOOL_SERVERS:
+        tools.append(
+            ToolUserResponse(
+                **{
+                    "id": f"server:{server['idx']}",
+                    "user_id": f"server:{server['idx']}",
+                    "name": server["openapi"]
+                    .get("info", {})
+                    .get("title", "Tool Server"),
+                    "meta": {
+                        "description": server["openapi"]
+                        .get("info", {})
+                        .get("description", ""),
+                    },
+                    "access_control": request.app.state.config.TOOL_SERVER_CONNECTIONS[
+                        server["idx"]
+                    ]
+                    .get("config", {})
+                    .get("access_control", None),
+                    "updated_at": int(time.time()),
+                    "created_at": int(time.time()),
+                }
+            )
+        )
+
+    if user.role != "admin":
+        tools = [
+            tool
+            for tool in tools
+            if tool.user_id == user.id
+            or has_access(user.id, "read", tool.access_control)
+        ]
+
     return tools
 
 
@@ -94,15 +137,15 @@ async def create_new_tools(
     if tools is None:
         try:
             form_data.content = replace_imports(form_data.content)
-            tools_module, frontmatter = load_tools_module_by_id(
+            tool_module, frontmatter = load_tool_module_by_id(
                 form_data.id, content=form_data.content
             )
             form_data.meta.manifest = frontmatter
 
             TOOLS = request.app.state.TOOLS
-            TOOLS[form_data.id] = tools_module
+            TOOLS[form_data.id] = tool_module
 
-            specs = get_tools_specs(TOOLS[form_data.id])
+            specs = get_tool_specs(TOOLS[form_data.id])
             tools = Tools.insert_new_tool(user.id, form_data, specs)
 
             tool_cache_dir = CACHE_DIR / "tools" / form_data.id
@@ -183,15 +226,13 @@ async def update_tools_by_id(
 
     try:
         form_data.content = replace_imports(form_data.content)
-        tools_module, frontmatter = load_tools_module_by_id(
-            id, content=form_data.content
-        )
+        tool_module, frontmatter = load_tool_module_by_id(id, content=form_data.content)
         form_data.meta.manifest = frontmatter
 
         TOOLS = request.app.state.TOOLS
-        TOOLS[id] = tools_module
+        TOOLS[id] = tool_module
 
-        specs = get_tools_specs(TOOLS[id])
+        specs = get_tool_specs(TOOLS[id])
 
         updated = {
             **form_data.model_dump(exclude={"id"}),
@@ -289,7 +330,7 @@ async def get_tools_valves_spec_by_id(
         if id in request.app.state.TOOLS:
             tools_module = request.app.state.TOOLS[id]
         else:
-            tools_module, _ = load_tools_module_by_id(id)
+            tools_module, _ = load_tool_module_by_id(id)
             request.app.state.TOOLS[id] = tools_module
 
         if hasattr(tools_module, "Valves"):
@@ -332,7 +373,7 @@ async def update_tools_valves_by_id(
     if id in request.app.state.TOOLS:
         tools_module = request.app.state.TOOLS[id]
     else:
-        tools_module, _ = load_tools_module_by_id(id)
+        tools_module, _ = load_tool_module_by_id(id)
         request.app.state.TOOLS[id] = tools_module
 
     if not hasattr(tools_module, "Valves"):
@@ -388,7 +429,7 @@ async def get_tools_user_valves_spec_by_id(
         if id in request.app.state.TOOLS:
             tools_module = request.app.state.TOOLS[id]
         else:
-            tools_module, _ = load_tools_module_by_id(id)
+            tools_module, _ = load_tool_module_by_id(id)
             request.app.state.TOOLS[id] = tools_module
 
         if hasattr(tools_module, "UserValves"):
@@ -412,7 +453,7 @@ async def update_tools_user_valves_by_id(
         if id in request.app.state.TOOLS:
             tools_module = request.app.state.TOOLS[id]
         else:
-            tools_module, _ = load_tools_module_by_id(id)
+            tools_module, _ = load_tool_module_by_id(id)
             request.app.state.TOOLS[id] = tools_module
 
         if hasattr(tools_module, "UserValves"):

backend/open_webui/routers/users.py

@@ -2,6 +2,7 @@ import logging
 from typing import Optional
 
 from open_webui.models.auths import Auths
+from open_webui.models.groups import Groups
 from open_webui.models.chats import Chats
 from open_webui.models.users import (
     UserModel,
@@ -17,7 +18,10 @@ from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import SRC_LOG_LEVELS
 from fastapi import APIRouter, Depends, HTTPException, Request, status
 from pydantic import BaseModel
+
 from open_webui.utils.auth import get_admin_user, get_password_hash, get_verified_user
+from open_webui.utils.access_control import get_permissions
+
 
 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MODELS"])
@@ -45,7 +49,7 @@ async def get_users(
 
 @router.get("/groups")
 async def get_user_groups(user=Depends(get_verified_user)):
-    return Users.get_user_groups(user.id)
+    return Groups.get_groups_by_member_id(user.id)
 
 
 ############################
@@ -54,8 +58,12 @@ async def get_user_groups(user=Depends(get_verified_user)):
 
 
 @router.get("/permissions")
-async def get_user_permissisions(user=Depends(get_verified_user)):
-    return Users.get_user_groups(user.id)
+async def get_user_permissisions(request: Request, user=Depends(get_verified_user)):
+    user_permissions = get_permissions(
+        user.id, request.app.state.config.USER_PERMISSIONS
+    )
+
+    return user_permissions
 
 
 ############################
@@ -68,15 +76,25 @@ class WorkspacePermissions(BaseModel):
     tools: bool = False
 
 
+class SharingPermissions(BaseModel):
+    public_models: bool = True
+    public_knowledge: bool = True
+    public_prompts: bool = True
+    public_tools: bool = True
+
+
 class ChatPermissions(BaseModel):
     controls: bool = True
     file_upload: bool = True
     delete: bool = True
     edit: bool = True
+    multiple_models: bool = True
     temporary: bool = True
+    temporary_enforced: bool = False
 
 
 class FeaturesPermissions(BaseModel):
+    direct_tool_servers: bool = False
     web_search: bool = True
     image_generation: bool = True
     code_interpreter: bool = True
@@ -84,16 +102,20 @@ class FeaturesPermissions(BaseModel):
 
 class UserPermissions(BaseModel):
     workspace: WorkspacePermissions
+    sharing: SharingPermissions
     chat: ChatPermissions
     features: FeaturesPermissions
 
 
 @router.get("/default/permissions", response_model=UserPermissions)
-async def get_user_permissions(request: Request, user=Depends(get_admin_user)):
+async def get_default_user_permissions(request: Request, user=Depends(get_admin_user)):
     return {
         "workspace": WorkspacePermissions(
             **request.app.state.config.USER_PERMISSIONS.get("workspace", {})
         ),
+        "sharing": SharingPermissions(
+            **request.app.state.config.USER_PERMISSIONS.get("sharing", {})
+        ),
         "chat": ChatPermissions(
             **request.app.state.config.USER_PERMISSIONS.get("chat", {})
         ),
@@ -104,7 +126,7 @@ async def get_user_permissions(request: Request, user=Depends(get_admin_user)):
 
 
 @router.post("/default/permissions")
-async def update_user_permissions(
+async def update_default_user_permissions(
     request: Request, form_data: UserPermissions, user=Depends(get_admin_user)
 ):
     request.app.state.config.USER_PERMISSIONS = form_data.model_dump()

backend/open_webui/socket/main.py

@@ -3,16 +3,23 @@ import socketio
 import logging
 import sys
 import time
+from redis import asyncio as aioredis
 
 from open_webui.models.users import Users, UserNameResponse
 from open_webui.models.channels import Channels
 from open_webui.models.chats import Chats
+from open_webui.utils.redis import (
+    get_sentinels_from_env,
+    get_sentinel_url_from_env,
+)
 
 from open_webui.env import (
     ENABLE_WEBSOCKET_SUPPORT,
     WEBSOCKET_MANAGER,
     WEBSOCKET_REDIS_URL,
     WEBSOCKET_REDIS_LOCK_TIMEOUT,
+    WEBSOCKET_SENTINEL_PORT,
+    WEBSOCKET_SENTINEL_HOSTS,
 )
 from open_webui.utils.auth import decode_token
 from open_webui.socket.utils import RedisDict, RedisLock
@@ -29,7 +36,14 @@ log.setLevel(SRC_LOG_LEVELS["SOCKET"])
 
 
 if WEBSOCKET_MANAGER == "redis":
-    mgr = socketio.AsyncRedisManager(WEBSOCKET_REDIS_URL)
+    if WEBSOCKET_SENTINEL_HOSTS:
+        mgr = socketio.AsyncRedisManager(
+            get_sentinel_url_from_env(
+                WEBSOCKET_REDIS_URL, WEBSOCKET_SENTINEL_HOSTS, WEBSOCKET_SENTINEL_PORT
+            )
+        )
+    else:
+        mgr = socketio.AsyncRedisManager(WEBSOCKET_REDIS_URL)
     sio = socketio.AsyncServer(
         cors_allowed_origins=[],
         async_mode="asgi",
@@ -55,14 +69,30 @@ TIMEOUT_DURATION = 3
 
 if WEBSOCKET_MANAGER == "redis":
     log.debug("Using Redis to manage websockets.")
-    SESSION_POOL = RedisDict("open-webui:session_pool", redis_url=WEBSOCKET_REDIS_URL)
-    USER_POOL = RedisDict("open-webui:user_pool", redis_url=WEBSOCKET_REDIS_URL)
-    USAGE_POOL = RedisDict("open-webui:usage_pool", redis_url=WEBSOCKET_REDIS_URL)
+    redis_sentinels = get_sentinels_from_env(
+        WEBSOCKET_SENTINEL_HOSTS, WEBSOCKET_SENTINEL_PORT
+    )
+    SESSION_POOL = RedisDict(
+        "open-webui:session_pool",
+        redis_url=WEBSOCKET_REDIS_URL,
+        redis_sentinels=redis_sentinels,
+    )
+    USER_POOL = RedisDict(
+        "open-webui:user_pool",
+        redis_url=WEBSOCKET_REDIS_URL,
+        redis_sentinels=redis_sentinels,
+    )
+    USAGE_POOL = RedisDict(
+        "open-webui:usage_pool",
+        redis_url=WEBSOCKET_REDIS_URL,
+        redis_sentinels=redis_sentinels,
+    )
 
     clean_up_lock = RedisLock(
         redis_url=WEBSOCKET_REDIS_URL,
         lock_name="usage_cleanup_lock",
         timeout_secs=WEBSOCKET_REDIS_LOCK_TIMEOUT,
+        redis_sentinels=redis_sentinels,
     )
     aquire_func = clean_up_lock.aquire_lock
     renew_func = clean_up_lock.renew_lock
@@ -269,11 +299,19 @@ async def disconnect(sid):
         # print(f"Unknown session ID {sid} disconnected")
 
 
-def get_event_emitter(request_info):
+def get_event_emitter(request_info, update_db=True):
     async def __event_emitter__(event_data):
         user_id = request_info["user_id"]
+
         session_ids = list(
-            set(USER_POOL.get(user_id, []) + [request_info["session_id"]])
+            set(
+                USER_POOL.get(user_id, [])
+                + (
+                    [request_info.get("session_id")]
+                    if request_info.get("session_id")
+                    else []
+                )
+            )
         )
 
         for session_id in session_ids:
@@ -287,40 +325,41 @@ def get_event_emitter(request_info):
                 to=session_id,
             )
 
-        if "type" in event_data and event_data["type"] == "status":
-            Chats.add_message_status_to_chat_by_id_and_message_id(
-                request_info["chat_id"],
-                request_info["message_id"],
-                event_data.get("data", {}),
-            )
+        if update_db:
+            if "type" in event_data and event_data["type"] == "status":
+                Chats.add_message_status_to_chat_by_id_and_message_id(
+                    request_info["chat_id"],
+                    request_info["message_id"],
+                    event_data.get("data", {}),
+                )
 
-        if "type" in event_data and event_data["type"] == "message":
-            message = Chats.get_message_by_id_and_message_id(
-                request_info["chat_id"],
-                request_info["message_id"],
-            )
+            if "type" in event_data and event_data["type"] == "message":
+                message = Chats.get_message_by_id_and_message_id(
+                    request_info["chat_id"],
+                    request_info["message_id"],
+                )
 
-            content = message.get("content", "")
-            content += event_data.get("data", {}).get("content", "")
+                content = message.get("content", "")
+                content += event_data.get("data", {}).get("content", "")
 
-            Chats.upsert_message_to_chat_by_id_and_message_id(
-                request_info["chat_id"],
-                request_info["message_id"],
-                {
-                    "content": content,
-                },
-            )
+                Chats.upsert_message_to_chat_by_id_and_message_id(
+                    request_info["chat_id"],
+                    request_info["message_id"],
+                    {
+                        "content": content,
+                    },
+                )
 
-        if "type" in event_data and event_data["type"] == "replace":
-            content = event_data.get("data", {}).get("content", "")
+            if "type" in event_data and event_data["type"] == "replace":
+                content = event_data.get("data", {}).get("content", "")
 
-            Chats.upsert_message_to_chat_by_id_and_message_id(
-                request_info["chat_id"],
-                request_info["message_id"],
-                {
-                    "content": content,
-                },
-            )
+                Chats.upsert_message_to_chat_by_id_and_message_id(
+                    request_info["chat_id"],
+                    request_info["message_id"],
+                    {
+                        "content": content,
+                    },
+                )
 
     return __event_emitter__
 

backend/open_webui/socket/utils.py

@@ -1,15 +1,17 @@
 import json
-import redis
 import uuid
+from open_webui.utils.redis import get_redis_connection
 
 
 class RedisLock:
-    def __init__(self, redis_url, lock_name, timeout_secs):
+    def __init__(self, redis_url, lock_name, timeout_secs, redis_sentinels=[]):
         self.lock_name = lock_name
         self.lock_id = str(uuid.uuid4())
         self.timeout_secs = timeout_secs
         self.lock_obtained = False
-        self.redis = redis.Redis.from_url(redis_url, decode_responses=True)
+        self.redis = get_redis_connection(
+            redis_url, redis_sentinels, decode_responses=True
+        )
 
     def aquire_lock(self):
         # nx=True will only set this key if it _hasn't_ already been set
@@ -31,9 +33,11 @@ class RedisLock:
 
 
 class RedisDict:
-    def __init__(self, name, redis_url):
+    def __init__(self, name, redis_url, redis_sentinels=[]):
         self.name = name
-        self.redis = redis.Redis.from_url(redis_url, decode_responses=True)
+        self.redis = get_redis_connection(
+            redis_url, redis_sentinels, decode_responses=True
+        )
 
     def __setitem__(self, key, value):
         serialized_value = json.dumps(value)

backend/open_webui/tasks.py

@@ -5,16 +5,23 @@ from uuid import uuid4
 
 # A dictionary to keep track of active tasks
 tasks: Dict[str, asyncio.Task] = {}
+chat_tasks = {}
 
 
-def cleanup_task(task_id: str):
+def cleanup_task(task_id: str, id=None):
     """
     Remove a completed or canceled task from the global `tasks` dictionary.
     """
     tasks.pop(task_id, None)  # Remove the task if it exists
 
+    # If an ID is provided, remove the task from the chat_tasks dictionary
+    if id and task_id in chat_tasks.get(id, []):
+        chat_tasks[id].remove(task_id)
+        if not chat_tasks[id]:  # If no tasks left for this ID, remove the entry
+            chat_tasks.pop(id, None)
 
-def create_task(coroutine):
+
+def create_task(coroutine, id=None):
     """
     Create a new asyncio task and add it to the global task dictionary.
     """
@@ -22,9 +29,15 @@ def create_task(coroutine):
     task = asyncio.create_task(coroutine)  # Create the task
 
     # Add a done callback for cleanup
-    task.add_done_callback(lambda t: cleanup_task(task_id))
-
+    task.add_done_callback(lambda t: cleanup_task(task_id, id))
     tasks[task_id] = task
+
+    # If an ID is provided, associate the task with that ID
+    if chat_tasks.get(id):
+        chat_tasks[id].append(task_id)
+    else:
+        chat_tasks[id] = [task_id]
+
     return task_id, task
 
 
@@ -42,6 +55,13 @@ def list_tasks():
     return list(tasks.keys())
 
 
+def list_task_ids_by_chat_id(id):
+    """
+    List all tasks associated with a specific ID.
+    """
+    return chat_tasks.get(id, [])
+
+
 async def stop_task(task_id: str):
     """
     Cancel a running task and remove it from the global task list.

backend/open_webui/utils/auth.py

@@ -8,7 +8,9 @@ import requests
 import os
 
 
-from datetime import UTC, datetime, timedelta
+from datetime import datetime, timedelta
+import pytz
+from pytz import UTC
 from typing import Optional, Union, List, Dict
 
 from open_webui.models.users import Users
@@ -141,12 +143,14 @@ def create_api_key():
     return f"sk-{key}"
 
 
-def get_http_authorization_cred(auth_header: str):
+def get_http_authorization_cred(auth_header: Optional[str]):
+    if not auth_header:
+        return None
     try:
         scheme, credentials = auth_header.split(" ")
         return HTTPAuthorizationCredentials(scheme=scheme, credentials=credentials)
     except Exception:
-        raise ValueError(ERROR_MESSAGES.INVALID_TOKEN)
+        return None
 
 
 def get_current_user(
@@ -180,7 +184,12 @@ def get_current_user(
                 ).split(",")
             ]
 
-            if request.url.path not in allowed_paths:
+            # Check if the request path matches any allowed endpoint.
+            if not any(
+                request.url.path == allowed
+                or request.url.path.startswith(allowed + "/")
+                for allowed in allowed_paths
+            ):
                 raise HTTPException(
                     status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.API_KEY_NOT_ALLOWED
                 )

backend/open_webui/utils/filter.py

@@ -12,9 +12,9 @@ log.setLevel(SRC_LOG_LEVELS["MAIN"])
 def get_sorted_filter_ids(model: dict):
     def get_priority(function_id):
         function = Functions.get_function_by_id(function_id)
-        if function is not None and hasattr(function, "valves"):
-            # TODO: Fix FunctionModel to include vavles
-            return (function.valves if function.valves else {}).get("priority", 0)
+        if function is not None:
+            valves = Functions.get_function_valves_by_id(function_id)
+            return valves.get("priority", 0) if valves else 0
         return 0
 
     filter_ids = [function.id for function in Functions.get_global_filter_functions()]
@@ -101,11 +101,12 @@ async def process_filter_functions(
                 form_data = handler(**params)
 
         except Exception as e:
-            log.exception(f"Error in {filter_type} handler {filter_id}: {e}")
+            log.debug(f"Error in {filter_type} handler {filter_id}: {e}")
             raise e
 
     # Handle file cleanup for inlet
     if skip_files and "files" in form_data.get("metadata", {}):
+        del form_data["files"]
         del form_data["metadata"]["files"]
 
     return form_data, {}

backend/open_webui/utils/middleware.py

@@ -18,9 +18,7 @@ from uuid import uuid4
 from concurrent.futures import ThreadPoolExecutor
 
 
-from fastapi import Request
-from fastapi import BackgroundTasks
-
+from fastapi import Request, HTTPException
 from starlette.responses import Response, StreamingResponse
 
 
@@ -100,7 +98,7 @@ log.setLevel(SRC_LOG_LEVELS["MAIN"])
 
 
 async def chat_completion_tools_handler(
-    request: Request, body: dict, user: UserModel, models, tools
+    request: Request, body: dict, extra_params: dict, user: UserModel, models, tools
 ) -> tuple[dict, dict]:
     async def get_content_from_response(response) -> Optional[str]:
         content = None
@@ -135,6 +133,9 @@ async def chat_completion_tools_handler(
             "metadata": {"task": str(TASKS.FUNCTION_CALLING)},
         }
 
+    event_caller = extra_params["__event_call__"]
+    metadata = extra_params["__metadata__"]
+
     task_model_id = get_task_model_id(
         body["model"],
         request.app.state.config.TASK_MODEL,
@@ -156,7 +157,6 @@ async def chat_completion_tools_handler(
     tools_function_calling_prompt = tools_function_calling_generation_template(
         template, tools_specs
     )
-    log.info(f"{tools_function_calling_prompt=}")
     payload = get_tools_function_calling_payload(
         body["messages"], task_model_id, tools_function_calling_prompt
     )
@@ -189,52 +189,83 @@ async def chat_completion_tools_handler(
                 tool_function_params = tool_call.get("parameters", {})
 
                 try:
-                    required_params = (
-                        tools[tool_function_name]
-                        .get("spec", {})
-                        .get("parameters", {})
-                        .get("required", [])
+                    tool = tools[tool_function_name]
+
+                    spec = tool.get("spec", {})
+                    allowed_params = (
+                        spec.get("parameters", {}).get("properties", {}).keys()
                     )
-                    tool_function = tools[tool_function_name]["callable"]
                     tool_function_params = {
                         k: v
                         for k, v in tool_function_params.items()
-                        if k in required_params
+                        if k in allowed_params
                     }
-                    tool_output = await tool_function(**tool_function_params)
 
-                except Exception as e:
-                    tool_output = str(e)
-
-                if isinstance(tool_output, str):
-                    if tools[tool_function_name]["citation"]:
-                        sources.append(
+                    if tool.get("direct", False):
+                        tool_result = await event_caller(
                             {
-                                "source": {
-                                    "name": f"TOOL:{tools[tool_function_name]['toolkit_id']}/{tool_function_name}"
+                                "type": "execute:tool",
+                                "data": {
+                                    "id": str(uuid4()),
+                                    "name": tool_function_name,
+                                    "params": tool_function_params,
+                                    "server": tool.get("server", {}),
+                                    "session_id": metadata.get("session_id", None),
                                 },
-                                "document": [tool_output],
-                                "metadata": [
-                                    {
-                                        "source": f"TOOL:{tools[tool_function_name]['toolkit_id']}/{tool_function_name}"
-                                    }
-                                ],
                             }
                         )
                     else:
+                        tool_function = tool["callable"]
+                        tool_result = await tool_function(**tool_function_params)
+
+                except Exception as e:
+                    tool_result = str(e)
+
+                tool_result_files = []
+                if isinstance(tool_result, list):
+                    for item in tool_result:
+                        # check if string
+                        if isinstance(item, str) and item.startswith("data:"):
+                            tool_result_files.append(item)
+                            tool_result.remove(item)
+
+                if isinstance(tool_result, dict) or isinstance(tool_result, list):
+                    tool_result = json.dumps(tool_result, indent=2)
+
+                if isinstance(tool_result, str):
+                    tool = tools[tool_function_name]
+                    tool_id = tool.get("tool_id", "")
+
+                    tool_name = (
+                        f"{tool_id}/{tool_function_name}"
+                        if tool_id
+                        else f"{tool_function_name}"
+                    )
+                    if tool.get("metadata", {}).get("citation", False) or tool.get(
+                        "direct", False
+                    ):
+                        # Citation is enabled for this tool
                         sources.append(
                             {
-                                "source": {},
-                                "document": [tool_output],
-                                "metadata": [
-                                    {
-                                        "source": f"TOOL:{tools[tool_function_name]['toolkit_id']}/{tool_function_name}"
-                                    }
-                                ],
+                                "source": {
+                                    "name": (f"TOOL:{tool_name}"),
+                                },
+                                "document": [tool_result],
+                                "metadata": [{"source": (f"TOOL:{tool_name}")}],
                             }
                         )
+                    else:
+                        # Citation is not enabled for this tool
+                        body["messages"] = add_or_update_user_message(
+                            f"\nTool `{tool_name}` Output: {tool_result}",
+                            body["messages"],
+                        )
 
-                    if tools[tool_function_name]["file_handler"]:
+                    if (
+                        tools[tool_function_name]
+                        .get("metadata", {})
+                        .get("file_handler", False)
+                    ):
                         skip_files = True
 
             # check if "tool_calls" in result
@@ -245,10 +276,10 @@ async def chat_completion_tools_handler(
                 await tool_call_handler(result)
 
         except Exception as e:
-            log.exception(f"Error: {e}")
+            log.debug(f"Error: {e}")
             content = None
     except Exception as e:
-        log.exception(f"Error: {e}")
+        log.debug(f"Error: {e}")
         content = None
 
     log.debug(f"tool_contexts: {sources}")
@@ -352,24 +383,44 @@ async def chat_web_search_handler(
                 all_results.append(results)
                 files = form_data.get("files", [])
 
-                if results.get("collection_name"):
-                    files.append(
-                        {
-                            "collection_name": results["collection_name"],
-                            "name": searchQuery,
-                            "type": "web_search",
-                            "urls": results["filenames"],
-                        }
-                    )
+                if results.get("collection_names"):
+                    for col_idx, collection_name in enumerate(
+                        results.get("collection_names")
+                    ):
+                        files.append(
+                            {
+                                "collection_name": collection_name,
+                                "name": searchQuery,
+                                "type": "web_search",
+                                "urls": [results["filenames"][col_idx]],
+                            }
+                        )
                 elif results.get("docs"):
-                    files.append(
-                        {
-                            "docs": results.get("docs", []),
-                            "name": searchQuery,
-                            "type": "web_search",
-                            "urls": results["filenames"],
-                        }
-                    )
+                    # Invoked when bypass embedding and retrieval is set to True
+                    docs = results["docs"]
+
+                    if len(docs) == len(results["filenames"]):
+                        # the number of docs and filenames (urls) should be the same
+                        for doc_idx, doc in enumerate(docs):
+                            files.append(
+                                {
+                                    "docs": [doc],
+                                    "name": searchQuery,
+                                    "type": "web_search",
+                                    "urls": [results["filenames"][doc_idx]],
+                                }
+                            )
+                    else:
+                        # edge case when the number of docs and filenames (urls) are not the same
+                        # this should not happen, but if it does, we will just append the docs
+                        files.append(
+                            {
+                                "docs": results.get("docs", []),
+                                "name": searchQuery,
+                                "type": "web_search",
+                                "urls": results["filenames"],
+                            }
+                        )
 
                 form_data["files"] = files
         except Exception as e:
@@ -562,11 +613,12 @@ async def chat_completion_files_handler(
                         request=request,
                         files=files,
                         queries=queries,
-                        embedding_function=lambda query: request.app.state.EMBEDDING_FUNCTION(
-                            query, user=user
+                        embedding_function=lambda query, prefix: request.app.state.EMBEDDING_FUNCTION(
+                            query, prefix=prefix, user=user
                         ),
                         k=request.app.state.config.TOP_K,
                         reranking_function=request.app.state.rf,
+                        k_reranker=request.app.state.config.TOP_K_RERANKER,
                         r=request.app.state.config.RELEVANCE_THRESHOLD,
                         hybrid_search=request.app.state.config.ENABLE_RAG_HYBRID_SEARCH,
                         full_context=request.app.state.config.RAG_FULL_CONTEXT,
@@ -591,27 +643,28 @@ def apply_params_to_form_data(form_data, model):
         if "keep_alive" in params:
             form_data["keep_alive"] = params["keep_alive"]
     else:
-        if "seed" in params:
+        if "seed" in params and params["seed"] is not None:
             form_data["seed"] = params["seed"]
 
-        if "stop" in params:
+        if "stop" in params and params["stop"] is not None:
             form_data["stop"] = params["stop"]
 
-        if "temperature" in params:
+        if "temperature" in params and params["temperature"] is not None:
             form_data["temperature"] = params["temperature"]
 
-        if "max_tokens" in params:
+        if "max_tokens" in params and params["max_tokens"] is not None:
             form_data["max_tokens"] = params["max_tokens"]
 
-        if "top_p" in params:
+        if "top_p" in params and params["top_p"] is not None:
             form_data["top_p"] = params["top_p"]
 
-        if "frequency_penalty" in params:
+        if "frequency_penalty" in params and params["frequency_penalty"] is not None:
             form_data["frequency_penalty"] = params["frequency_penalty"]
 
-        if "reasoning_effort" in params:
+        if "reasoning_effort" in params and params["reasoning_effort"] is not None:
             form_data["reasoning_effort"] = params["reasoning_effort"]
-        if "logit_bias" in params:
+
+        if "logit_bias" in params and params["logit_bias"] is not None:
             try:
                 form_data["logit_bias"] = json.loads(
                     convert_logit_bias_input_to_json(params["logit_bias"])
@@ -766,12 +819,18 @@ async def process_chat_payload(request, form_data, user, metadata, model):
     }
     form_data["metadata"] = metadata
 
+    # Server side tools
     tool_ids = metadata.get("tool_ids", None)
+    # Client side tools
+    tool_servers = metadata.get("tool_servers", None)
+
     log.debug(f"{tool_ids=}")
+    log.debug(f"{tool_servers=}")
+
+    tools_dict = {}
 
     if tool_ids:
-        # If tool_ids field is present, then get the tools
-        tools = get_tools(
+        tools_dict = get_tools(
             request,
             tool_ids,
             user,
@@ -782,20 +841,31 @@ async def process_chat_payload(request, form_data, user, metadata, model):
                 "__files__": metadata.get("files", []),
             },
         )
-        log.info(f"{tools=}")
 
+    if tool_servers:
+        for tool_server in tool_servers:
+            tool_specs = tool_server.pop("specs", [])
+
+            for tool in tool_specs:
+                tools_dict[tool["name"]] = {
+                    "spec": tool,
+                    "direct": True,
+                    "server": tool_server,
+                }
+
+    if tools_dict:
         if metadata.get("function_calling") == "native":
             # If the function calling is native, then call the tools function calling handler
-            metadata["tools"] = tools
+            metadata["tools"] = tools_dict
             form_data["tools"] = [
                 {"type": "function", "function": tool.get("spec", {})}
-                for tool in tools.values()
+                for tool in tools_dict.values()
             ]
         else:
             # If the function calling is not native, then call the tools function calling handler
             try:
                 form_data, flags = await chat_completion_tools_handler(
-                    request, form_data, user, models, tools
+                    request, form_data, extra_params, user, models, tools_dict
                 )
                 sources.extend(flags.get("sources", []))
 
@@ -811,10 +881,16 @@ async def process_chat_payload(request, form_data, user, metadata, model):
     # If context is not empty, insert it into the messages
     if len(sources) > 0:
         context_string = ""
-        for source_idx, source in enumerate(sources):
+        citated_file_idx = {}
+        for _, source in enumerate(sources, 1):
             if "document" in source:
-                for doc_idx, doc_context in enumerate(source["document"]):
-                    context_string += f"<source><source_id>{source_idx}</source_id><source_context>{doc_context}</source_context></source>\n"
+                for doc_context, doc_meta in zip(
+                    source["document"], source["metadata"]
+                ):
+                    file_id = doc_meta.get("file_id")
+                    if file_id not in citated_file_idx:
+                        citated_file_idx[file_id] = len(citated_file_idx) + 1
+                    context_string += f'<source id="{citated_file_idx[file_id]}">{doc_context}</source>\n'
 
         context_string = context_string.strip()
         prompt = get_last_user_message(form_data["messages"])
@@ -991,6 +1067,16 @@ async def process_chat_response(
     # Non-streaming response
     if not isinstance(response, StreamingResponse):
         if event_emitter:
+            if "error" in response:
+                error = response["error"].get("detail", response["error"])
+                Chats.upsert_message_to_chat_by_id_and_message_id(
+                    metadata["chat_id"],
+                    metadata["message_id"],
+                    {
+                        "error": {"content": error},
+                    },
+                )
+
             if "selected_model_id" in response:
                 Chats.upsert_message_to_chat_by_id_and_message_id(
                     metadata["chat_id"],
@@ -1000,7 +1086,8 @@ async def process_chat_response(
                     },
                 )
 
-            if response.get("choices", [])[0].get("message", {}).get("content"):
+            choices = response.get("choices", [])
+            if choices and choices[0].get("message", {}).get("content"):
                 content = response["choices"][0]["message"]["content"]
 
                 if content:
@@ -1081,8 +1168,6 @@ async def process_chat_response(
         for filter_id in get_sorted_filter_ids(model)
     ]
 
-    print(f"{filter_functions=}")
-
     # Streaming response
     if event_emitter and event_caller:
         task_id = str(uuid4())  # Create a unique task ID.
@@ -1121,36 +1206,53 @@ async def process_chat_response(
                     elif block["type"] == "tool_calls":
                         attributes = block.get("attributes", {})
 
-                        block_content = block.get("content", [])
+                        tool_calls = block.get("content", [])
                         results = block.get("results", [])
 
                         if results:
 
-                            result_display_content = ""
+                            tool_calls_display_content = ""
+                            for tool_call in tool_calls:
 
-                            for result in results:
-                                tool_call_id = result.get("tool_call_id", "")
-                                tool_name = ""
+                                tool_call_id = tool_call.get("id", "")
+                                tool_name = tool_call.get("function", {}).get(
+                                    "name", ""
+                                )
+                                tool_arguments = tool_call.get("function", {}).get(
+                                    "arguments", ""
+                                )
 
-                                for tool_call in block_content:
-                                    if tool_call.get("id", "") == tool_call_id:
-                                        tool_name = tool_call.get("function", {}).get(
-                                            "name", ""
-                                        )
+                                tool_result = None
+                                tool_result_files = None
+                                for result in results:
+                                    if tool_call_id == result.get("tool_call_id", ""):
+                                        tool_result = result.get("content", None)
+                                        tool_result_files = result.get("files", None)
                                         break
 
-                                result_display_content = f"{result_display_content}\n> {tool_name}: {result.get('content', '')}"
+                                if tool_result:
+                                    tool_calls_display_content = f'{tool_calls_display_content}\n<details type="tool_calls" done="true" id="{tool_call_id}" name="{tool_name}" arguments="{html.escape(json.dumps(tool_arguments))}" result="{html.escape(json.dumps(tool_result))}" files="{html.escape(json.dumps(tool_result_files)) if tool_result_files else ""}">\n<summary>Tool Executed</summary>\n</details>\n'
+                                else:
+                                    tool_calls_display_content = f'{tool_calls_display_content}\n<details type="tool_calls" done="false" id="{tool_call_id}" name="{tool_name}" arguments="{html.escape(json.dumps(tool_arguments))}">\n<summary>Executing...</summary>\n</details>'
 
                             if not raw:
-                                content = f'{content}\n<details type="tool_calls" done="true" content="{html.escape(json.dumps(block_content))}" results="{html.escape(json.dumps(results))}">\n<summary>Tool Executed</summary>\n{result_display_content}\n</details>\n'
+                                content = f"{content}\n{tool_calls_display_content}\n\n"
                         else:
                             tool_calls_display_content = ""
 
-                            for tool_call in block_content:
-                                tool_calls_display_content = f"{tool_calls_display_content}\n> Executing {tool_call.get('function', {}).get('name', '')}"
+                            for tool_call in tool_calls:
+                                tool_call_id = tool_call.get("id", "")
+                                tool_name = tool_call.get("function", {}).get(
+                                    "name", ""
+                                )
+                                tool_arguments = tool_call.get("function", {}).get(
+                                    "arguments", ""
+                                )
+
+                                tool_calls_display_content = f'{tool_calls_display_content}\n<details type="tool_calls" done="false" id="{tool_call_id}" name="{tool_name}" arguments="{html.escape(json.dumps(tool_arguments))}">\n<summary>Executing...</summary>\n</details>'
 
                             if not raw:
-                                content = f'{content}\n<details type="tool_calls" done="false" content="{html.escape(json.dumps(block_content))}">\n<summary>Tool Executing...</summary>\n{tool_calls_display_content}\n</details>\n'
+                                content = f"{content}\n{tool_calls_display_content}\n\n"
 
                     elif block["type"] == "reasoning":
                         reasoning_display_content = "\n".join(
@@ -1495,6 +1597,9 @@ async def process_chat_response(
                             )
 
                             if data:
+                                if "event" in data:
+                                    await event_emitter(data.get("event", {}))
+
                                 if "selected_model_id" in data:
                                     model_id = data["selected_model_id"]
                                     Chats.upsert_message_to_chat_by_id_and_message_id(
@@ -1507,6 +1612,16 @@ async def process_chat_response(
                                 else:
                                     choices = data.get("choices", [])
                                     if not choices:
+                                        error = data.get("error", {})
+                                        if error:
+                                            await event_emitter(
+                                                {
+                                                    "type": "chat:completion",
+                                                    "data": {
+                                                        "error": error,
+                                                    },
+                                                }
+                                            )
                                         usage = data.get("usage", {})
                                         if usage:
                                             await event_emitter(
@@ -1529,14 +1644,27 @@ async def process_chat_response(
                                             )
 
                                             if tool_call_index is not None:
-                                                if (
-                                                    len(response_tool_calls)
-                                                    <= tool_call_index
-                                                ):
+                                                # Check if the tool call already exists
+                                                current_response_tool_call = None
+                                                for (
+                                                    response_tool_call
+                                                ) in response_tool_calls:
+                                                    if (
+                                                        response_tool_call.get("index")
+                                                        == tool_call_index
+                                                    ):
+                                                        current_response_tool_call = (
+                                                            response_tool_call
+                                                        )
+                                                        break
+
+                                                if current_response_tool_call is None:
+                                                    # Add the new tool call
                                                     response_tool_calls.append(
                                                         delta_tool_call
                                                     )
                                                 else:
+                                                    # Update the existing tool call
                                                     delta_name = delta_tool_call.get(
                                                         "function", {}
                                                     ).get("name")
@@ -1547,22 +1675,22 @@ async def process_chat_response(
                                                     )
 
                                                     if delta_name:
-                                                        response_tool_calls[
-                                                            tool_call_index
-                                                        ]["function"][
-                                                            "name"
-                                                        ] += delta_name
+                                                        current_response_tool_call[
+                                                            "function"
+                                                        ]["name"] += delta_name
 
                                                     if delta_arguments:
-                                                        response_tool_calls[
-                                                            tool_call_index
-                                                        ]["function"][
+                                                        current_response_tool_call[
+                                                            "function"
+                                                        ][
                                                             "arguments"
                                                         ] += delta_arguments
 
                                     value = delta.get("content")
 
-                                    reasoning_content = delta.get("reasoning_content")
+                                    reasoning_content = delta.get(
+                                        "reasoning_content"
+                                    ) or delta.get("reasoning")
                                     if reasoning_content:
                                         if (
                                             not content_blocks
@@ -1718,7 +1846,7 @@ async def process_chat_response(
 
                 await stream_body_handler(response)
 
-                MAX_TOOL_CALL_RETRIES = 5
+                MAX_TOOL_CALL_RETRIES = 10
                 tool_call_retries = 0
 
                 while len(tool_calls) > 0 and tool_call_retries < MAX_TOOL_CALL_RETRIES:
@@ -1757,6 +1885,15 @@ async def process_chat_response(
                             )
                         except Exception as e:
                             log.debug(e)
+                            # Fallback to JSON parsing
+                            try:
+                                tool_function_params = json.loads(
+                                    tool_call.get("function", {}).get("arguments", "{}")
+                                )
+                            except Exception as e:
+                                log.debug(
+                                    f"Error parsing tool call arguments: {tool_call.get('function', {}).get('arguments', '{}')}"
+                                )
 
                         tool_result = None
 
@@ -1765,25 +1902,65 @@ async def process_chat_response(
                             spec = tool.get("spec", {})
 
                             try:
-                                required_params = spec.get("parameters", {}).get(
-                                    "required", []
+                                allowed_params = (
+                                    spec.get("parameters", {})
+                                    .get("properties", {})
+                                    .keys()
                                 )
-                                tool_function = tool["callable"]
+
                                 tool_function_params = {
                                     k: v
                                     for k, v in tool_function_params.items()
-                                    if k in required_params
+                                    if k in allowed_params
                                 }
-                                tool_result = await tool_function(
-                                    **tool_function_params
-                                )
+
+                                if tool.get("direct", False):
+                                    tool_result = await event_caller(
+                                        {
+                                            "type": "execute:tool",
+                                            "data": {
+                                                "id": str(uuid4()),
+                                                "name": tool_name,
+                                                "params": tool_function_params,
+                                                "server": tool.get("server", {}),
+                                                "session_id": metadata.get(
+                                                    "session_id", None
+                                                ),
+                                            },
+                                        }
+                                    )
+
+                                else:
+                                    tool_function = tool["callable"]
+                                    tool_result = await tool_function(
+                                        **tool_function_params
+                                    )
+
                             except Exception as e:
                                 tool_result = str(e)
 
+                        tool_result_files = []
+                        if isinstance(tool_result, list):
+                            for item in tool_result:
+                                # check if string
+                                if isinstance(item, str) and item.startswith("data:"):
+                                    tool_result_files.append(item)
+                                    tool_result.remove(item)
+
+                        if isinstance(tool_result, dict) or isinstance(
+                            tool_result, list
+                        ):
+                            tool_result = json.dumps(tool_result, indent=2)
+
                         results.append(
                             {
                                 "tool_call_id": tool_call_id,
                                 "content": tool_result,
+                                **(
+                                    {"files": tool_result_files}
+                                    if tool_result_files
+                                    else {}
+                                ),
                             }
                         )
 
@@ -1982,11 +2159,6 @@ async def process_chat_response(
                             }
                         )
 
-                        log.info(f"content_blocks={content_blocks}")
-                        log.info(
-                            f"serialize_content_blocks={serialize_content_blocks(content_blocks)}"
-                        )
-
                         try:
                             res = await generate_chat_completion(
                                 request,
@@ -2073,7 +2245,9 @@ async def process_chat_response(
                 await response.background()
 
         # background_tasks.add_task(post_response_handler, response, events)
-        task_id, _ = create_task(post_response_handler(response, events))
+        task_id, _ = create_task(
+            post_response_handler(response, events), id=metadata["chat_id"]
+        )
         return {"status": True, "task_id": task_id}
 
     else:

backend/open_webui/utils/models.py

@@ -49,6 +49,7 @@ async def get_all_base_models(request: Request, user: UserModel = None):
                 "created": int(time.time()),
                 "owned_by": "ollama",
                 "ollama": model,
+                "tags": model.get("tags", []),
             }
             for model in ollama_models["models"]
         ]
@@ -113,9 +114,12 @@ async def get_all_models(request, user: UserModel = None):
     for custom_model in custom_models:
         if custom_model.base_model_id is None:
             for model in models:
-                if (
-                    custom_model.id == model["id"]
-                    or custom_model.id == model["id"].split(":")[0]
+                if custom_model.id == model["id"] or (
+                    model.get("owned_by") == "ollama"
+                    and custom_model.id
+                    == model["id"].split(":")[
+                        0
+                    ]  # Ollama may return model ids in different formats (e.g., 'llama3' vs. 'llama3:7b')
                 ):
                     if custom_model.is_active:
                         model["name"] = custom_model.name

backend/open_webui/utils/oauth.py

@@ -94,7 +94,7 @@ class OAuthManager:
             oauth_claim = auth_manager_config.OAUTH_ROLES_CLAIM
             oauth_allowed_roles = auth_manager_config.OAUTH_ALLOWED_ROLES
             oauth_admin_roles = auth_manager_config.OAUTH_ADMIN_ROLES
-            oauth_roles = None
+            oauth_roles = []
             # Default/fallback role if no matching roles are found
             role = auth_manager_config.DEFAULT_USER_ROLE
 
@@ -104,7 +104,7 @@ class OAuthManager:
                 nested_claims = oauth_claim.split(".")
                 for nested_claim in nested_claims:
                     claim_data = claim_data.get(nested_claim, {})
-                oauth_roles = claim_data if isinstance(claim_data, list) else None
+                oauth_roles = claim_data if isinstance(claim_data, list) else []
 
             log.debug(f"Oauth Roles claim: {oauth_claim}")
             log.debug(f"User roles from oauth: {oauth_roles}")
@@ -140,6 +140,7 @@ class OAuthManager:
         log.debug("Running OAUTH Group management")
         oauth_claim = auth_manager_config.OAUTH_GROUPS_CLAIM
 
+        user_oauth_groups = []
         # Nested claim search for groups claim
         if oauth_claim:
             claim_data = user_data
@@ -160,7 +161,7 @@ class OAuthManager:
 
         # Remove groups that user is no longer a part of
         for group_model in user_current_groups:
-            if group_model.name not in user_oauth_groups:
+            if user_oauth_groups and group_model.name not in user_oauth_groups:
                 # Remove group from user
                 log.debug(
                     f"Removing user from group {group_model.name} as it is no longer in their oauth groups"
@@ -186,8 +187,10 @@ class OAuthManager:
 
         # Add user to new groups
         for group_model in all_available_groups:
-            if group_model.name in user_oauth_groups and not any(
-                gm.name == group_model.name for gm in user_current_groups
+            if (
+                user_oauth_groups
+                and group_model.name in user_oauth_groups
+                and not any(gm.name == group_model.name for gm in user_current_groups)
             ):
                 # Add user to group
                 log.debug(
@@ -323,40 +326,45 @@ class OAuthManager:
                     raise HTTPException(400, detail=ERROR_MESSAGES.EMAIL_TAKEN)
 
                 picture_claim = auth_manager_config.OAUTH_PICTURE_CLAIM
-                picture_url = user_data.get(
-                    picture_claim, OAUTH_PROVIDERS[provider].get("picture_url", "")
-                )
-                if picture_url:
-                    # Download the profile image into a base64 string
-                    try:
-                        access_token = token.get("access_token")
-                        get_kwargs = {}
-                        if access_token:
-                            get_kwargs["headers"] = {
-                                "Authorization": f"Bearer {access_token}",
-                            }
-                        async with aiohttp.ClientSession() as session:
-                            async with session.get(picture_url, **get_kwargs) as resp:
-                                if resp.ok:
-                                    picture = await resp.read()
-                                    base64_encoded_picture = base64.b64encode(
-                                        picture
-                                    ).decode("utf-8")
-                                    guessed_mime_type = mimetypes.guess_type(
-                                        picture_url
-                                    )[0]
-                                    if guessed_mime_type is None:
-                                        # assume JPG, browsers are tolerant enough of image formats
-                                        guessed_mime_type = "image/jpeg"
-                                    picture_url = f"data:{guessed_mime_type};base64,{base64_encoded_picture}"
-                                else:
-                                    picture_url = "/user.png"
-                    except Exception as e:
-                        log.error(
-                            f"Error downloading profile image '{picture_url}': {e}"
-                        )
+                if picture_claim:
+                    picture_url = user_data.get(
+                        picture_claim, OAUTH_PROVIDERS[provider].get("picture_url", "")
+                    )
+                    if picture_url:
+                        # Download the profile image into a base64 string
+                        try:
+                            access_token = token.get("access_token")
+                            get_kwargs = {}
+                            if access_token:
+                                get_kwargs["headers"] = {
+                                    "Authorization": f"Bearer {access_token}",
+                                }
+                            async with aiohttp.ClientSession() as session:
+                                async with session.get(
+                                    picture_url, **get_kwargs
+                                ) as resp:
+                                    if resp.ok:
+                                        picture = await resp.read()
+                                        base64_encoded_picture = base64.b64encode(
+                                            picture
+                                        ).decode("utf-8")
+                                        guessed_mime_type = mimetypes.guess_type(
+                                            picture_url
+                                        )[0]
+                                        if guessed_mime_type is None:
+                                            # assume JPG, browsers are tolerant enough of image formats
+                                            guessed_mime_type = "image/jpeg"
+                                        picture_url = f"data:{guessed_mime_type};base64,{base64_encoded_picture}"
+                                    else:
+                                        picture_url = "/user.png"
+                        except Exception as e:
+                            log.error(
+                                f"Error downloading profile image '{picture_url}': {e}"
+                            )
+                            picture_url = "/user.png"
+                    if not picture_url:
                         picture_url = "/user.png"
-                if not picture_url:
+                else:
                     picture_url = "/user.png"
 
                 username_claim = auth_manager_config.OAUTH_USERNAME_CLAIM

backend/open_webui/utils/payload.py

@@ -63,6 +63,7 @@ def apply_model_params_to_body_openai(params: dict, form_data: dict) -> dict:
         "seed": lambda x: x,
         "stop": lambda x: [bytes(s, "utf-8").decode("unicode_escape") for s in x],
         "logit_bias": lambda x: x,
+        "response_format": dict,
     }
     return apply_model_params_to_body(params, form_data, mappings)
 
@@ -110,6 +111,15 @@ 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"]
+
+    if "options" in form_data and "format" in form_data["options"]:
+        form_data["format"] = form_data["options"]["format"]
+        del form_data["options"]["format"]
+
     return apply_model_params_to_body(params, form_data, mappings)
 
 
@@ -231,6 +241,11 @@ def convert_payload_openai_to_ollama(openai_payload: dict) -> dict:
                 "system"
             ]  # To prevent Ollama warning of invalid option provided
 
+        # 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"]
+
     # If there is the "stop" parameter in the openai_payload, remap it to the ollama_payload.options
     if "stop" in openai_payload:
         ollama_options = ollama_payload.get("options", {})
@@ -240,4 +255,13 @@ def convert_payload_openai_to_ollama(openai_payload: dict) -> dict:
     if "metadata" in openai_payload:
         ollama_payload["metadata"] = openai_payload["metadata"]
 
+    if "response_format" in openai_payload:
+        response_format = openai_payload["response_format"]
+        format_type = response_format.get("type", None)
+
+        schema = response_format.get(format_type, None)
+        if schema:
+            format = schema.get("schema", None)
+            ollama_payload["format"] = format
+
     return ollama_payload

backend/open_webui/utils/plugin.py

@@ -7,7 +7,7 @@ import types
 import tempfile
 import logging
 
-from open_webui.env import SRC_LOG_LEVELS
+from open_webui.env import SRC_LOG_LEVELS, PIP_OPTIONS, PIP_PACKAGE_INDEX_OPTIONS
 from open_webui.models.functions import Functions
 from open_webui.models.tools import Tools
 
@@ -68,23 +68,23 @@ def replace_imports(content):
     return content
 
 
-def load_tools_module_by_id(toolkit_id, content=None):
+def load_tool_module_by_id(tool_id, content=None):
 
     if content is None:
-        tool = Tools.get_tool_by_id(toolkit_id)
+        tool = Tools.get_tool_by_id(tool_id)
         if not tool:
-            raise Exception(f"Toolkit not found: {toolkit_id}")
+            raise Exception(f"Toolkit not found: {tool_id}")
 
         content = tool.content
 
         content = replace_imports(content)
-        Tools.update_tool_by_id(toolkit_id, {"content": content})
+        Tools.update_tool_by_id(tool_id, {"content": content})
     else:
         frontmatter = extract_frontmatter(content)
         # Install required packages found within the frontmatter
         install_frontmatter_requirements(frontmatter.get("requirements", ""))
 
-    module_name = f"tool_{toolkit_id}"
+    module_name = f"tool_{tool_id}"
     module = types.ModuleType(module_name)
     sys.modules[module_name] = module
 
@@ -108,7 +108,7 @@ def load_tools_module_by_id(toolkit_id, content=None):
         else:
             raise Exception("No Tools class found in the module")
     except Exception as e:
-        log.error(f"Error loading module: {toolkit_id}: {e}")
+        log.error(f"Error loading module: {tool_id}: {e}")
         del sys.modules[module_name]  # Clean up
         raise e
     finally:
@@ -165,15 +165,19 @@ def load_function_module_by_id(function_id, content=None):
         os.unlink(temp_file.name)
 
 
-def install_frontmatter_requirements(requirements):
+def install_frontmatter_requirements(requirements: str):
     if requirements:
         try:
             req_list = [req.strip() for req in requirements.split(",")]
-            for req in req_list:
-                log.info(f"Installing requirement: {req}")
-                subprocess.check_call([sys.executable, "-m", "pip", "install", req])
+            log.info(f"Installing requirements: {' '.join(req_list)}")
+            subprocess.check_call(
+                [sys.executable, "-m", "pip", "install"]
+                + PIP_OPTIONS
+                + req_list
+                + PIP_PACKAGE_INDEX_OPTIONS
+            )
         except Exception as e:
-            log.error(f"Error installing package: {req}")
+            log.error(f"Error installing packages: {' '.join(req_list)}")
             raise e
 
     else:

backend/open_webui/utils/redis.py

@@ -0,0 +1,58 @@
+import socketio
+import redis
+from redis import asyncio as aioredis
+from urllib.parse import urlparse
+
+
+def parse_redis_service_url(redis_url):
+    parsed_url = urlparse(redis_url)
+    if parsed_url.scheme != "redis":
+        raise ValueError("Invalid Redis URL scheme. Must be 'redis'.")
+
+    return {
+        "username": parsed_url.username or None,
+        "password": parsed_url.password or None,
+        "service": parsed_url.hostname or "mymaster",
+        "port": parsed_url.port or 6379,
+        "db": int(parsed_url.path.lstrip("/") or 0),
+    }
+
+
+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,
+        )
+
+        # Get a master connection from Sentinel
+        return sentinel.master_for(redis_config["service"])
+    else:
+        # Standard Redis connection
+        return redis.Redis.from_url(redis_url, decode_responses=decode_responses)
+
+
+def get_sentinels_from_env(sentinel_hosts_env, sentinel_port_env):
+    if sentinel_hosts_env:
+        sentinel_hosts = sentinel_hosts_env.split(",")
+        sentinel_port = int(sentinel_port_env)
+        return [(host, sentinel_port) for host in sentinel_hosts]
+    return []
+
+
+def get_sentinel_url_from_env(redis_url, sentinel_hosts_env, sentinel_port_env):
+    redis_config = parse_redis_service_url(redis_url)
+    username = redis_config["username"] or ""
+    password = redis_config["password"] or ""
+    auth_part = ""
+    if username or password:
+        auth_part = f"{username}:{password}@"
+    hosts_part = ",".join(
+        f"{host}:{sentinel_port_env}" for host in sentinel_hosts_env.split(",")
+    )
+    return f"redis+sentinel://{auth_part}{hosts_part}/{redis_config['db']}/{redis_config['service']}"

backend/open_webui/utils/task.py

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

backend/open_webui/utils/telemetry/constants.py

@@ -0,0 +1,26 @@
+from opentelemetry.semconv.trace import SpanAttributes as _SpanAttributes
+
+# Span Tags
+SPAN_DB_TYPE = "mysql"
+SPAN_REDIS_TYPE = "redis"
+SPAN_DURATION = "duration"
+SPAN_SQL_STR = "sql"
+SPAN_SQL_EXPLAIN = "explain"
+SPAN_ERROR_TYPE = "error"
+
+
+class SpanAttributes(_SpanAttributes):
+    """
+    Span Attributes
+    """
+
+    DB_INSTANCE = "db.instance"
+    DB_TYPE = "db.type"
+    DB_IP = "db.ip"
+    DB_PORT = "db.port"
+    ERROR_KIND = "error.kind"
+    ERROR_OBJECT = "error.object"
+    ERROR_MESSAGE = "error.message"
+    RESULT_CODE = "result.code"
+    RESULT_MESSAGE = "result.message"
+    RESULT_ERRORS = "result.errors"

backend/open_webui/utils/telemetry/exporters.py

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

backend/open_webui/utils/telemetry/instrumentors.py

@@ -0,0 +1,202 @@
+import logging
+import traceback
+from typing import Collection, Union
+
+from aiohttp import (
+    TraceRequestStartParams,
+    TraceRequestEndParams,
+    TraceRequestExceptionParams,
+)
+from chromadb.telemetry.opentelemetry.fastapi import instrument_fastapi
+from fastapi import FastAPI
+from opentelemetry.instrumentation.httpx import (
+    HTTPXClientInstrumentor,
+    RequestInfo,
+    ResponseInfo,
+)
+from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
+from opentelemetry.instrumentation.logging import LoggingInstrumentor
+from opentelemetry.instrumentation.redis import RedisInstrumentor
+from opentelemetry.instrumentation.requests import RequestsInstrumentor
+from opentelemetry.instrumentation.sqlalchemy import SQLAlchemyInstrumentor
+from opentelemetry.instrumentation.aiohttp_client import AioHttpClientInstrumentor
+from opentelemetry.trace import Span, StatusCode
+from redis import Redis
+from requests import PreparedRequest, Response
+from sqlalchemy import Engine
+from fastapi import status
+
+from open_webui.utils.telemetry.constants import SPAN_REDIS_TYPE, SpanAttributes
+
+from open_webui.env import SRC_LOG_LEVELS
+
+logger = logging.getLogger(__name__)
+logger.setLevel(SRC_LOG_LEVELS["MAIN"])
+
+
+def requests_hook(span: Span, request: PreparedRequest):
+    """
+    Http Request Hook
+    """
+
+    span.update_name(f"{request.method} {request.url}")
+    span.set_attributes(
+        attributes={
+            SpanAttributes.HTTP_URL: request.url,
+            SpanAttributes.HTTP_METHOD: request.method,
+        }
+    )
+
+
+def response_hook(span: Span, request: PreparedRequest, response: Response):
+    """
+    HTTP Response Hook
+    """
+
+    span.set_attributes(
+        attributes={
+            SpanAttributes.HTTP_STATUS_CODE: response.status_code,
+        }
+    )
+    span.set_status(StatusCode.ERROR if response.status_code >= 400 else StatusCode.OK)
+
+
+def redis_request_hook(span: Span, instance: Redis, args, kwargs):
+    """
+    Redis Request Hook
+    """
+
+    try:
+        connection_kwargs: dict = instance.connection_pool.connection_kwargs
+        host = connection_kwargs.get("host")
+        port = connection_kwargs.get("port")
+        db = connection_kwargs.get("db")
+        span.set_attributes(
+            {
+                SpanAttributes.DB_INSTANCE: f"{host}/{db}",
+                SpanAttributes.DB_NAME: f"{host}/{db}",
+                SpanAttributes.DB_TYPE: SPAN_REDIS_TYPE,
+                SpanAttributes.DB_PORT: port,
+                SpanAttributes.DB_IP: host,
+                SpanAttributes.DB_STATEMENT: " ".join([str(i) for i in args]),
+                SpanAttributes.DB_OPERATION: str(args[0]),
+            }
+        )
+    except Exception:  # pylint: disable=W0718
+        logger.error(traceback.format_exc())
+
+
+def httpx_request_hook(span: Span, request: RequestInfo):
+    """
+    HTTPX Request Hook
+    """
+
+    span.update_name(f"{request.method.decode()} {str(request.url)}")
+    span.set_attributes(
+        attributes={
+            SpanAttributes.HTTP_URL: str(request.url),
+            SpanAttributes.HTTP_METHOD: request.method.decode(),
+        }
+    )
+
+
+def httpx_response_hook(span: Span, request: RequestInfo, response: ResponseInfo):
+    """
+    HTTPX Response Hook
+    """
+
+    span.set_attribute(SpanAttributes.HTTP_STATUS_CODE, response.status_code)
+    span.set_status(
+        StatusCode.ERROR
+        if response.status_code >= status.HTTP_400_BAD_REQUEST
+        else StatusCode.OK
+    )
+
+
+async def httpx_async_request_hook(span: Span, request: RequestInfo):
+    """
+    Async Request Hook
+    """
+
+    httpx_request_hook(span, request)
+
+
+async def httpx_async_response_hook(
+    span: Span, request: RequestInfo, response: ResponseInfo
+):
+    """
+    Async Response Hook
+    """
+
+    httpx_response_hook(span, request, response)
+
+
+def aiohttp_request_hook(span: Span, request: TraceRequestStartParams):
+    """
+    Aiohttp Request Hook
+    """
+
+    span.update_name(f"{request.method} {str(request.url)}")
+    span.set_attributes(
+        attributes={
+            SpanAttributes.HTTP_URL: str(request.url),
+            SpanAttributes.HTTP_METHOD: request.method,
+        }
+    )
+
+
+def aiohttp_response_hook(
+    span: Span, response: Union[TraceRequestExceptionParams, TraceRequestEndParams]
+):
+    """
+    Aiohttp Response Hook
+    """
+
+    if isinstance(response, TraceRequestEndParams):
+        span.set_attribute(SpanAttributes.HTTP_STATUS_CODE, response.response.status)
+        span.set_status(
+            StatusCode.ERROR
+            if response.response.status >= status.HTTP_400_BAD_REQUEST
+            else StatusCode.OK
+        )
+    elif isinstance(response, TraceRequestExceptionParams):
+        span.set_status(StatusCode.ERROR)
+        span.set_attribute(SpanAttributes.ERROR_MESSAGE, str(response.exception))
+
+
+class Instrumentor(BaseInstrumentor):
+    """
+    Instrument OT
+    """
+
+    def __init__(self, app: FastAPI, db_engine: Engine):
+        self.app = app
+        self.db_engine = db_engine
+
+    def instrumentation_dependencies(self) -> Collection[str]:
+        return []
+
+    def _instrument(self, **kwargs):
+        instrument_fastapi(app=self.app)
+        SQLAlchemyInstrumentor().instrument(engine=self.db_engine)
+        RedisInstrumentor().instrument(request_hook=redis_request_hook)
+        RequestsInstrumentor().instrument(
+            request_hook=requests_hook, response_hook=response_hook
+        )
+        LoggingInstrumentor().instrument()
+        HTTPXClientInstrumentor().instrument(
+            request_hook=httpx_request_hook,
+            response_hook=httpx_response_hook,
+            async_request_hook=httpx_async_request_hook,
+            async_response_hook=httpx_async_response_hook,
+        )
+        AioHttpClientInstrumentor().instrument(
+            request_hook=aiohttp_request_hook,
+            response_hook=aiohttp_response_hook,
+        )
+
+    def _uninstrument(self, **kwargs):
+        if getattr(self, "instrumentors", None) is None:
+            return
+        for instrumentor in self.instrumentors:
+            instrumentor.uninstrument()

backend/open_webui/utils/telemetry/setup.py

@@ -0,0 +1,23 @@
+from fastapi import FastAPI
+from opentelemetry import trace
+from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
+from opentelemetry.sdk.resources import SERVICE_NAME, Resource
+from opentelemetry.sdk.trace import TracerProvider
+from sqlalchemy import Engine
+
+from open_webui.utils.telemetry.exporters import LazyBatchSpanProcessor
+from open_webui.utils.telemetry.instrumentors import Instrumentor
+from open_webui.env import OTEL_SERVICE_NAME, OTEL_EXPORTER_OTLP_ENDPOINT
+
+
+def setup(app: FastAPI, db_engine: Engine):
+    # set up trace
+    trace.set_tracer_provider(
+        TracerProvider(
+            resource=Resource.create(attributes={SERVICE_NAME: OTEL_SERVICE_NAME})
+        )
+    )
+    # otlp export
+    exporter = OTLPSpanExporter(endpoint=OTEL_EXPORTER_OTLP_ENDPOINT)
+    trace.get_tracer_provider().add_span_processor(LazyBatchSpanProcessor(exporter))
+    Instrumentor(app=app, db_engine=db_engine).instrument()

backend/open_webui/utils/tools.py

@@ -1,108 +1,204 @@
 import inspect
 import logging
 import re
-from typing import Any, Awaitable, Callable, get_type_hints
+import inspect
+import aiohttp
+import asyncio
+import yaml
+
+from pydantic import BaseModel
+from pydantic.fields import FieldInfo
+from typing import (
+    Any,
+    Awaitable,
+    Callable,
+    get_type_hints,
+    get_args,
+    get_origin,
+    Dict,
+    List,
+    Tuple,
+    Union,
+    Optional,
+    Type,
+)
 from functools import update_wrapper, partial
 
 
 from fastapi import Request
 from pydantic import BaseModel, Field, create_model
-from langchain_core.utils.function_calling import convert_to_openai_function
+
+from langchain_core.utils.function_calling import (
+    convert_to_openai_function as convert_pydantic_model_to_openai_function_spec,
+)
 
 
 from open_webui.models.tools import Tools
 from open_webui.models.users import UserModel
-from open_webui.utils.plugin import load_tools_module_by_id
+from open_webui.utils.plugin import load_tool_module_by_id
+from open_webui.env import AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA
+
+import copy
 
 log = logging.getLogger(__name__)
 
 
-def apply_extra_params_to_tool_function(
+def get_async_tool_function_and_apply_extra_params(
     function: Callable, extra_params: dict
 ) -> Callable[..., Awaitable]:
     sig = inspect.signature(function)
     extra_params = {k: v for k, v in extra_params.items() if k in sig.parameters}
     partial_func = partial(function, **extra_params)
+
     if inspect.iscoroutinefunction(function):
         update_wrapper(partial_func, function)
         return partial_func
+    else:
+        # Make it a coroutine function
+        async def new_function(*args, **kwargs):
+            return partial_func(*args, **kwargs)
 
-    async def new_function(*args, **kwargs):
-        return partial_func(*args, **kwargs)
-
-    update_wrapper(new_function, function)
-    return new_function
+        update_wrapper(new_function, function)
+        return new_function
 
 
-# Mutation on extra_params
 def get_tools(
     request: Request, tool_ids: list[str], user: UserModel, extra_params: dict
 ) -> dict[str, dict]:
     tools_dict = {}
 
     for tool_id in tool_ids:
-        tools = Tools.get_tool_by_id(tool_id)
-        if tools is None:
-            continue
+        tool = Tools.get_tool_by_id(tool_id)
+        if tool is None:
+            if tool_id.startswith("server:"):
+                server_idx = int(tool_id.split(":")[1])
+                tool_server_connection = (
+                    request.app.state.config.TOOL_SERVER_CONNECTIONS[server_idx]
+                )
+                tool_server_data = None
+                for server in request.app.state.TOOL_SERVERS:
+                    if server["idx"] == server_idx:
+                        tool_server_data = server
+                        break
+                assert tool_server_data is not None
+                specs = tool_server_data.get("specs", [])
 
-        module = request.app.state.TOOLS.get(tool_id, None)
-        if module is None:
-            module, _ = load_tools_module_by_id(tool_id)
-            request.app.state.TOOLS[tool_id] = module
+                for spec in specs:
+                    function_name = spec["name"]
 
-        extra_params["__id__"] = tool_id
-        if hasattr(module, "valves") and hasattr(module, "Valves"):
-            valves = Tools.get_tool_valves_by_id(tool_id) or {}
-            module.valves = module.Valves(**valves)
+                    auth_type = tool_server_connection.get("auth_type", "bearer")
+                    token = None
 
-        if hasattr(module, "UserValves"):
-            extra_params["__user__"]["valves"] = module.UserValves(  # type: ignore
-                **Tools.get_user_valves_by_id_and_user_id(tool_id, user.id)
-            )
+                    if auth_type == "bearer":
+                        token = tool_server_connection.get("key", "")
+                    elif auth_type == "session":
+                        token = request.state.token.credentials
 
-        for spec in tools.specs:
-            # TODO: Fix hack for OpenAI API
-            # Some times breaks OpenAI but others don't. Leaving the comment
-            for val in spec.get("parameters", {}).get("properties", {}).values():
-                if val["type"] == "str":
-                    val["type"] = "string"
+                    def make_tool_function(function_name, token, tool_server_data):
+                        async def tool_function(**kwargs):
+                            print(
+                                f"Executing tool function {function_name} with params: {kwargs}"
+                            )
+                            return await execute_tool_server(
+                                token=token,
+                                url=tool_server_data["url"],
+                                name=function_name,
+                                params=kwargs,
+                                server_data=tool_server_data,
+                            )
 
-            # Remove internal parameters
-            spec["parameters"]["properties"] = {
-                key: val
-                for key, val in spec["parameters"]["properties"].items()
-                if not key.startswith("__")
-            }
+                        return tool_function
 
-            function_name = spec["name"]
+                    tool_function = make_tool_function(
+                        function_name, token, tool_server_data
+                    )
 
-            # convert to function that takes only model params and inserts custom params
-            original_func = getattr(module, function_name)
-            callable = apply_extra_params_to_tool_function(original_func, extra_params)
+                    callable = get_async_tool_function_and_apply_extra_params(
+                        tool_function,
+                        {},
+                    )
 
-            if callable.__doc__ and callable.__doc__.strip() != "":
-                s = re.split(":(param|return)", callable.__doc__, 1)
-                spec["description"] = s[0]
+                    tool_dict = {
+                        "tool_id": tool_id,
+                        "callable": callable,
+                        "spec": spec,
+                    }
+
+                    # TODO: if collision, prepend toolkit name
+                    if function_name in tools_dict:
+                        log.warning(
+                            f"Tool {function_name} already exists in another tools!"
+                        )
+                        log.warning(f"Discarding {tool_id}.{function_name}")
+                    else:
+                        tools_dict[function_name] = tool_dict
             else:
-                spec["description"] = function_name
+                continue
+        else:
+            module = request.app.state.TOOLS.get(tool_id, None)
+            if module is None:
+                module, _ = load_tool_module_by_id(tool_id)
+                request.app.state.TOOLS[tool_id] = module
 
-            # TODO: This needs to be a pydantic model
-            tool_dict = {
-                "toolkit_id": tool_id,
-                "callable": callable,
-                "spec": spec,
-                "pydantic_model": function_to_pydantic_model(callable),
-                "file_handler": hasattr(module, "file_handler") and module.file_handler,
-                "citation": hasattr(module, "citation") and module.citation,
-            }
+            extra_params["__id__"] = tool_id
 
-            # TODO: if collision, prepend toolkit name
-            if function_name in tools_dict:
-                log.warning(f"Tool {function_name} already exists in another tools!")
-                log.warning(f"Collision between {tools} and {tool_id}.")
-                log.warning(f"Discarding {tools}.{function_name}")
-            else:
-                tools_dict[function_name] = tool_dict
+            # Set valves for the tool
+            if hasattr(module, "valves") and hasattr(module, "Valves"):
+                valves = Tools.get_tool_valves_by_id(tool_id) or {}
+                module.valves = module.Valves(**valves)
+            if hasattr(module, "UserValves"):
+                extra_params["__user__"]["valves"] = module.UserValves(  # type: ignore
+                    **Tools.get_user_valves_by_id_and_user_id(tool_id, user.id)
+                )
+
+            for spec in tool.specs:
+                # TODO: Fix hack for OpenAI API
+                # Some times breaks OpenAI but others don't. Leaving the comment
+                for val in spec.get("parameters", {}).get("properties", {}).values():
+                    if val["type"] == "str":
+                        val["type"] = "string"
+
+                # Remove internal reserved parameters (e.g. __id__, __user__)
+                spec["parameters"]["properties"] = {
+                    key: val
+                    for key, val in spec["parameters"]["properties"].items()
+                    if not key.startswith("__")
+                }
+
+                # convert to function that takes only model params and inserts custom params
+                function_name = spec["name"]
+                tool_function = getattr(module, function_name)
+                callable = get_async_tool_function_and_apply_extra_params(
+                    tool_function, extra_params
+                )
+
+                # TODO: Support Pydantic models as parameters
+                if callable.__doc__ and callable.__doc__.strip() != "":
+                    s = re.split(":(param|return)", callable.__doc__, 1)
+                    spec["description"] = s[0]
+                else:
+                    spec["description"] = function_name
+
+                tool_dict = {
+                    "tool_id": tool_id,
+                    "callable": callable,
+                    "spec": spec,
+                    # Misc info
+                    "metadata": {
+                        "file_handler": hasattr(module, "file_handler")
+                        and module.file_handler,
+                        "citation": hasattr(module, "citation") and module.citation,
+                    },
+                }
+
+                # TODO: if collision, prepend toolkit name
+                if function_name in tools_dict:
+                    log.warning(
+                        f"Tool {function_name} already exists in another tools!"
+                    )
+                    log.warning(f"Discarding {tool_id}.{function_name}")
+                else:
+                    tools_dict[function_name] = tool_dict
 
     return tools_dict
 
@@ -162,7 +258,7 @@ def parse_docstring(docstring):
     return param_descriptions
 
 
-def function_to_pydantic_model(func: Callable) -> type[BaseModel]:
+def convert_function_to_pydantic_model(func: Callable) -> type[BaseModel]:
     """
     Converts a Python function's type hints and docstring to a Pydantic model,
     including support for nested types, default values, and descriptions.
@@ -179,37 +275,323 @@ def function_to_pydantic_model(func: Callable) -> type[BaseModel]:
     parameters = signature.parameters
 
     docstring = func.__doc__
-    descriptions = parse_docstring(docstring)
 
-    tool_description = parse_description(docstring)
+    description = parse_description(docstring)
+    function_descriptions = parse_docstring(docstring)
 
     field_defs = {}
     for name, param in parameters.items():
+
         type_hint = type_hints.get(name, Any)
         default_value = param.default if param.default is not param.empty else ...
-        description = descriptions.get(name, None)
-        if not description:
+
+        description = function_descriptions.get(name, None)
+
+        if description:
+            field_defs[name] = type_hint, Field(default_value, description=description)
+        else:
             field_defs[name] = type_hint, default_value
-            continue
-        field_defs[name] = type_hint, Field(default_value, description=description)
 
     model = create_model(func.__name__, **field_defs)
-    model.__doc__ = tool_description
+    model.__doc__ = description
 
     return model
 
 
-def get_callable_attributes(tool: object) -> list[Callable]:
+def get_functions_from_tool(tool: object) -> list[Callable]:
     return [
         getattr(tool, func)
         for func in dir(tool)
-        if callable(getattr(tool, func))
-        and not func.startswith("__")
-        and not inspect.isclass(getattr(tool, func))
+        if callable(
+            getattr(tool, func)
+        )  # checks if the attribute is callable (a method or function).
+        and not func.startswith(
+            "__"
+        )  # filters out special (dunder) methods like init, str, etc. — these are usually built-in functions of an object that you might not need to use directly.
+        and not inspect.isclass(
+            getattr(tool, func)
+        )  # ensures that the callable is not a class itself, just a method or function.
     ]
 
 
-def get_tools_specs(tool_class: object) -> list[dict]:
-    function_list = get_callable_attributes(tool_class)
-    models = map(function_to_pydantic_model, function_list)
-    return [convert_to_openai_function(tool) for tool in models]
+def get_tool_specs(tool_module: object) -> list[dict]:
+    function_models = map(
+        convert_function_to_pydantic_model, get_functions_from_tool(tool_module)
+    )
+
+    specs = [
+        convert_pydantic_model_to_openai_function_spec(function_model)
+        for function_model in function_models
+    ]
+
+    return specs
+
+
+def resolve_schema(schema, components):
+    """
+    Recursively resolves a JSON schema using OpenAPI components.
+    """
+    if not schema:
+        return {}
+
+    if "$ref" in schema:
+        ref_path = schema["$ref"]
+        ref_parts = ref_path.strip("#/").split("/")
+        resolved = components
+        for part in ref_parts[1:]:  # Skip the initial 'components'
+            resolved = resolved.get(part, {})
+        return resolve_schema(resolved, components)
+
+    resolved_schema = copy.deepcopy(schema)
+
+    # Recursively resolve inner schemas
+    if "properties" in resolved_schema:
+        for prop, prop_schema in resolved_schema["properties"].items():
+            resolved_schema["properties"][prop] = resolve_schema(
+                prop_schema, components
+            )
+
+    if "items" in resolved_schema:
+        resolved_schema["items"] = resolve_schema(resolved_schema["items"], components)
+
+    return resolved_schema
+
+
+def convert_openapi_to_tool_payload(openapi_spec):
+    """
+    Converts an OpenAPI specification into a custom tool payload structure.
+
+    Args:
+        openapi_spec (dict): The OpenAPI specification as a Python dict.
+
+    Returns:
+        list: A list of tool payloads.
+    """
+    tool_payload = []
+
+    for path, methods in openapi_spec.get("paths", {}).items():
+        for method, operation in methods.items():
+            tool = {
+                "type": "function",
+                "name": operation.get("operationId"),
+                "description": operation.get(
+                    "description", operation.get("summary", "No description available.")
+                ),
+                "parameters": {"type": "object", "properties": {}, "required": []},
+            }
+
+            # Extract path and query parameters
+            for param in operation.get("parameters", []):
+                param_name = param["name"]
+                param_schema = param.get("schema", {})
+                tool["parameters"]["properties"][param_name] = {
+                    "type": param_schema.get("type"),
+                    "description": param_schema.get("description", ""),
+                }
+                if param.get("required"):
+                    tool["parameters"]["required"].append(param_name)
+
+            # Extract and resolve requestBody if available
+            request_body = operation.get("requestBody")
+            if request_body:
+                content = request_body.get("content", {})
+                json_schema = content.get("application/json", {}).get("schema")
+                if json_schema:
+                    resolved_schema = resolve_schema(
+                        json_schema, openapi_spec.get("components", {})
+                    )
+
+                    if resolved_schema.get("properties"):
+                        tool["parameters"]["properties"].update(
+                            resolved_schema["properties"]
+                        )
+                        if "required" in resolved_schema:
+                            tool["parameters"]["required"] = list(
+                                set(
+                                    tool["parameters"]["required"]
+                                    + resolved_schema["required"]
+                                )
+                            )
+                    elif resolved_schema.get("type") == "array":
+                        tool["parameters"] = resolved_schema  # special case for array
+
+            tool_payload.append(tool)
+
+    return tool_payload
+
+
+async def get_tool_server_data(token: str, url: str) -> Dict[str, Any]:
+    headers = {
+        "Accept": "application/json",
+        "Content-Type": "application/json",
+    }
+    if token:
+        headers["Authorization"] = f"Bearer {token}"
+
+    error = None
+    try:
+        timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT_TOOL_SERVER_DATA)
+        async with aiohttp.ClientSession(timeout=timeout) as session:
+            async with session.get(url, headers=headers) as response:
+                if response.status != 200:
+                    error_body = await response.json()
+                    raise Exception(error_body)
+
+                # Check if URL ends with .yaml or .yml to determine format
+                if url.lower().endswith((".yaml", ".yml")):
+                    text_content = await response.text()
+                    res = yaml.safe_load(text_content)
+                else:
+                    res = await response.json()
+    except Exception as err:
+        log.exception(f"Could not fetch tool server spec from {url}")
+        if isinstance(err, dict) and "detail" in err:
+            error = err["detail"]
+        else:
+            error = str(err)
+        raise Exception(error)
+
+    data = {
+        "openapi": res,
+        "info": res.get("info", {}),
+        "specs": convert_openapi_to_tool_payload(res),
+    }
+
+    print("Fetched data:", data)
+    return data
+
+
+async def get_tool_servers_data(
+    servers: List[Dict[str, Any]], session_token: Optional[str] = None
+) -> List[Dict[str, Any]]:
+    # Prepare list of enabled servers along with their original index
+    server_entries = []
+    for idx, server in enumerate(servers):
+        if server.get("config", {}).get("enable"):
+            url_path = server.get("path", "openapi.json")
+            full_url = f"{server.get('url')}/{url_path}"
+
+            auth_type = server.get("auth_type", "bearer")
+            token = None
+
+            if auth_type == "bearer":
+                token = server.get("key", "")
+            elif auth_type == "session":
+                token = session_token
+            server_entries.append((idx, server, full_url, token))
+
+    # Create async tasks to fetch data
+    tasks = [get_tool_server_data(token, url) for (_, _, url, token) in server_entries]
+
+    # Execute tasks concurrently
+    responses = await asyncio.gather(*tasks, return_exceptions=True)
+
+    # Build final results with index and server metadata
+    results = []
+    for (idx, server, url, _), response in zip(server_entries, responses):
+        if isinstance(response, Exception):
+            print(f"Failed to connect to {url} OpenAPI tool server")
+            continue
+
+        results.append(
+            {
+                "idx": idx,
+                "url": server.get("url"),
+                "openapi": response.get("openapi"),
+                "info": response.get("info"),
+                "specs": response.get("specs"),
+            }
+        )
+
+    return results
+
+
+async def execute_tool_server(
+    token: str, url: str, name: str, params: Dict[str, Any], server_data: Dict[str, Any]
+) -> Any:
+    error = None
+    try:
+        openapi = server_data.get("openapi", {})
+        paths = openapi.get("paths", {})
+
+        matching_route = None
+        for route_path, methods in paths.items():
+            for http_method, operation in methods.items():
+                if isinstance(operation, dict) and operation.get("operationId") == name:
+                    matching_route = (route_path, methods)
+                    break
+            if matching_route:
+                break
+
+        if not matching_route:
+            raise Exception(f"No matching route found for operationId: {name}")
+
+        route_path, methods = matching_route
+
+        method_entry = None
+        for http_method, operation in methods.items():
+            if operation.get("operationId") == name:
+                method_entry = (http_method.lower(), operation)
+                break
+
+        if not method_entry:
+            raise Exception(f"No matching method found for operationId: {name}")
+
+        http_method, operation = method_entry
+
+        path_params = {}
+        query_params = {}
+        body_params = {}
+
+        for param in operation.get("parameters", []):
+            param_name = param["name"]
+            param_in = param["in"]
+            if param_name in params:
+                if param_in == "path":
+                    path_params[param_name] = params[param_name]
+                elif param_in == "query":
+                    query_params[param_name] = params[param_name]
+
+        final_url = f"{url}{route_path}"
+        for key, value in path_params.items():
+            final_url = final_url.replace(f"{{{key}}}", str(value))
+
+        if query_params:
+            query_string = "&".join(f"{k}={v}" for k, v in query_params.items())
+            final_url = f"{final_url}?{query_string}"
+
+        if operation.get("requestBody", {}).get("content"):
+            if params:
+                body_params = params
+            else:
+                raise Exception(
+                    f"Request body expected for operation '{name}' but none found."
+                )
+
+        headers = {"Content-Type": "application/json"}
+
+        if token:
+            headers["Authorization"] = f"Bearer {token}"
+
+        async with aiohttp.ClientSession() as session:
+            request_method = getattr(session, http_method.lower())
+
+            if http_method in ["post", "put", "patch"]:
+                async with request_method(
+                    final_url, json=body_params, headers=headers
+                ) as response:
+                    if response.status >= 400:
+                        text = await response.text()
+                        raise Exception(f"HTTP error {response.status}: {text}")
+                    return await response.json()
+            else:
+                async with request_method(final_url, headers=headers) as response:
+                    if response.status >= 400:
+                        text = await response.text()
+                        raise Exception(f"HTTP error {response.status}: {text}")
+                    return await response.json()
+
+    except Exception as err:
+        error = str(err)
+        print("API Request Error:", error)
+        return {"error": error}

backend/requirements.txt

@@ -1,9 +1,9 @@
 fastapi==0.115.7
 uvicorn[standard]==0.34.0
 pydantic==2.10.6
-python-multipart==0.0.18
+python-multipart==0.0.20
 
-python-socketio==5.11.3
+python-socketio==5.13.0
 python-jose==3.4.0
 passlib[bcrypt]==1.7.4
 
@@ -18,7 +18,7 @@ alembic==1.14.0
 peewee==3.17.9
 peewee-migrate==1.12.2
 psycopg2-binary==2.9.9
-pgvector==0.3.5
+pgvector==0.4.0
 PyMySQL==1.1.1
 bcrypt==4.3.0
 
@@ -37,14 +37,14 @@ asgiref==3.8.1
 # AI libraries
 openai
 anthropic
-google-generativeai==0.7.2
+google-generativeai==0.8.4
 tiktoken
 
 langchain==0.3.19
 langchain-community==0.3.18
 
-fake-useragent==1.5.1
-chromadb==0.6.2
+fake-useragent==2.1.0
+chromadb==0.6.3
 pymilvus==2.5.0
 qdrant-client~=1.12.0
 opensearch-py==2.8.0
@@ -54,8 +54,9 @@ elasticsearch==8.17.1
 
 transformers
 sentence-transformers==3.3.1
+accelerate
 colbert-ai==0.2.21
-einops==0.8.0
+einops==0.8.1
 
 
 ftfy==6.2.3
@@ -67,7 +68,7 @@ python-pptx==1.0.0
 unstructured==0.16.17
 nltk==3.9.1
 Markdown==3.7
-pypandoc==1.13
+pypandoc==1.15
 pandas==2.2.3
 openpyxl==3.1.5
 pyxlsb==1.0.10
@@ -78,10 +79,13 @@ sentencepiece
 soundfile==0.13.1
 azure-ai-documentintelligence==1.0.0
 
+pillow==11.1.0
 opencv-python-headless==4.11.0.86
 rapidocr-onnxruntime==1.3.24
 rank-bm25==0.2.2
 
+onnxruntime==1.20.1
+
 faster-whisper==1.1.1
 
 PyJWT[crypto]==2.10.1
@@ -89,12 +93,12 @@ authlib==1.4.1
 
 black==25.1.0
 langfuse==2.44.0
-youtube-transcript-api==0.6.3
+youtube-transcript-api==1.0.3
 pytube==15.0.0
 
 extract_msg
 pydub
-duckduckgo-search~=7.3.2
+duckduckgo-search~=8.0.0
 
 ## Google Drive
 google-api-python-client
@@ -109,7 +113,7 @@ pytest-docker~=3.1.1
 googleapis-common-protos==1.63.2
 google-cloud-storage==2.19.0
 
-azure-identity==1.20.0
+azure-identity==1.21.0
 azure-storage-blob==12.24.1
 
 
@@ -118,3 +122,19 @@ ldap3==2.9.1
 
 ## Firecrawl
 firecrawl-py==1.12.0
+
+# Sougou API SDK(Tencentcloud SDK)
+tencentcloud-sdk-python==3.0.1336
+
+## Trace
+opentelemetry-api==1.31.1
+opentelemetry-sdk==1.31.1
+opentelemetry-exporter-otlp==1.31.1
+opentelemetry-instrumentation==0.52b1
+opentelemetry-instrumentation-fastapi==0.52b1
+opentelemetry-instrumentation-sqlalchemy==0.52b1
+opentelemetry-instrumentation-redis==0.52b1
+opentelemetry-instrumentation-requests==0.52b1
+opentelemetry-instrumentation-logging==0.52b1
+opentelemetry-instrumentation-httpx==0.52b1
+opentelemetry-instrumentation-aiohttp-client==0.52b1

backend/start.sh

@@ -4,8 +4,8 @@ SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
 cd "$SCRIPT_DIR" || exit
 
 # Add conditional Playwright browser installation
-if [[ "${RAG_WEB_LOADER_ENGINE,,}" == "playwright" ]]; then
-    if [[ -z "${PLAYWRIGHT_WS_URI}" ]]; then
+if [[ "${WEB_LOADER_ENGINE,,}" == "playwright" ]]; then
+    if [[ -z "${PLAYWRIGHT_WS_URL}" ]]; then
         echo "Installing Playwright browsers..."
         playwright install chromium
         playwright install-deps chromium
@@ -65,4 +65,4 @@ if [ -n "$SPACE_ID" ]; then
   export WEBUI_URL=${SPACE_HOST}
 fi
 
-WEBUI_SECRET_KEY="$WEBUI_SECRET_KEY" exec uvicorn open_webui.main:app --host "$HOST" --port "$PORT" --forwarded-allow-ips '*'
+WEBUI_SECRET_KEY="$WEBUI_SECRET_KEY" exec uvicorn open_webui.main:app --host "$HOST" --port "$PORT" --forwarded-allow-ips '*' --workers "${UVICORN_WORKERS:-1}"

backend/start_windows.bat

@@ -7,8 +7,8 @@ SET "SCRIPT_DIR=%~dp0"
 cd /d "%SCRIPT_DIR%" || exit /b
 
 :: Add conditional Playwright browser installation
-IF /I "%RAG_WEB_LOADER_ENGINE%" == "playwright" (
-    IF "%PLAYWRIGHT_WS_URI%" == "" (
+IF /I "%WEB_LOADER_ENGINE%" == "playwright" (
+    IF "%PLAYWRIGHT_WS_URL%" == "" (
         echo Installing Playwright browsers...
         playwright install chromium
         playwright install-deps chromium
@@ -41,4 +41,6 @@ IF "%WEBUI_SECRET_KEY%%WEBUI_JWT_SECRET_KEY%" == " " (
 
 :: Execute uvicorn
 SET "WEBUI_SECRET_KEY=%WEBUI_SECRET_KEY%"
-uvicorn open_webui.main:app --host "%HOST%" --port "%PORT%" --forwarded-allow-ips '*'
+IF "%UVICORN_WORKERS%"=="" SET UVICORN_WORKERS=1
+uvicorn open_webui.main:app --host "%HOST%" --port "%PORT%" --forwarded-allow-ips '*' --workers %UVICORN_WORKERS% --ws auto
+:: For ssl user uvicorn open_webui.main:app --host "%HOST%" --port "%PORT%" --forwarded-allow-ips '*' --ssl-keyfile "key.pem" --ssl-certfile "cert.pem" --ws auto

docker-compose.playwright.yaml

@@ -6,5 +6,5 @@ services:
 
   open-webui:
     environment:
-      - 'RAG_WEB_LOADER_ENGINE=playwright'
-      - 'PLAYWRIGHT_WS_URI=ws://playwright:3000'
+      - 'WEB_LOADER_ENGINE=playwright'
+      - 'PLAYWRIGHT_WS_URL=ws://playwright:3000'

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "open-webui",
-	"version": "0.5.20",
+	"version": "0.6.3",
 	"private": true,
 	"scripts": {
 		"dev": "npm run pyodide:fetch && vite dev --host",
@@ -66,13 +66,14 @@
 		"@tiptap/extension-highlight": "^2.10.0",
 		"@tiptap/extension-placeholder": "^2.10.0",
 		"@tiptap/extension-typography": "^2.10.0",
-		"@tiptap/pm": "^2.10.0",
+		"@tiptap/pm": "^2.11.7",
 		"@tiptap/starter-kit": "^2.10.0",
 		"@xyflow/svelte": "^0.1.19",
 		"async": "^3.2.5",
 		"bits-ui": "^0.19.7",
 		"codemirror": "^6.0.1",
 		"codemirror-lang-hcl": "^0.0.0-beta.2",
+		"codemirror-lang-elixir": "^4.0.0",
 		"crc-32": "^1.2.2",
 		"dayjs": "^1.11.10",
 		"dompurify": "^3.1.6",
@@ -80,6 +81,8 @@
 		"file-saver": "^2.0.5",
 		"fuse.js": "^7.0.0",
 		"highlight.js": "^11.9.0",
+		"html-entities": "^2.5.3",
+		"html2canvas-pro": "^1.5.8",
 		"i18next": "^23.10.0",
 		"i18next-browser-languagedetector": "^7.2.0",
 		"i18next-resources-to-backend": "^1.2.0",
@@ -102,7 +105,7 @@
 		"prosemirror-schema-list": "^1.4.1",
 		"prosemirror-state": "^1.4.3",
 		"prosemirror-view": "^1.34.3",
-		"pyodide": "^0.27.2",
+		"pyodide": "^0.27.3",
 		"socket.io-client": "^4.2.0",
 		"sortablejs": "^1.15.2",
 		"svelte-sonner": "^0.3.19",
@@ -110,7 +113,8 @@
 		"turndown": "^7.2.0",
 		"undici": "^7.3.0",
 		"uuid": "^9.0.1",
-		"vite-plugin-static-copy": "^2.2.0"
+		"vite-plugin-static-copy": "^2.2.0",
+		"yaml": "^2.7.1"
 	},
 	"engines": {
 		"node": ">=18.13.0 <=22.x.x",

pyproject.toml

@@ -9,9 +9,9 @@ dependencies = [
     "fastapi==0.115.7",
     "uvicorn[standard]==0.34.0",
     "pydantic==2.10.6",
-    "python-multipart==0.0.18",
+    "python-multipart==0.0.20",
 
-    "python-socketio==5.11.3",
+    "python-socketio==5.13.0",
     "python-jose==3.4.0",
     "passlib[bcrypt]==1.7.4",
 
@@ -26,7 +26,7 @@ dependencies = [
     "peewee==3.17.9",
     "peewee-migrate==1.12.2",
     "psycopg2-binary==2.9.9",
-    "pgvector==0.3.5",
+    "pgvector==0.4.0",
     "PyMySQL==1.1.1",
     "bcrypt==4.3.0",
 
@@ -45,14 +45,14 @@ dependencies = [
 
     "openai",
     "anthropic",
-    "google-generativeai==0.7.2",
+    "google-generativeai==0.8.4",
     "tiktoken",
 
     "langchain==0.3.19",
     "langchain-community==0.3.18",
 
-    "fake-useragent==1.5.1",
-    "chromadb==0.6.2",
+    "fake-useragent==2.1.0",
+    "chromadb==0.6.3",
     "pymilvus==2.5.0",
     "qdrant-client~=1.12.0",
     "opensearch-py==2.8.0",
@@ -61,8 +61,9 @@ dependencies = [
 
     "transformers",
     "sentence-transformers==3.3.1",
+    "accelerate",
     "colbert-ai==0.2.21",
-    "einops==0.8.0",
+    "einops==0.8.1",
 
     "ftfy==6.2.3",
     "pypdf==4.3.1",
@@ -73,7 +74,7 @@ dependencies = [
     "unstructured==0.16.17",
     "nltk==3.9.1",
     "Markdown==3.7",
-    "pypandoc==1.13",
+    "pypandoc==1.15",
     "pandas==2.2.3",
     "openpyxl==3.1.5",
     "pyxlsb==1.0.10",
@@ -84,10 +85,13 @@ dependencies = [
     "soundfile==0.13.1",
     "azure-ai-documentintelligence==1.0.0",
 
+    "pillow==11.1.0",
     "opencv-python-headless==4.11.0.86",
     "rapidocr-onnxruntime==1.3.24",
     "rank-bm25==0.2.2",
 
+    "onnxruntime==1.20.1",
+
     "faster-whisper==1.1.1",
 
     "PyJWT[crypto]==2.10.1",
@@ -95,12 +99,12 @@ dependencies = [
 
     "black==25.1.0",
     "langfuse==2.44.0",
-    "youtube-transcript-api==0.6.3",
+    "youtube-transcript-api==1.0.3",
     "pytube==15.0.0",
 
     "extract_msg",
     "pydub",
-    "duckduckgo-search~=7.3.2",
+    "duckduckgo-search~=8.0.0",
 
     "google-api-python-client",
     "google-auth-httplib2",
@@ -109,7 +113,6 @@ dependencies = [
     "docker~=7.1.0",
     "pytest~=8.3.2",
     "pytest-docker~=3.1.1",
-    "moto[s3]>=5.0.26",
 
     "googleapis-common-protos==1.63.2",
     "google-cloud-storage==2.19.0",
@@ -121,7 +124,12 @@ dependencies = [
 
     "firecrawl-py==1.12.0",
 
+    "tencentcloud-sdk-python==3.0.1336",
+
     "gcp-storage-emulator>=2024.8.3",
+
+    "moto[s3]>=5.0.26",
+
 ]
 readme = "README.md"
 requires-python = ">= 3.11, < 3.13.0a1"

src/app.css

@@ -46,6 +46,14 @@ math {
 	@apply rounded-lg;
 }
 
+input::placeholder {
+	direction: auto;
+}
+
+textarea::placeholder {
+	direction: auto;
+}
+
 .input-prose {
 	@apply prose dark:prose-invert prose-headings:font-semibold prose-hr:my-4 prose-hr:border-gray-100 prose-hr:dark:border-gray-800 prose-p:my-0 prose-img:my-1 prose-headings:my-1 prose-pre:my-0 prose-table:my-0 prose-blockquote:my-0 prose-ul:-my-0 prose-ol:-my-0 prose-li:-my-0 whitespace-pre-line;
 }
@@ -83,10 +91,6 @@ math {
 	-webkit-app-region: no-drag;
 }
 
-iframe {
-	@apply rounded-lg;
-}
-
 li p {
 	display: inline;
 }
@@ -106,7 +110,7 @@ li p {
 }
 
 ::-webkit-scrollbar {
-	height: 0.4rem;
+	height: 0.8rem;
 	width: 0.4rem;
 }
 
@@ -214,6 +218,28 @@ input[type='number'] {
 	-moz-appearance: textfield; /* Firefox */
 }
 
+.katex-display {
+	@apply overflow-y-hidden overflow-x-auto max-w-full;
+}
+
+.katex-display::-webkit-scrollbar {
+	height: 0.4rem;
+	width: 0.4rem;
+}
+
+.katex-display:active::-webkit-scrollbar-thumb,
+.katex-display:focus::-webkit-scrollbar-thumb,
+.katex-display:hover::-webkit-scrollbar-thumb {
+	visibility: visible;
+}
+.katex-display::-webkit-scrollbar-thumb {
+	visibility: hidden;
+}
+
+.katex-display::-webkit-scrollbar-corner {
+	display: none;
+}
+
 .cm-editor {
 	height: 100%;
 	width: 100%;

src/app.html

@@ -9,7 +9,7 @@
 		<link rel="apple-touch-icon" sizes="180x180" href="/static/apple-touch-icon.png" />
 		<meta name="apple-mobile-web-app-title" content="Open WebUI" />
 
-		<link rel="manifest" href="/manifest.json" />
+		<link rel="manifest" href="/manifest.json" crossorigin="use-credentials" />
 		<meta
 			name="viewport"
 			content="width=device-width, initial-scale=1, maximum-scale=1, viewport-fit=cover"

src/lib/apis/configs/index.ts

@@ -115,6 +115,93 @@ export const setDirectConnectionsConfig = async (token: string, config: object)
 	return res;
 };
 
+export const getToolServerConnections = async (token: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/tool_servers`, {
+		method: 'GET',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		}
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			error = err.detail;
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const setToolServerConnections = async (token: string, connections: object) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/tool_servers`, {
+		method: 'POST',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			...connections
+		})
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			error = err.detail;
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const verifyToolServerConnection = async (token: string, connection: object) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/tool_servers/verify`, {
+		method: 'POST',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			...connection
+		})
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			error = err.detail;
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
 export const getCodeExecutionConfig = async (token: string) => {
 	let error = null;
 

src/lib/apis/index.ts

@@ -1,6 +1,10 @@
 import { WEBUI_API_BASE_URL, WEBUI_BASE_URL } from '$lib/constants';
+import { convertOpenApiToToolPayload } from '$lib/utils';
 import { getOpenAIModelsDirect } from './openai';
 
+import { parse } from 'yaml';
+import { toast } from 'svelte-sonner';
+
 export const getModels = async (
 	token: string = '',
 	connections: object | null = null,
@@ -114,6 +118,13 @@ export const getModels = async (
 					}
 				}
 
+				const tags = apiConfig.tags;
+				if (tags) {
+					for (const model of models) {
+						model.tags = tags;
+					}
+				}
+
 				localModels = localModels.concat(models);
 			}
 		}
@@ -249,6 +260,221 @@ export const stopTask = async (token: string, id: string) => {
 	return res;
 };
 
+export const getTaskIdsByChatId = async (token: string, chat_id: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_BASE_URL}/api/tasks/chat/${chat_id}`, {
+		method: 'GET',
+		headers: {
+			Accept: 'application/json',
+			'Content-Type': 'application/json',
+			...(token && { authorization: `Bearer ${token}` })
+		}
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			if ('detail' in err) {
+				error = err.detail;
+			} else {
+				error = err;
+			}
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const getToolServerData = async (token: string, url: string) => {
+	let error = null;
+
+	const res = await fetch(`${url}`, {
+		method: 'GET',
+		headers: {
+			Accept: 'application/json',
+			'Content-Type': 'application/json',
+			...(token && { authorization: `Bearer ${token}` })
+		}
+	})
+		.then(async (res) => {
+			// Check if URL ends with .yaml or .yml to determine format
+			if (url.toLowerCase().endsWith('.yaml') || url.toLowerCase().endsWith('.yml')) {
+				if (!res.ok) throw await res.text();
+				const text = await res.text();
+				return parse(text);
+			} else {
+				if (!res.ok) throw await res.json();
+				return res.json();
+			}
+		})
+		.catch((err) => {
+			console.log(err);
+			if ('detail' in err) {
+				error = err.detail;
+			} else {
+				error = err;
+			}
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	const data = {
+		openapi: res,
+		info: res.info,
+		specs: convertOpenApiToToolPayload(res)
+	};
+
+	console.log(data);
+	return data;
+};
+
+export const getToolServersData = async (i18n, servers: object[]) => {
+	return (
+		await Promise.all(
+			servers
+				.filter((server) => server?.config?.enable)
+				.map(async (server) => {
+					const data = await getToolServerData(
+						(server?.auth_type ?? 'bearer') === 'bearer' ? server?.key : localStorage.token,
+						server?.url + '/' + (server?.path ?? 'openapi.json')
+					).catch((err) => {
+						toast.error(
+							i18n.t(`Failed to connect to {{URL}} OpenAPI tool server`, {
+								URL: server?.url + '/' + (server?.path ?? 'openapi.json')
+							})
+						);
+						return null;
+					});
+
+					if (data) {
+						const { openapi, info, specs } = data;
+						return {
+							url: server?.url,
+							openapi: openapi,
+							info: info,
+							specs: specs
+						};
+					}
+				})
+		)
+	).filter((server) => server);
+};
+
+export const executeToolServer = async (
+	token: string,
+	url: string,
+	name: string,
+	params: Record<string, any>,
+	serverData: { openapi: any; info: any; specs: any }
+) => {
+	let error = null;
+
+	try {
+		// Find the matching operationId in the OpenAPI spec
+		const matchingRoute = Object.entries(serverData.openapi.paths).find(([_, methods]) =>
+			Object.entries(methods as any).some(([__, operation]: any) => operation.operationId === name)
+		);
+
+		if (!matchingRoute) {
+			throw new Error(`No matching route found for operationId: ${name}`);
+		}
+
+		const [routePath, methods] = matchingRoute;
+
+		const methodEntry = Object.entries(methods as any).find(
+			([_, operation]: any) => operation.operationId === name
+		);
+
+		if (!methodEntry) {
+			throw new Error(`No matching method found for operationId: ${name}`);
+		}
+
+		const [httpMethod, operation]: [string, any] = methodEntry;
+
+		// Split parameters by type
+		const pathParams: Record<string, any> = {};
+		const queryParams: Record<string, any> = {};
+		let bodyParams: any = {};
+
+		if (operation.parameters) {
+			operation.parameters.forEach((param: any) => {
+				const paramName = param.name;
+				const paramIn = param.in;
+				if (params.hasOwnProperty(paramName)) {
+					if (paramIn === 'path') {
+						pathParams[paramName] = params[paramName];
+					} else if (paramIn === 'query') {
+						queryParams[paramName] = params[paramName];
+					}
+				}
+			});
+		}
+
+		let finalUrl = `${url}${routePath}`;
+
+		// Replace path parameters (`{param}`)
+		Object.entries(pathParams).forEach(([key, value]) => {
+			finalUrl = finalUrl.replace(new RegExp(`{${key}}`, 'g'), encodeURIComponent(value));
+		});
+
+		// Append query parameters to URL if any
+		if (Object.keys(queryParams).length > 0) {
+			const queryString = new URLSearchParams(
+				Object.entries(queryParams).map(([k, v]) => [k, String(v)])
+			).toString();
+			finalUrl += `?${queryString}`;
+		}
+
+		// Handle requestBody composite
+		if (operation.requestBody && operation.requestBody.content) {
+			const contentType = Object.keys(operation.requestBody.content)[0];
+			if (params !== undefined) {
+				bodyParams = params;
+			} else {
+				// Optional: Fallback or explicit error if body is expected but not provided
+				throw new Error(`Request body expected for operation '${name}' but none found.`);
+			}
+		}
+
+		// Prepare headers and request options
+		const headers: Record<string, string> = {
+			'Content-Type': 'application/json',
+			...(token && { authorization: `Bearer ${token}` })
+		};
+
+		let requestOptions: RequestInit = {
+			method: httpMethod.toUpperCase(),
+			headers
+		};
+
+		if (['post', 'put', 'patch'].includes(httpMethod.toLowerCase()) && operation.requestBody) {
+			requestOptions.body = JSON.stringify(bodyParams);
+		}
+
+		const res = await fetch(finalUrl, requestOptions);
+		if (!res.ok) {
+			const resText = await res.text();
+			throw new Error(`HTTP error! Status: ${res.status}. Message: ${resText}`);
+		}
+
+		return await res.json();
+	} catch (err: any) {
+		error = err.message;
+		console.error('API Request Error:', error);
+		return { error };
+	}
+};
+
 export const getTaskConfig = async (token: string = '') => {
 	let error = null;
 

src/lib/apis/knowledge/index.ts

@@ -345,3 +345,31 @@ export const deleteKnowledgeById = async (token: string, id: string) => {
 
 	return res;
 };
+
+export const reindexKnowledgeFiles = async (token: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/knowledge/reindex`, {
+		method: 'POST',
+		headers: {
+			Accept: 'application/json',
+			'Content-Type': 'application/json',
+			authorization: `Bearer ${token}`
+		}
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			error = err.detail;
+			console.log(err);
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};

src/lib/apis/retrieval/index.ts

@@ -50,9 +50,9 @@ type YoutubeConfigForm = {
 };
 
 type RAGConfigForm = {
-	pdf_extract_images?: boolean;
-	enable_google_drive_integration?: boolean;
-	enable_onedrive_integration?: boolean;
+	PDF_EXTRACT_IMAGES?: boolean;
+	ENABLE_GOOGLE_DRIVE_INTEGRATION?: boolean;
+	ENABLE_ONEDRIVE_INTEGRATION?: boolean;
 	chunk?: ChunkConfigForm;
 	content_extraction?: ContentExtractConfigForm;
 	web_loader_ssl_verification?: boolean;
@@ -89,33 +89,6 @@ export const updateRAGConfig = async (token: string, payload: RAGConfigForm) =>
 	return res;
 };
 
-export const getRAGTemplate = async (token: string) => {
-	let error = null;
-
-	const res = await fetch(`${RETRIEVAL_API_BASE_URL}/template`, {
-		method: 'GET',
-		headers: {
-			'Content-Type': 'application/json',
-			Authorization: `Bearer ${token}`
-		}
-	})
-		.then(async (res) => {
-			if (!res.ok) throw await res.json();
-			return res.json();
-		})
-		.catch((err) => {
-			console.log(err);
-			error = err.detail;
-			return null;
-		});
-
-	if (error) {
-		throw error;
-	}
-
-	return res?.template ?? '';
-};
-
 export const getQuerySettings = async (token: string) => {
 	let error = null;
 

src/lib/components/AddConnectionModal.svelte

@@ -14,6 +14,7 @@
 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
 	import Tooltip from '$lib/components/common/Tooltip.svelte';
 	import Switch from '$lib/components/common/Switch.svelte';
+	import Tags from './common/Tags.svelte';
 
 	export let onSubmit: Function = () => {};
 	export let onDelete: Function = () => {};
@@ -31,6 +32,7 @@
 
 	let prefixId = '';
 	let enable = true;
+	let tags = [];
 
 	let modelId = '';
 	let modelIds = [];
@@ -77,17 +79,21 @@
 	const submitHandler = async () => {
 		loading = true;
 
-		if (!ollama && (!url || !key)) {
+		if (!ollama && !url) {
 			loading = false;
-			toast.error('URL and Key are required');
+			toast.error('URL is required');
 			return;
 		}
 
+		// remove trailing slash from url
+		url = url.replace(/\/$/, '');
+
 		const connection = {
 			url,
 			key,
 			config: {
 				enable: enable,
+				tags: tags,
 				prefix_id: prefixId,
 				model_ids: modelIds
 			}
@@ -101,6 +107,7 @@
 		url = '';
 		key = '';
 		prefixId = '';
+		tags = [];
 		modelIds = [];
 	};
 
@@ -110,6 +117,7 @@
 			key = connection.key;
 
 			enable = connection.config?.enable ?? true;
+			tags = connection.config?.tags ?? [];
 			prefixId = connection.config?.prefix_id ?? '';
 			modelIds = connection.config?.model_ids ?? [];
 		}
@@ -179,7 +187,7 @@
 								</div>
 							</div>
 
-							<Tooltip content="Verify Connection" className="self-end -mb-1">
+							<Tooltip content={$i18n.t('Verify Connection')} className="self-end -mb-1">
 								<button
 									class="self-center p-1 bg-transparent hover:bg-gray-100 dark:bg-gray-900 dark:hover:bg-gray-850 rounded-lg transition"
 									on:click={() => {
@@ -218,7 +226,7 @@
 										className="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
 										bind:value={key}
 										placeholder={$i18n.t('API Key')}
-										required={!ollama}
+										required={false}
 									/>
 								</div>
 							</div>
@@ -244,6 +252,29 @@
 							</div>
 						</div>
 
+						<div class="flex gap-2 mt-2">
+							<div class="flex flex-col w-full">
+								<div class=" mb-1.5 text-xs text-gray-500">{$i18n.t('Tags')}</div>
+
+								<div class="flex-1">
+									<Tags
+										bind:tags
+										on:add={(e) => {
+											tags = [
+												...tags,
+												{
+													name: e.detail
+												}
+											];
+										}}
+										on:delete={(e) => {
+											tags = tags.filter((tag) => tag.name !== e.detail);
+										}}
+									/>
+								</div>
+							</div>
+						</div>
+
 						<hr class=" border-gray-100 dark:border-gray-700/10 my-2.5 w-full" />
 
 						<div class="flex flex-col w-full">
@@ -274,12 +305,12 @@
 							{:else}
 								<div class="text-gray-500 text-xs text-center py-2 px-10">
 									{#if ollama}
-										{$i18n.t('Leave empty to include all models from "{{URL}}/api/tags" endpoint', {
-											URL: url
+										{$i18n.t('Leave empty to include all models from "{{url}}/api/tags" endpoint', {
+											url: url
 										})}
 									{:else}
-										{$i18n.t('Leave empty to include all models from "{{URL}}/models" endpoint', {
-											URL: url
+										{$i18n.t('Leave empty to include all models from "{{url}}/models" endpoint', {
+											url: url
 										})}
 									{/if}
 								</div>

src/lib/components/AddServerModal.svelte

@@ -0,0 +1,347 @@
+<script lang="ts">
+	import { toast } from 'svelte-sonner';
+	import { getContext, onMount } from 'svelte';
+	const i18n = getContext('i18n');
+
+	import { models } from '$lib/stores';
+	import { verifyOpenAIConnection } from '$lib/apis/openai';
+	import { verifyOllamaConnection } from '$lib/apis/ollama';
+
+	import Modal from '$lib/components/common/Modal.svelte';
+	import Plus from '$lib/components/icons/Plus.svelte';
+	import Minus from '$lib/components/icons/Minus.svelte';
+	import PencilSolid from '$lib/components/icons/PencilSolid.svelte';
+	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
+	import Tooltip from '$lib/components/common/Tooltip.svelte';
+	import Switch from '$lib/components/common/Switch.svelte';
+	import Tags from './common/Tags.svelte';
+	import { getToolServerData } from '$lib/apis';
+	import { verifyToolServerConnection } from '$lib/apis/configs';
+	import AccessControl from './workspace/common/AccessControl.svelte';
+
+	export let onSubmit: Function = () => {};
+	export let onDelete: Function = () => {};
+
+	export let show = false;
+	export let edit = false;
+
+	export let direct = false;
+
+	export let connection = null;
+
+	let url = '';
+	let path = 'openapi.json';
+
+	let auth_type = 'bearer';
+	let key = '';
+
+	let accessControl = {};
+
+	let enable = true;
+
+	let loading = false;
+
+	const verifyHandler = async () => {
+		if (url === '') {
+			toast.error($i18n.t('Please enter a valid URL'));
+			return;
+		}
+
+		if (path === '') {
+			toast.error($i18n.t('Please enter a valid path'));
+			return;
+		}
+
+		if (direct) {
+			const res = await getToolServerData(
+				auth_type === 'bearer' ? key : localStorage.token,
+				`${url}/${path}`
+			).catch((err) => {
+				toast.error($i18n.t('Connection failed'));
+			});
+
+			if (res) {
+				toast.success($i18n.t('Connection successful'));
+				console.debug('Connection successful', res);
+			}
+		} else {
+			const res = await verifyToolServerConnection(localStorage.token, {
+				url,
+				path,
+				auth_type,
+				key,
+				config: {
+					enable: enable,
+					access_control: accessControl
+				}
+			}).catch((err) => {
+				toast.error($i18n.t('Connection failed'));
+			});
+
+			if (res) {
+				toast.success($i18n.t('Connection successful'));
+				console.debug('Connection successful', res);
+			}
+		}
+	};
+
+	const submitHandler = async () => {
+		loading = true;
+
+		// remove trailing slash from url
+		url = url.replace(/\/$/, '');
+
+		const connection = {
+			url,
+			path,
+			auth_type,
+			key,
+			config: {
+				enable: enable,
+				access_control: accessControl
+			}
+		};
+
+		await onSubmit(connection);
+
+		loading = false;
+		show = false;
+
+		url = '';
+		path = 'openapi.json';
+		key = '';
+		auth_type = 'bearer';
+
+		enable = true;
+		accessControl = null;
+	};
+
+	const init = () => {
+		if (connection) {
+			url = connection.url;
+			path = connection?.path ?? 'openapi.json';
+
+			auth_type = connection?.auth_type ?? 'bearer';
+			key = connection?.key ?? '';
+
+			enable = connection.config?.enable ?? true;
+			accessControl = connection.config?.access_control ?? null;
+		}
+	};
+
+	$: if (show) {
+		init();
+	}
+
+	onMount(() => {
+		init();
+	});
+</script>
+
+<Modal size="sm" bind:show>
+	<div>
+		<div class=" flex justify-between dark:text-gray-100 px-5 pt-4 pb-2">
+			<div class=" text-lg font-medium self-center font-primary">
+				{#if edit}
+					{$i18n.t('Edit Connection')}
+				{:else}
+					{$i18n.t('Add Connection')}
+				{/if}
+			</div>
+			<button
+				class="self-center"
+				on:click={() => {
+					show = false;
+				}}
+			>
+				<svg
+					xmlns="http://www.w3.org/2000/svg"
+					viewBox="0 0 20 20"
+					fill="currentColor"
+					class="w-5 h-5"
+				>
+					<path
+						d="M6.28 5.22a.75.75 0 00-1.06 1.06L8.94 10l-3.72 3.72a.75.75 0 101.06 1.06L10 11.06l3.72 3.72a.75.75 0 101.06-1.06L11.06 10l3.72-3.72a.75.75 0 00-1.06-1.06L10 8.94 6.28 5.22z"
+					/>
+				</svg>
+			</button>
+		</div>
+
+		<div class="flex flex-col md:flex-row w-full px-4 pb-4 md:space-x-4 dark:text-gray-200">
+			<div class=" flex flex-col w-full sm:flex-row sm:justify-center sm:space-x-6">
+				<form
+					class="flex flex-col w-full"
+					on:submit={(e) => {
+						e.preventDefault();
+						submitHandler();
+					}}
+				>
+					<div class="px-1">
+						<div class="flex gap-2">
+							<div class="flex flex-col w-full">
+								<div class="flex justify-between mb-0.5">
+									<div class=" text-xs text-gray-500">{$i18n.t('URL')}</div>
+								</div>
+
+								<div class="flex flex-1 items-center">
+									<input
+										class="w-full flex-1 text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
+										type="text"
+										bind:value={url}
+										placeholder={$i18n.t('API Base URL')}
+										autocomplete="off"
+										required
+									/>
+
+									<Tooltip
+										content={$i18n.t('Verify Connection')}
+										className="shrink-0 flex items-center mr-1"
+									>
+										<button
+											class="self-center p-1 bg-transparent hover:bg-gray-100 dark:bg-gray-900 dark:hover:bg-gray-850 rounded-lg transition"
+											on:click={() => {
+												verifyHandler();
+											}}
+											type="button"
+										>
+											<svg
+												xmlns="http://www.w3.org/2000/svg"
+												viewBox="0 0 20 20"
+												fill="currentColor"
+												class="w-4 h-4"
+											>
+												<path
+													fill-rule="evenodd"
+													d="M15.312 11.424a5.5 5.5 0 01-9.201 2.466l-.312-.311h2.433a.75.75 0 000-1.5H3.989a.75.75 0 00-.75.75v4.242a.75.75 0 001.5 0v-2.43l.31.31a7 7 0 0011.712-3.138.75.75 0 00-1.449-.39zm1.23-3.723a.75.75 0 00.219-.53V2.929a.75.75 0 00-1.5 0V5.36l-.31-.31A7 7 0 003.239 8.188a.75.75 0 101.448.389A5.5 5.5 0 0113.89 6.11l.311.31h-2.432a.75.75 0 000 1.5h4.243a.75.75 0 00.53-.219z"
+													clip-rule="evenodd"
+												/>
+											</svg>
+										</button>
+									</Tooltip>
+
+									<Tooltip content={enable ? $i18n.t('Enabled') : $i18n.t('Disabled')}>
+										<Switch bind:state={enable} />
+									</Tooltip>
+								</div>
+
+								<div class="flex-1 flex items-center">
+									<div class="text-sm">/</div>
+									<input
+										class="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
+										type="text"
+										bind:value={path}
+										placeholder={$i18n.t('openapi.json Path')}
+										autocomplete="off"
+										required
+									/>
+								</div>
+							</div>
+						</div>
+
+						<div class="text-xs text-gray-500 mt-1">
+							{$i18n.t(`WebUI will make requests to "{{url}}"`, {
+								url: `${url}/${path}`
+							})}
+						</div>
+
+						<div class="flex gap-2 mt-2">
+							<div class="flex flex-col w-full">
+								<div class="  text-xs text-gray-500">{$i18n.t('Auth')}</div>
+
+								<div class="flex gap-2">
+									<div class="flex-shrink-0 self-start">
+										<select
+											class="w-full text-sm bg-transparent dark:bg-gray-900 placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden pr-5"
+											bind:value={auth_type}
+										>
+											<option value="bearer">Bearer</option>
+											<option value="session">Session</option>
+										</select>
+									</div>
+
+									<div class="flex flex-1 items-center">
+										{#if auth_type === 'bearer'}
+											<SensitiveInput
+												className="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
+												bind:value={key}
+												placeholder={$i18n.t('API Key')}
+												required={false}
+											/>
+										{:else if auth_type === 'session'}
+											<div class="text-xs text-gray-500 self-center translate-y-[1px]">
+												{$i18n.t('Forwards system user session credentials to authenticate')}
+											</div>
+										{/if}
+									</div>
+								</div>
+							</div>
+						</div>
+
+						{#if !direct}
+							<hr class=" border-gray-100 dark:border-gray-700/10 my-2.5 w-full" />
+
+							<div class="my-2 -mx-2">
+								<div class="px-3 py-2 bg-gray-50 dark:bg-gray-950 rounded-lg">
+									<AccessControl bind:accessControl />
+								</div>
+							</div>
+						{/if}
+					</div>
+
+					<div class="flex justify-end pt-3 text-sm font-medium gap-1.5">
+						{#if edit}
+							<button
+								class="px-3.5 py-1.5 text-sm font-medium dark:bg-black dark:hover:bg-gray-900 dark:text-white bg-white text-black hover:bg-gray-100 transition rounded-full flex flex-row space-x-1 items-center"
+								type="button"
+								on:click={() => {
+									onDelete();
+									show = false;
+								}}
+							>
+								{$i18n.t('Delete')}
+							</button>
+						{/if}
+
+						<button
+							class="px-3.5 py-1.5 text-sm font-medium bg-black hover:bg-gray-900 text-white dark:bg-white dark:text-black dark:hover:bg-gray-100 transition rounded-full flex flex-row space-x-1 items-center {loading
+								? ' cursor-not-allowed'
+								: ''}"
+							type="submit"
+							disabled={loading}
+						>
+							{$i18n.t('Save')}
+
+							{#if loading}
+								<div class="ml-2 self-center">
+									<svg
+										class=" w-4 h-4"
+										viewBox="0 0 24 24"
+										fill="currentColor"
+										xmlns="http://www.w3.org/2000/svg"
+										><style>
+											.spinner_ajPY {
+												transform-origin: center;
+												animation: spinner_AtaB 0.75s infinite linear;
+											}
+											@keyframes spinner_AtaB {
+												100% {
+													transform: rotate(360deg);
+												}
+											}
+										</style><path
+											d="M12,1A11,11,0,1,0,23,12,11,11,0,0,0,12,1Zm0,19a8,8,0,1,1,8-8A8,8,0,0,1,12,20Z"
+											opacity=".25"
+										/><path
+											d="M10.14,1.16a11,11,0,0,0-9,8.92A1.59,1.59,0,0,0,2.46,12,1.52,1.52,0,0,0,4.11,10.7a8,8,0,0,1,6.66-6.61A1.42,1.42,0,0,0,12,2.69h0A1.57,1.57,0,0,0,10.14,1.16Z"
+											class="spinner_ajPY"
+										/></svg
+									>
+								</div>
+							{/if}
+						</button>
+					</div>
+				</form>
+			</div>
+		</div>
+	</div>
+</Modal>

src/lib/components/admin/Evaluations/Feedbacks.svelte

@@ -115,7 +115,7 @@
 		<span class="text-lg font-medium text-gray-500 dark:text-gray-300">{feedbacks.length}</span>
 	</div>
 
-	<div>
+	{#if feedbacks.length > 0}
 		<div>
 			<Tooltip content={$i18n.t('Export')}>
 				<button
@@ -128,7 +128,7 @@
 				</button>
 			</Tooltip>
 		</div>
-	</div>
+	{/if}
 </div>
 
 <div

src/lib/components/admin/Functions.svelte

@@ -430,39 +430,41 @@
 			</div>
 		</button>
 
-		<button
-			class="flex text-xs items-center space-x-1 px-3 py-1.5 rounded-xl bg-gray-50 hover:bg-gray-100 dark:bg-gray-800 dark:hover:bg-gray-700 dark:text-gray-200 transition"
-			on:click={async () => {
-				const _functions = await exportFunctions(localStorage.token).catch((error) => {
-					toast.error(`${error}`);
-					return null;
-				});
-
-				if (_functions) {
-					let blob = new Blob([JSON.stringify(_functions)], {
-						type: 'application/json'
+		{#if $functions.length}
+			<button
+				class="flex text-xs items-center space-x-1 px-3 py-1.5 rounded-xl bg-gray-50 hover:bg-gray-100 dark:bg-gray-800 dark:hover:bg-gray-700 dark:text-gray-200 transition"
+				on:click={async () => {
+					const _functions = await exportFunctions(localStorage.token).catch((error) => {
+						toast.error(`${error}`);
+						return null;
 					});
-					saveAs(blob, `functions-export-${Date.now()}.json`);
-				}
-			}}
-		>
-			<div class=" self-center mr-2 font-medium line-clamp-1">{$i18n.t('Export Functions')}</div>
 
-			<div class=" self-center">
-				<svg
-					xmlns="http://www.w3.org/2000/svg"
-					viewBox="0 0 16 16"
-					fill="currentColor"
-					class="w-4 h-4"
-				>
-					<path
-						fill-rule="evenodd"
-						d="M4 2a1.5 1.5 0 0 0-1.5 1.5v9A1.5 1.5 0 0 0 4 14h8a1.5 1.5 0 0 0 1.5-1.5V6.621a1.5 1.5 0 0 0-.44-1.06L9.94 2.439A1.5 1.5 0 0 0 8.878 2H4Zm4 3.5a.75.75 0 0 1 .75.75v2.69l.72-.72a.75.75 0 1 1 1.06 1.06l-2 2a.75.75 0 0 1-1.06 0l-2-2a.75.75 0 0 1 1.06-1.06l.72.72V6.25A.75.75 0 0 1 8 5.5Z"
-						clip-rule="evenodd"
-					/>
-				</svg>
-			</div>
-		</button>
+					if (_functions) {
+						let blob = new Blob([JSON.stringify(_functions)], {
+							type: 'application/json'
+						});
+						saveAs(blob, `functions-export-${Date.now()}.json`);
+					}
+				}}
+			>
+				<div class=" self-center mr-2 font-medium line-clamp-1">{$i18n.t('Export Functions')}</div>
+
+				<div class=" self-center">
+					<svg
+						xmlns="http://www.w3.org/2000/svg"
+						viewBox="0 0 16 16"
+						fill="currentColor"
+						class="w-4 h-4"
+					>
+						<path
+							fill-rule="evenodd"
+							d="M4 2a1.5 1.5 0 0 0-1.5 1.5v9A1.5 1.5 0 0 0 4 14h8a1.5 1.5 0 0 0 1.5-1.5V6.621a1.5 1.5 0 0 0-.44-1.06L9.94 2.439A1.5 1.5 0 0 0 8.878 2H4Zm4 3.5a.75.75 0 0 1 .75.75v2.69l.72-.72a.75.75 0 1 1 1.06 1.06l-2 2a.75.75 0 0 1-1.06 0l-2-2a.75.75 0 0 1 1.06-1.06l.72.72V6.25A.75.75 0 0 1 8 5.5Z"
+							clip-rule="evenodd"
+						/>
+					</svg>
+				</div>
+			</button>
+		{/if}
 	</div>
 </div>
 

src/lib/components/admin/Settings.svelte

@@ -20,6 +20,7 @@
 	import DocumentChartBar from '../icons/DocumentChartBar.svelte';
 	import Evaluations from './Settings/Evaluations.svelte';
 	import CodeExecution from './Settings/CodeExecution.svelte';
+	import Tools from './Settings/Tools.svelte';
 
 	const i18n = getContext('i18n');
 
@@ -71,7 +72,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -95,7 +96,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -121,7 +122,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -136,7 +137,33 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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';
+			}}
+		>
+			<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="M12 6.75a5.25 5.25 0 0 1 6.775-5.025.75.75 0 0 1 .313 1.248l-3.32 3.319c.063.475.276.934.641 1.299.365.365.824.578 1.3.64l3.318-3.319a.75.75 0 0 1 1.248.313 5.25 5.25 0 0 1-5.472 6.756c-1.018-.086-1.87.1-2.309.634L7.344 21.3A3.298 3.298 0 1 1 2.7 16.657l8.684-7.151c.533-.44.72-1.291.634-2.309A5.342 5.342 0 0 1 12 6.75ZM4.117 19.125a.75.75 0 0 1 .75-.75h.008a.75.75 0 0 1 .75.75v.008a.75.75 0 0 1-.75.75h-.008a.75.75 0 0 1-.75-.75v-.008Z"
+						clip-rule="evenodd"
+					/>
+				</svg>
+			</div>
+			<div class=" self-center">{$i18n.t('Tools')}</div>
+		</button>
+
+		<button
+			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'}"
@@ -166,7 +193,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -190,7 +217,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -216,7 +243,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -242,7 +269,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -269,7 +296,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -295,7 +322,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -325,7 +352,7 @@
 		</button>
 
 		<button
-			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			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'}"
@@ -373,6 +400,8 @@
 			<Models />
 		{:else if selectedTab === 'evaluations'}
 			<Evaluations />
+		{:else if selectedTab === 'tools'}
+			<Tools />
 		{:else if selectedTab === 'documents'}
 			<Documents
 				on:save={async () => {

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

@@ -39,6 +39,9 @@
 	let STT_ENGINE = '';
 	let STT_MODEL = '';
 	let STT_WHISPER_MODEL = '';
+	let STT_AZURE_API_KEY = '';
+	let STT_AZURE_REGION = '';
+	let STT_AZURE_LOCALES = '';
 	let STT_DEEPGRAM_API_KEY = '';
 
 	let STT_WHISPER_MODEL_LOADING = false;
@@ -108,7 +111,10 @@
 				ENGINE: STT_ENGINE,
 				MODEL: STT_MODEL,
 				WHISPER_MODEL: STT_WHISPER_MODEL,
-				DEEPGRAM_API_KEY: STT_DEEPGRAM_API_KEY
+				DEEPGRAM_API_KEY: STT_DEEPGRAM_API_KEY,
+				AZURE_API_KEY: STT_AZURE_API_KEY,
+				AZURE_REGION: STT_AZURE_REGION,
+				AZURE_LOCALES: STT_AZURE_LOCALES
 			}
 		});
 
@@ -148,6 +154,9 @@
 			STT_ENGINE = res.stt.ENGINE;
 			STT_MODEL = res.stt.MODEL;
 			STT_WHISPER_MODEL = res.stt.WHISPER_MODEL;
+			STT_AZURE_API_KEY = res.stt.AZURE_API_KEY;
+			STT_AZURE_REGION = res.stt.AZURE_REGION;
+			STT_AZURE_LOCALES = res.stt.AZURE_LOCALES;
 			STT_DEEPGRAM_API_KEY = res.stt.DEEPGRAM_API_KEY;
 		}
 
@@ -180,6 +189,7 @@
 							<option value="openai">OpenAI</option>
 							<option value="web">{$i18n.t('Web API')}</option>
 							<option value="deepgram">Deepgram</option>
+							<option value="azure">Azure AI Speech</option>
 						</select>
 					</div>
 				</div>
@@ -248,6 +258,37 @@
 							</a>
 						</div>
 					</div>
+				{:else if STT_ENGINE === 'azure'}
+					<div>
+						<div class="mt-1 flex gap-2 mb-1">
+							<SensitiveInput
+								placeholder={$i18n.t('API Key')}
+								bind:value={STT_AZURE_API_KEY}
+								required
+							/>
+							<input
+								class="flex-1 w-full rounded-lg py-2 pl-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+								placeholder={$i18n.t('Azure Region')}
+								bind:value={STT_AZURE_REGION}
+								required
+							/>
+						</div>
+
+						<hr class="border-gray-100 dark:border-gray-850 my-2" />
+
+						<div>
+							<div class=" mb-1.5 text-sm font-medium">{$i18n.t('Language Locales')}</div>
+							<div class="flex w-full">
+								<div class="flex-1">
+									<input
+										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={STT_AZURE_LOCALES}
+										placeholder={$i18n.t('e.g., en-US,ja-JP (leave blank for auto-detect)')}
+									/>
+								</div>
+							</div>
+						</div>
+					</div>
 				{:else if STT_ENGINE === ''}
 					<div>
 						<div class=" mb-1.5 text-sm font-medium">{$i18n.t('STT Model')}</div>

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

@@ -136,7 +136,7 @@
 	};
 
 	onMount(async () => {
-		if ($user.role === 'admin') {
+		if ($user?.role === 'admin') {
 			let ollamaConfig = {};
 			let openaiConfig = {};
 

src/lib/components/admin/Settings/Connections/OllamaConnection.svelte

@@ -5,6 +5,7 @@
 	import Tooltip from '$lib/components/common/Tooltip.svelte';
 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
 	import AddConnectionModal from '$lib/components/AddConnectionModal.svelte';
+	import ConfirmDialog from '$lib/components/common/ConfirmDialog.svelte';
 
 	import Cog6 from '$lib/components/icons/Cog6.svelte';
 	import Wrench from '$lib/components/icons/Wrench.svelte';
@@ -20,6 +21,7 @@
 
 	let showManageModal = false;
 	let showConfigModal = false;
+	let showDeleteConfirmDialog = false;
 </script>
 
 <AddConnectionModal
@@ -31,7 +33,9 @@
 		key: config?.key ?? '',
 		config: config
 	}}
-	{onDelete}
+	onDelete={() => {
+		showDeleteConfirmDialog = true;
+	}}
 	onSubmit={(connection) => {
 		url = connection.url;
 		config = { ...connection.config, key: connection.key };
@@ -39,6 +43,14 @@
 	}}
 />
 
+<ConfirmDialog
+	bind:show={showDeleteConfirmDialog}
+	on:confirm={() => {
+		onDelete();
+		showConfigModal = false;
+	}}
+/>
+
 <ManageOllamaModal bind:show={showManageModal} urlIdx={idx} />
 
 <div class="flex gap-1.5">

src/lib/components/admin/Settings/Connections/OpenAIConnection.svelte

@@ -6,6 +6,7 @@
 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
 	import Cog6 from '$lib/components/icons/Cog6.svelte';
 	import AddConnectionModal from '$lib/components/AddConnectionModal.svelte';
+	import ConfirmDialog from '$lib/components/common/ConfirmDialog.svelte';
 
 	import { connect } from 'socket.io-client';
 
@@ -19,8 +20,16 @@
 	export let config = {};
 
 	let showConfigModal = false;
+	let showDeleteConfirmDialog = false;
 </script>
 
+<ConfirmDialog
+	bind:show={showDeleteConfirmDialog}
+	on:confirm={() => {
+		onDelete();
+	}}
+/>
+
 <AddConnectionModal
 	edit
 	bind:show={showConfigModal}
@@ -29,7 +38,9 @@
 		key,
 		config
 	}}
-	{onDelete}
+	onDelete={() => {
+		showDeleteConfirmDialog = true;
+	}}
 	onSubmit={(connection) => {
 		url = connection.url;
 		key = connection.key;

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

@@ -77,7 +77,7 @@
 	};
 
 	onMount(async () => {
-		if ($user.role === 'admin') {
+		if ($user?.role === 'admin') {
 			evaluationConfig = await getConfig(localStorage.token).catch((err) => {
 				toast.error(err);
 				return null;

src/lib/components/admin/Settings/Evaluations/ArenaModelModal.svelte

@@ -10,6 +10,7 @@
 	import PencilSolid from '$lib/components/icons/PencilSolid.svelte';
 	import { toast } from 'svelte-sonner';
 	import AccessControl from '$lib/components/workspace/common/AccessControl.svelte';
+	import ConfirmDialog from '$lib/components/common/ConfirmDialog.svelte';
 
 	export let show = false;
 	export let edit = false;
@@ -44,6 +45,7 @@
 
 	let imageInputElement;
 	let loading = false;
+	let showDeleteConfirmDialog = false;
 
 	const addModelHandler = () => {
 		if (selectedModelId) {
@@ -115,6 +117,14 @@
 	});
 </script>
 
+<ConfirmDialog
+	bind:show={showDeleteConfirmDialog}
+	on:confirm={() => {
+		dispatch('delete', model);
+		show = false;
+	}}
+/>
+
 <Modal size="sm" bind:show>
 	<div>
 		<div class=" flex justify-between dark:text-gray-100 px-5 pt-4 pb-2">
@@ -378,8 +388,7 @@
 								class="px-3.5 py-1.5 text-sm font-medium dark:bg-black dark:hover:bg-gray-950 dark:text-white bg-white text-black hover:bg-gray-100 transition rounded-full flex flex-row space-x-1 items-center"
 								type="button"
 								on:click={() => {
-									dispatch('delete', model);
-									show = false;
+									showDeleteConfirmDialog = true;
 								}}
 							>
 								{$i18n.t('Delete')}

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

@@ -554,7 +554,6 @@
 													</div>
 													<input
 														class="w-full bg-transparent outline-hidden py-0.5"
-														required
 														placeholder={$i18n.t('Enter certificate path')}
 														bind:value={LDAP_SERVER.certificate_path}
 													/>
@@ -610,6 +609,14 @@
 						<Switch bind:state={adminConfig.ENABLE_CHANNELS} />
 					</div>
 
+					<div class="mb-2.5 flex w-full items-center justify-between pr-2">
+						<div class=" self-center text-xs font-medium">
+							{$i18n.t('User Webhooks')}
+						</div>
+
+						<Switch bind:state={adminConfig.ENABLE_USER_WEBHOOKS} />
+					</div>
+
 					<div class="mb-2.5 w-full justify-between">
 						<div class="flex w-full justify-between">
 							<div class=" self-center text-xs font-medium">{$i18n.t('WebUI URL')}</div>

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

@@ -176,7 +176,7 @@
 	};
 
 	onMount(async () => {
-		if ($user.role === 'admin') {
+		if ($user?.role === 'admin') {
 			const res = await getConfig(localStorage.token).catch((error) => {
 				toast.error(`${error}`);
 				return null;
@@ -191,11 +191,15 @@
 			}
 
 			if (config.comfyui.COMFYUI_WORKFLOW) {
-				config.comfyui.COMFYUI_WORKFLOW = JSON.stringify(
-					JSON.parse(config.comfyui.COMFYUI_WORKFLOW),
-					null,
-					2
-				);
+				try {
+					config.comfyui.COMFYUI_WORKFLOW = JSON.stringify(
+						JSON.parse(config.comfyui.COMFYUI_WORKFLOW),
+						null,
+						2
+					);
+				} catch (e) {
+					console.log(e);
+				}
 			}
 
 			requiredWorkflowNodes = requiredWorkflowNodes.map((node) => {

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

@@ -380,7 +380,7 @@
 					</div>
 				</div>
 
-				{#if $user.role === 'admin'}
+				{#if $user?.role === 'admin'}
 					<div class=" space-y-3">
 						<div class="flex w-full justify-between mb-2">
 							<div class=" self-center text-sm font-semibold">

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

@@ -29,6 +29,12 @@
 	import Wrench from '$lib/components/icons/Wrench.svelte';
 	import ArrowDownTray from '$lib/components/icons/ArrowDownTray.svelte';
 	import ManageModelsModal from './Models/ManageModelsModal.svelte';
+	import ModelMenu from '$lib/components/admin/Settings/Models/ModelMenu.svelte';
+	import EllipsisHorizontal from '$lib/components/icons/EllipsisHorizontal.svelte';
+	import EyeSlash from '$lib/components/icons/EyeSlash.svelte';
+	import Eye from '$lib/components/icons/Eye.svelte';
+
+	let shiftKey = false;
 
 	let importFiles;
 	let modelsImportInputElement: HTMLInputElement;
@@ -102,7 +108,15 @@
 				toast.success($i18n.t('Model updated successfully'));
 			}
 		} else {
-			const res = await createNewModel(localStorage.token, model).catch((error) => {
+			const res = await createNewModel(localStorage.token, {
+				meta: {},
+				id: model.id,
+				name: model.name,
+				base_model_id: null,
+				params: {},
+				access_control: {},
+				...model
+			}).catch((error) => {
 				return null;
 			});
 
@@ -146,8 +160,62 @@
 		);
 	};
 
+	const hideModelHandler = async (model) => {
+		model.meta = {
+			...model.meta,
+			hidden: !(model?.meta?.hidden ?? false)
+		};
+
+		console.log(model);
+
+		toast.success(
+			model.meta.hidden
+				? $i18n.t(`Model {{name}} is now hidden`, {
+						name: model.id
+					})
+				: $i18n.t(`Model {{name}} is now visible`, {
+						name: model.id
+					})
+		);
+
+		upsertModelHandler(model);
+	};
+
+	const exportModelHandler = async (model) => {
+		let blob = new Blob([JSON.stringify([model])], {
+			type: 'application/json'
+		});
+		saveAs(blob, `${model.id}-${Date.now()}.json`);
+	};
+
 	onMount(async () => {
-		init();
+		await init();
+
+		const onKeyDown = (event) => {
+			if (event.key === 'Shift') {
+				shiftKey = true;
+			}
+		};
+
+		const onKeyUp = (event) => {
+			if (event.key === 'Shift') {
+				shiftKey = false;
+			}
+		};
+
+		const onBlur = () => {
+			shiftKey = false;
+		};
+
+		window.addEventListener('keydown', onKeyDown);
+		window.addEventListener('keyup', onKeyUp);
+		window.addEventListener('blur-sm', onBlur);
+
+		return () => {
+			window.removeEventListener('keydown', onKeyDown);
+			window.removeEventListener('keyup', onKeyUp);
+			window.removeEventListener('blur-sm', onBlur);
+		};
 	});
 </script>
 
@@ -211,7 +279,10 @@
 			{#if models.length > 0}
 				{#each filteredModels as model, modelIdx (model.id)}
 					<div
-						class=" flex space-x-4 cursor-pointer w-full px-3 py-2 dark:hover:bg-white/5 hover:bg-black/5 rounded-lg transition"
+						class=" flex space-x-4 cursor-pointer w-full px-3 py-2 dark:hover:bg-white/5 hover:bg-black/5 rounded-lg transition {model
+							?.meta?.hidden
+							? 'opacity-50 dark:opacity-50'
+							: ''}"
 						id="model-item-{model.id}"
 					>
 						<button
@@ -261,41 +332,78 @@
 							</div>
 						</button>
 						<div class="flex flex-row gap-0.5 items-center self-center">
-							<button
-								class="self-center w-fit text-sm px-2 py-2 dark:text-gray-300 dark:hover:text-white hover:bg-black/5 dark:hover:bg-white/5 rounded-xl"
-								type="button"
-								on:click={() => {
-									selectedModelId = model.id;
-								}}
-							>
-								<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="m16.862 4.487 1.687-1.688a1.875 1.875 0 1 1 2.652 2.652L6.832 19.82a4.5 4.5 0 0 1-1.897 1.13l-2.685.8.8-2.685a4.5 4.5 0 0 1 1.13-1.897L16.863 4.487Zm0 0L19.5 7.125"
-									/>
-								</svg>
-							</button>
-
-							<div class="ml-1">
-								<Tooltip
-									content={(model?.is_active ?? true) ? $i18n.t('Enabled') : $i18n.t('Disabled')}
-								>
-									<Switch
-										bind:state={model.is_active}
-										on:change={async () => {
-											toggleModelHandler(model);
+							{#if shiftKey}
+								<Tooltip content={model?.meta?.hidden ? $i18n.t('Show') : $i18n.t('Hide')}>
+									<button
+										class="self-center w-fit text-sm px-2 py-2 dark:text-gray-300 dark:hover:text-white hover:bg-black/5 dark:hover:bg-white/5 rounded-xl"
+										type="button"
+										on:click={() => {
+											hideModelHandler(model);
 										}}
-									/>
+									>
+										{#if model?.meta?.hidden}
+											<EyeSlash />
+										{:else}
+											<Eye />
+										{/if}
+									</button>
 								</Tooltip>
-							</div>
+							{:else}
+								<button
+									class="self-center w-fit text-sm px-2 py-2 dark:text-gray-300 dark:hover:text-white hover:bg-black/5 dark:hover:bg-white/5 rounded-xl"
+									type="button"
+									on:click={() => {
+										selectedModelId = model.id;
+									}}
+								>
+									<svg
+										xmlns="http://www.w3.org/2000/svg"
+										fill="none"
+										viewBox="0 0 24 24"
+										stroke-width="1.5"
+										stroke="currentColor"
+										class="w-4 h-4"
+									>
+										<path
+											stroke-linecap="round"
+											stroke-linejoin="round"
+											d="m16.862 4.487 1.687-1.688a1.875 1.875 0 1 1 2.652 2.652L6.832 19.82a4.5 4.5 0 0 1-1.897 1.13l-2.685.8.8-2.685a4.5 4.5 0 0 1 1.13-1.897L16.863 4.487Zm0 0L19.5 7.125"
+										/>
+									</svg>
+								</button>
+
+								<ModelMenu
+									user={$user}
+									{model}
+									exportHandler={() => {
+										exportModelHandler(model);
+									}}
+									hideHandler={() => {
+										hideModelHandler(model);
+									}}
+									onClose={() => {}}
+								>
+									<button
+										class="self-center w-fit text-sm p-1.5 dark:text-gray-300 dark:hover:text-white hover:bg-black/5 dark:hover:bg-white/5 rounded-xl"
+										type="button"
+									>
+										<EllipsisHorizontal className="size-5" />
+									</button>
+								</ModelMenu>
+
+								<div class="ml-1">
+									<Tooltip
+										content={(model?.is_active ?? true) ? $i18n.t('Enabled') : $i18n.t('Disabled')}
+									>
+										<Switch
+											bind:state={model.is_active}
+											on:change={async () => {
+												toggleModelHandler(model);
+											}}
+										/>
+									</Tooltip>
+								</div>
+							{/if}
 						</div>
 					</div>
 				{/each}

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

@@ -19,7 +19,7 @@
 	let ollamaConfig = null;
 
 	onMount(async () => {
-		if ($user.role === 'admin') {
+		if ($user?.role === 'admin') {
 			await Promise.all([
 				(async () => {
 					ollamaConfig = await getOllamaConfig(localStorage.token);

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

@@ -33,6 +33,7 @@
 		if (modelListElement) {
 			sortable = Sortable.create(modelListElement, {
 				animation: 150,
+				handle: '.item-handle',
 				onUpdate: async (event) => {
 					positionChangeHandler();
 				}
@@ -47,7 +48,7 @@
 			<div class=" flex gap-2 w-full justify-between items-center" id="model-item-{modelId}">
 				<Tooltip content={modelId} placement="top-start">
 					<div class="flex items-center gap-1">
-						<EllipsisVertical className="size-4 cursor-move" />
+						<EllipsisVertical className="size-4 cursor-move item-handle" />
 
 						<div class=" text-sm flex-1 py-1 rounded-lg">
 							{#if $models.find((model) => model.id === modelId)}

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

@@ -0,0 +1,116 @@
+<script lang="ts">
+	import { DropdownMenu } from 'bits-ui';
+	import { flyAndScale } from '$lib/utils/transitions';
+	import { getContext } from 'svelte';
+
+	import Dropdown from '$lib/components/common/Dropdown.svelte';
+	import GarbageBin from '$lib/components/icons/GarbageBin.svelte';
+	import Pencil from '$lib/components/icons/Pencil.svelte';
+	import Tooltip from '$lib/components/common/Tooltip.svelte';
+	import Tags from '$lib/components/chat/Tags.svelte';
+	import Share from '$lib/components/icons/Share.svelte';
+	import ArchiveBox from '$lib/components/icons/ArchiveBox.svelte';
+	import DocumentDuplicate from '$lib/components/icons/DocumentDuplicate.svelte';
+	import ArrowDownTray from '$lib/components/icons/ArrowDownTray.svelte';
+	import ArrowUpCircle from '$lib/components/icons/ArrowUpCircle.svelte';
+
+	import { config } from '$lib/stores';
+
+	const i18n = getContext('i18n');
+
+	export let user;
+	export let model;
+
+	export let exportHandler: Function;
+	export let hideHandler: Function;
+
+	export let onClose: Function;
+
+	let show = false;
+</script>
+
+<Dropdown
+	bind:show
+	on:change={(e) => {
+		if (e.detail === false) {
+			onClose();
+		}
+	}}
+>
+	<Tooltip content={$i18n.t('More')}>
+		<slot />
+	</Tooltip>
+
+	<div slot="content">
+		<DropdownMenu.Content
+			class="w-full max-w-[160px] rounded-xl px-1 py-1.5 border border-gray-300/30 dark:border-gray-700/50 z-50 bg-white dark:bg-gray-850 dark:text-white shadow-sm"
+			sideOffset={-2}
+			side="bottom"
+			align="start"
+			transition={flyAndScale}
+		>
+			<DropdownMenu.Item
+				class="flex  gap-2  items-center px-3 py-2 text-sm  font-medium cursor-pointer hover:bg-gray-50 dark:hover:bg-gray-800 rounded-md"
+				on:click={() => {
+					hideHandler();
+				}}
+			>
+				{#if model?.meta?.hidden ?? false}
+					<svg
+						xmlns="http://www.w3.org/2000/svg"
+						fill="none"
+						viewBox="0 0 24 24"
+						stroke-width="1.5"
+						stroke="currentColor"
+						class="size-4"
+					>
+						<path
+							stroke-linecap="round"
+							stroke-linejoin="round"
+							d="M3.98 8.223A10.477 10.477 0 0 0 1.934 12C3.226 16.338 7.244 19.5 12 19.5c.993 0 1.953-.138 2.863-.395M6.228 6.228A10.451 10.451 0 0 1 12 4.5c4.756 0 8.773 3.162 10.065 7.498a10.522 10.522 0 0 1-4.293 5.774M6.228 6.228 3 3m3.228 3.228 3.65 3.65m7.894 7.894L21 21m-3.228-3.228-3.65-3.65m0 0a3 3 0 1 0-4.243-4.243m4.242 4.242L9.88 9.88"
+						/>
+					</svg>
+				{:else}
+					<svg
+						xmlns="http://www.w3.org/2000/svg"
+						fill="none"
+						viewBox="0 0 24 24"
+						stroke-width="1.5"
+						stroke="currentColor"
+						class="size-4"
+					>
+						<path
+							stroke-linecap="round"
+							stroke-linejoin="round"
+							d="M2.036 12.322a1.012 1.012 0 0 1 0-.639C3.423 7.51 7.36 4.5 12 4.5c4.638 0 8.573 3.007 9.963 7.178.07.207.07.431 0 .639C20.577 16.49 16.64 19.5 12 19.5c-4.638 0-8.573-3.007-9.963-7.178Z"
+						/>
+						<path
+							stroke-linecap="round"
+							stroke-linejoin="round"
+							d="M15 12a3 3 0 1 1-6 0 3 3 0 0 1 6 0Z"
+						/>
+					</svg>
+				{/if}
+
+				<div class="flex items-center">
+					{#if model?.meta?.hidden ?? false}
+						{$i18n.t('Show Model')}
+					{:else}
+						{$i18n.t('Hide Model')}
+					{/if}
+				</div>
+			</DropdownMenu.Item>
+
+			<DropdownMenu.Item
+				class="flex gap-2 items-center px-3 py-2 text-sm  font-medium cursor-pointer hover:bg-gray-50 dark:hover:bg-gray-800 rounded-md"
+				on:click={() => {
+					exportHandler();
+				}}
+			>
+				<ArrowDownTray />
+
+				<div class="flex items-center">{$i18n.t('Export')}</div>
+			</DropdownMenu.Item>
+		</DropdownMenu.Content>
+	</div>
+</Dropdown>

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

@@ -0,0 +1,134 @@
+<script lang="ts">
+	import { toast } from 'svelte-sonner';
+	import { createEventDispatcher, onMount, getContext, tick } from 'svelte';
+	import { getModels as _getModels } from '$lib/apis';
+
+	const dispatch = createEventDispatcher();
+	const i18n = getContext('i18n');
+
+	import { models, settings, user } from '$lib/stores';
+
+	import Switch from '$lib/components/common/Switch.svelte';
+	import Spinner from '$lib/components/common/Spinner.svelte';
+	import Tooltip from '$lib/components/common/Tooltip.svelte';
+	import Plus from '$lib/components/icons/Plus.svelte';
+	import Connection from '$lib/components/chat/Settings/Tools/Connection.svelte';
+
+	import AddServerModal from '$lib/components/AddServerModal.svelte';
+	import { getToolServerConnections, setToolServerConnections } from '$lib/apis/configs';
+
+	export let saveSettings: Function;
+
+	let servers = null;
+	let showConnectionModal = false;
+
+	const addConnectionHandler = async (server) => {
+		servers = [...servers, server];
+		await updateHandler();
+	};
+
+	const updateHandler = async () => {
+		const res = await setToolServerConnections(localStorage.token, {
+			TOOL_SERVER_CONNECTIONS: servers
+		}).catch((err) => {
+			toast.error($i18n.t('Failed to save connections'));
+
+			return null;
+		});
+
+		if (res) {
+			toast.success($i18n.t('Connections saved successfully'));
+		}
+	};
+
+	onMount(async () => {
+		const res = await getToolServerConnections(localStorage.token);
+		servers = res.TOOL_SERVER_CONNECTIONS;
+	});
+</script>
+
+<AddServerModal bind:show={showConnectionModal} onSubmit={addConnectionHandler} />
+
+<form
+	class="flex flex-col h-full justify-between text-sm"
+	on:submit|preventDefault={() => {
+		updateHandler();
+	}}
+>
+	<div class=" overflow-y-scroll scrollbar-hidden h-full">
+		{#if servers !== null}
+			<div class="">
+				<div class="mb-3">
+					<div class=" mb-2.5 text-base font-medium">{$i18n.t('General')}</div>
+
+					<hr class=" border-gray-100 dark:border-gray-850 my-2" />
+
+					<div class="mb-2.5 flex flex-col w-full justify-between">
+						<!-- {$i18n.t(`Failed to connect to {{URL}} OpenAPI tool server`, {
+							URL: 'server?.url'
+						})} -->
+						<div class="flex justify-between items-center mb-0.5">
+							<div class="font-medium">{$i18n.t('Manage Tool Servers')}</div>
+
+							<Tooltip content={$i18n.t(`Add Connection`)}>
+								<button
+									class="px-1"
+									on:click={() => {
+										showConnectionModal = true;
+									}}
+									type="button"
+								>
+									<Plus />
+								</button>
+							</Tooltip>
+						</div>
+
+						<div class="flex flex-col gap-1.5">
+							{#each servers as server, idx}
+								<Connection
+									bind:connection={server}
+									onSubmit={() => {
+										updateHandler();
+									}}
+									onDelete={() => {
+										servers = servers.filter((_, i) => i !== idx);
+										updateHandler();
+									}}
+								/>
+							{/each}
+						</div>
+
+						<div class="my-1.5">
+							<div class="text-xs text-gray-500">
+								{$i18n.t('Connect to your own OpenAPI compatible external tool servers.')}
+							</div>
+						</div>
+					</div>
+
+					<!-- <div class="mb-2.5 flex w-full justify-between">
+						<div class=" text-xs font-medium">{$i18n.t('Arena Models')}</div>
+
+						<Tooltip content={$i18n.t(`Message rating should be enabled to use this feature`)}>
+							<Switch bind:state={evaluationConfig.ENABLE_EVALUATION_ARENA_MODELS} />
+						</Tooltip>
+					</div> -->
+				</div>
+			</div>
+		{:else}
+			<div class="flex h-full justify-center">
+				<div class="my-auto">
+					<Spinner className="size-6" />
+				</div>
+			</div>
+		{/if}
+	</div>
+
+	<div class="flex justify-end pt-3 text-sm font-medium">
+		<button
+			class="px-3.5 py-1.5 text-sm font-medium bg-black hover:bg-gray-900 text-white dark:bg-white dark:text-black dark:hover:bg-gray-100 transition rounded-full"
+			type="submit"
+		>
+			{$i18n.t('Save')}
+		</button>
+	</div>
+</form>

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

@@ -12,7 +12,6 @@
 
 	export let saveHandler: Function;
 
-	let webConfig = null;
 	let webSearchEngines = [
 		'searxng',
 		'google_pse',
@@ -30,34 +29,38 @@
 		'jina',
 		'bing',
 		'exa',
-		'perplexity'
+		'perplexity',
+		'sougou'
 	];
+	let webLoaderEngines = ['playwright', 'firecrawl', 'tavily'];
 
-	let youtubeLanguage = 'en';
-	let youtubeTranslation = null;
-	let youtubeProxyUrl = '';
+	let webConfig = null;
 
 	const submitHandler = async () => {
 		// Convert domain filter string to array before sending
-		if (webConfig.search.domain_filter_list) {
-			webConfig.search.domain_filter_list = webConfig.search.domain_filter_list
-				.split(',')
+		if (webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST) {
+			webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST = webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST.split(',')
 				.map((domain) => domain.trim())
 				.filter((domain) => domain.length > 0);
 		} else {
-			webConfig.search.domain_filter_list = [];
+			webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST = [];
+		}
+
+		// Convert Youtube loader language string to array before sending
+		if (webConfig.YOUTUBE_LOADER_LANGUAGE) {
+			webConfig.YOUTUBE_LOADER_LANGUAGE = webConfig.YOUTUBE_LOADER_LANGUAGE.split(',')
+				.map((lang) => lang.trim())
+				.filter((lang) => lang.length > 0);
+		} else {
+			webConfig.YOUTUBE_LOADER_LANGUAGE = [];
 		}
 
 		const res = await updateRAGConfig(localStorage.token, {
-			web: webConfig,
-			youtube: {
-				language: youtubeLanguage.split(',').map((lang) => lang.trim()),
-				translation: youtubeTranslation,
-				proxy_url: youtubeProxyUrl
-			}
+			web: webConfig
 		});
 
-		webConfig.search.domain_filter_list = webConfig.search.domain_filter_list.join(', ');
+		webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST = webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST.join(',');
+		webConfig.YOUTUBE_LOADER_LANGUAGE = webConfig.YOUTUBE_LOADER_LANGUAGE.join(',');
 	};
 
 	onMount(async () => {
@@ -65,14 +68,13 @@
 
 		if (res) {
 			webConfig = res.web;
+
 			// Convert array back to comma-separated string for display
-			if (webConfig?.search?.domain_filter_list) {
-				webConfig.search.domain_filter_list = webConfig.search.domain_filter_list.join(', ');
+			if (webConfig?.WEB_SEARCH_DOMAIN_FILTER_LIST) {
+				webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST = webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST.join(',');
 			}
 
-			youtubeLanguage = res.youtube.language.join(',');
-			youtubeTranslation = res.youtube.translation;
-			youtubeProxyUrl = res.youtube.proxy_url;
+			webConfig.YOUTUBE_LOADER_LANGUAGE = webConfig.YOUTUBE_LOADER_LANGUAGE.join(',');
 		}
 	});
 </script>
@@ -97,7 +99,7 @@
 							{$i18n.t('Web Search')}
 						</div>
 						<div class="flex items-center relative">
-							<Switch bind:state={webConfig.search.enabled} />
+							<Switch bind:state={webConfig.ENABLE_WEB_SEARCH} />
 						</div>
 					</div>
 
@@ -108,7 +110,7 @@
 						<div class="flex items-center relative">
 							<select
 								class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
-								bind:value={webConfig.search.engine}
+								bind:value={webConfig.WEB_SEARCH_ENGINE}
 								placeholder={$i18n.t('Select a engine')}
 								required
 							>
@@ -120,8 +122,8 @@
 						</div>
 					</div>
 
-					{#if webConfig.search.engine !== ''}
-						{#if webConfig.search.engine === 'searxng'}
+					{#if webConfig.WEB_SEARCH_ENGINE !== ''}
+						{#if webConfig.WEB_SEARCH_ENGINE === 'searxng'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -134,14 +136,14 @@
 												class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 												type="text"
 												placeholder={$i18n.t('Enter Searxng Query URL')}
-												bind:value={webConfig.search.searxng_query_url}
+												bind:value={webConfig.SEARXNG_QUERY_URL}
 												autocomplete="off"
 											/>
 										</div>
 									</div>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'google_pse'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'google_pse'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -150,7 +152,7 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Google PSE API Key')}
-										bind:value={webConfig.search.google_pse_api_key}
+										bind:value={webConfig.GOOGLE_PSE_API_KEY}
 									/>
 								</div>
 								<div class="mt-1.5">
@@ -164,14 +166,14 @@
 												class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 												type="text"
 												placeholder={$i18n.t('Enter Google PSE Engine Id')}
-												bind:value={webConfig.search.google_pse_engine_id}
+												bind:value={webConfig.GOOGLE_PSE_ENGINE_ID}
 												autocomplete="off"
 											/>
 										</div>
 									</div>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'brave'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'brave'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -180,11 +182,11 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Brave Search API Key')}
-										bind:value={webConfig.search.brave_search_api_key}
+										bind:value={webConfig.BRAVE_SEARCH_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'kagi'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'kagi'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -193,12 +195,12 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Kagi Search API Key')}
-										bind:value={webConfig.search.kagi_search_api_key}
+										bind:value={webConfig.KAGI_SEARCH_API_KEY}
 									/>
 								</div>
 								.
 							</div>
-						{:else if webConfig.search.engine === 'mojeek'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'mojeek'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -207,11 +209,11 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Mojeek Search API Key')}
-										bind:value={webConfig.search.mojeek_search_api_key}
+										bind:value={webConfig.MOJEEK_SEARCH_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'bocha'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'bocha'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -220,11 +222,11 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Bocha Search API Key')}
-										bind:value={webConfig.search.bocha_search_api_key}
+										bind:value={webConfig.BOCHA_SEARCH_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'serpstack'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'serpstack'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -233,11 +235,11 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Serpstack API Key')}
-										bind:value={webConfig.search.serpstack_api_key}
+										bind:value={webConfig.SERPSTACK_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'serper'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'serper'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -246,11 +248,11 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Serper API Key')}
-										bind:value={webConfig.search.serper_api_key}
+										bind:value={webConfig.SERPER_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'serply'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'serply'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -259,11 +261,24 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Serply API Key')}
-										bind:value={webConfig.search.serply_api_key}
+										bind:value={webConfig.SERPLY_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'searchapi'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'tavily'}
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Tavily API Key')}
+									</div>
+
+									<SensitiveInput
+										placeholder={$i18n.t('Enter Tavily API Key')}
+										bind:value={webConfig.TAVILY_API_KEY}
+									/>
+								</div>
+							</div>
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'searchapi'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -272,7 +287,7 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter SearchApi API Key')}
-										bind:value={webConfig.search.searchapi_api_key}
+										bind:value={webConfig.SEARCHAPI_API_KEY}
 									/>
 								</div>
 								<div class="mt-1.5">
@@ -286,14 +301,14 @@
 												class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 												type="text"
 												placeholder={$i18n.t('Enter SearchApi Engine')}
-												bind:value={webConfig.search.searchapi_engine}
+												bind:value={webConfig.SEARCHAPI_ENGINE}
 												autocomplete="off"
 											/>
 										</div>
 									</div>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'serpapi'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'serpapi'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -302,7 +317,7 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter SerpApi API Key')}
-										bind:value={webConfig.search.serpapi_api_key}
+										bind:value={webConfig.SERPAPI_API_KEY}
 									/>
 								</div>
 								<div class="mt-1.5">
@@ -316,27 +331,14 @@
 												class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 												type="text"
 												placeholder={$i18n.t('Enter SerpApi Engine')}
-												bind:value={webConfig.search.serpapi_engine}
+												bind:value={webConfig.SERPAPI_ENGINE}
 												autocomplete="off"
 											/>
 										</div>
 									</div>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'tavily'}
-							<div class="mb-2.5 flex w-full flex-col">
-								<div>
-									<div class=" self-center text-xs font-medium mb-1">
-										{$i18n.t('Tavily API Key')}
-									</div>
-
-									<SensitiveInput
-										placeholder={$i18n.t('Enter Tavily API Key')}
-										bind:value={webConfig.search.tavily_api_key}
-									/>
-								</div>
-							</div>
-						{:else if webConfig.search.engine === 'jina'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'jina'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -345,35 +347,11 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Jina API Key')}
-										bind:value={webConfig.search.jina_api_key}
+										bind:value={webConfig.JINA_API_KEY}
 									/>
 								</div>
 							</div>
-						{:else if webConfig.search.engine === 'exa'}
-							<div class="mb-2.5 flex w-full flex-col">
-								<div>
-									<div class=" self-center text-xs font-medium mb-1">
-										{$i18n.t('Exa API Key')}
-									</div>
-
-									<SensitiveInput
-										placeholder={$i18n.t('Enter Exa API Key')}
-										bind:value={webConfig.search.exa_api_key}
-									/>
-								</div>
-							</div>
-						{:else if webConfig.search.engine === 'perplexity'}
-							<div>
-								<div class=" self-center text-xs font-medium mb-1">
-									{$i18n.t('Perplexity API Key')}
-								</div>
-
-								<SensitiveInput
-									placeholder={$i18n.t('Enter Perplexity API Key')}
-									bind:value={webConfig.search.perplexity_api_key}
-								/>
-							</div>
-						{:else if webConfig.search.engine === 'bing'}
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'bing'}
 							<div class="mb-2.5 flex w-full flex-col">
 								<div>
 									<div class=" self-center text-xs font-medium mb-1">
@@ -386,7 +364,7 @@
 												class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 												type="text"
 												placeholder={$i18n.t('Enter Bing Search V7 Endpoint')}
-												bind:value={webConfig.search.bing_search_v7_endpoint}
+												bind:value={webConfig.BING_SEARCH_V7_ENDPOINT}
 												autocomplete="off"
 											/>
 										</div>
@@ -400,14 +378,63 @@
 
 									<SensitiveInput
 										placeholder={$i18n.t('Enter Bing Search V7 Subscription Key')}
-										bind:value={webConfig.search.bing_search_v7_subscription_key}
+										bind:value={webConfig.BING_SEARCH_V7_SUBSCRIPTION_KEY}
+									/>
+								</div>
+							</div>
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'exa'}
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Exa API Key')}
+									</div>
+
+									<SensitiveInput
+										placeholder={$i18n.t('Enter Exa API Key')}
+										bind:value={webConfig.EXA_API_KEY}
+									/>
+								</div>
+							</div>
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'perplexity'}
+							<div>
+								<div class=" self-center text-xs font-medium mb-1">
+									{$i18n.t('Perplexity API Key')}
+								</div>
+
+								<SensitiveInput
+									placeholder={$i18n.t('Enter Perplexity API Key')}
+									bind:value={webConfig.PERPLEXITY_API_KEY}
+								/>
+							</div>
+						{:else if webConfig.WEB_SEARCH_ENGINE === 'sougou'}
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Sougou Search API sID')}
+									</div>
+
+									<SensitiveInput
+										placeholder={$i18n.t('Enter Sougou Search API sID')}
+										bind:value={webConfig.SOUGOU_API_SID}
+									/>
+								</div>
+							</div>
+							<div class="mb-2.5 flex w-full flex-col">
+								<div>
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Sougou Search API SK')}
+									</div>
+
+									<SensitiveInput
+										placeholder={$i18n.t('Enter Sougou Search API SK')}
+										bind:value={webConfig.SOUGOU_API_SK}
 									/>
 								</div>
 							</div>
 						{/if}
 					{/if}
 
-					{#if webConfig.search.enabled}
+					{#if webConfig.ENABLE_WEB_SEARCH}
 						<div class="mb-2.5 flex w-full flex-col">
 							<div class="flex gap-2">
 								<div class="w-full">
@@ -418,7 +445,7 @@
 									<input
 										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 										placeholder={$i18n.t('Search Result Count')}
-										bind:value={webConfig.search.result_count}
+										bind:value={webConfig.WEB_SEARCH_RESULT_COUNT}
 										required
 									/>
 								</div>
@@ -431,7 +458,7 @@
 									<input
 										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 										placeholder={$i18n.t('Concurrent Requests')}
-										bind:value={webConfig.search.concurrent_requests}
+										bind:value={webConfig.WEB_SEARCH_CONCURRENT_REQUESTS}
 										required
 									/>
 								</div>
@@ -448,7 +475,7 @@
 								placeholder={$i18n.t(
 									'Enter domains separated by commas (e.g., example.com,site.org)'
 								)}
-								bind:value={webConfig.search.domain_filter_list}
+								bind:value={webConfig.WEB_SEARCH_DOMAIN_FILTER_LIST}
 							/>
 						</div>
 					{/if}
@@ -462,8 +489,12 @@
 						<div class="flex items-center relative">
 							<Tooltip
 								content={webConfig.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL
-									? 'Inject the entire content as context for comprehensive processing, this is recommended for complex queries.'
-									: 'Default to segmented retrieval for focused and relevant content extraction, this is recommended for most cases.'}
+									? $i18n.t(
+											'Inject the entire content as context for comprehensive processing, this is recommended for complex queries.'
+										)
+									: $i18n.t(
+											'Default to segmented retrieval for focused and relevant content extraction, this is recommended for most cases.'
+										)}
 							>
 								<Switch bind:state={webConfig.BYPASS_WEB_SEARCH_EMBEDDING_AND_RETRIEVAL} />
 							</Tooltip>
@@ -476,11 +507,13 @@
 						</div>
 						<div class="flex items-center relative">
 							<Tooltip
-								content={webConfig.search.trust_env
-									? 'Use proxy designated by http_proxy and https_proxy environment variables to fetch page contents'
-									: 'Use no proxy to fetch page contents.'}
+								content={webConfig.WEB_SEARCH_TRUST_ENV
+									? $i18n.t(
+											'Use proxy designated by http_proxy and https_proxy environment variables to fetch page contents.'
+										)
+									: $i18n.t('Use no proxy to fetch page contents.')}
 							>
-								<Switch bind:state={webConfig.search.trust_env} />
+								<Switch bind:state={webConfig.WEB_SEARCH_TRUST_ENV} />
 							</Tooltip>
 						</div>
 					</div>
@@ -493,13 +526,134 @@
 
 					<div class="  mb-2.5 flex w-full justify-between">
 						<div class=" self-center text-xs font-medium">
-							{$i18n.t('Bypass SSL verification for Websites')}
+							{$i18n.t('Web Loader Engine')}
 						</div>
 						<div class="flex items-center relative">
-							<Switch bind:state={webConfig.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION} />
+							<select
+								class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
+								bind:value={webConfig.WEB_LOADER_ENGINE}
+								placeholder={$i18n.t('Select a engine')}
+							>
+								<option value="">{$i18n.t('Default')}</option>
+								{#each webLoaderEngines as engine}
+									<option value={engine}>{engine}</option>
+								{/each}
+							</select>
 						</div>
 					</div>
 
+					{#if webConfig.WEB_LOADER_ENGINE === '' || webConfig.WEB_LOADER_ENGINE === 'safe_web'}
+						<div class="  mb-2.5 flex w-full justify-between">
+							<div class=" self-center text-xs font-medium">
+								{$i18n.t('Verify SSL Certificate')}
+							</div>
+							<div class="flex items-center relative">
+								<Switch bind:state={webConfig.ENABLE_WEB_LOADER_SSL_VERIFICATION} />
+							</div>
+						</div>
+					{:else if webConfig.WEB_LOADER_ENGINE === 'playwright'}
+						<div class="mb-2.5 flex w-full flex-col">
+							<div>
+								<div class=" self-center text-xs font-medium mb-1">
+									{$i18n.t('Playwright WebSocket URL')}
+								</div>
+
+								<div class="flex w-full">
+									<div class="flex-1">
+										<input
+											class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+											type="text"
+											placeholder={$i18n.t('Enter Playwright WebSocket URL')}
+											bind:value={webConfig.PLAYWRIGHT_WS_URL}
+											autocomplete="off"
+										/>
+									</div>
+								</div>
+							</div>
+
+							<div class="mt-2">
+								<div class=" self-center text-xs font-medium mb-1">
+									{$i18n.t('Playwright Timeout (ms)')}
+								</div>
+
+								<div class="flex w-full">
+									<div class="flex-1">
+										<input
+											class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+											placeholder={$i18n.t('Enter Playwright Timeout')}
+											bind:value={webConfig.PLAYWRIGHT_TIMEOUT}
+											autocomplete="off"
+										/>
+									</div>
+								</div>
+							</div>
+						</div>
+					{:else if webConfig.WEB_LOADER_ENGINE === 'firecrawl'}
+						<div class="mb-2.5 flex w-full flex-col">
+							<div>
+								<div class=" self-center text-xs font-medium mb-1">
+									{$i18n.t('Firecrawl API Base URL')}
+								</div>
+
+								<div class="flex w-full">
+									<div class="flex-1">
+										<input
+											class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+											type="text"
+											placeholder={$i18n.t('Enter Firecrawl API Base URL')}
+											bind:value={webConfig.FIRECRAWL_API_BASE_URL}
+											autocomplete="off"
+										/>
+									</div>
+								</div>
+							</div>
+
+							<div class="mt-2">
+								<div class=" self-center text-xs font-medium mb-1">
+									{$i18n.t('Firecrawl API Key')}
+								</div>
+
+								<SensitiveInput
+									placeholder={$i18n.t('Enter Firecrawl API Key')}
+									bind:value={webConfig.FIRECRAWL_API_KEY}
+								/>
+							</div>
+						</div>
+					{:else if webConfig.WEB_LOADER_ENGINE === 'tavily'}
+						<div class="mb-2.5 flex w-full flex-col">
+							<div>
+								<div class=" self-center text-xs font-medium mb-1">
+									{$i18n.t('Tavily Extract Depth')}
+								</div>
+
+								<div class="flex w-full">
+									<div class="flex-1">
+										<input
+											class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+											type="text"
+											placeholder={$i18n.t('Enter Tavily Extract Depth')}
+											bind:value={webConfig.TAVILY_EXTRACT_DEPTH}
+											autocomplete="off"
+										/>
+									</div>
+								</div>
+							</div>
+
+							{#if webConfig.WEB_SEARCH_ENGINE !== 'tavily'}
+								<div class="mt-2">
+									<div class=" self-center text-xs font-medium mb-1">
+										{$i18n.t('Tavily API Key')}
+									</div>
+
+									<SensitiveInput
+										placeholder={$i18n.t('Enter Tavily API Key')}
+										bind:value={webConfig.TAVILY_API_KEY}
+									/>
+								</div>
+							{/if}
+						</div>
+					{/if}
+
 					<div class="  mb-2.5 flex w-full justify-between">
 						<div class=" self-center text-xs font-medium">
 							{$i18n.t('Youtube Language')}
@@ -509,7 +663,7 @@
 								class="flex-1 w-full rounded-lg text-sm bg-transparent outline-hidden"
 								type="text"
 								placeholder={$i18n.t('Enter language codes')}
-								bind:value={youtubeLanguage}
+								bind:value={webConfig.YOUTUBE_LOADER_LANGUAGE}
 								autocomplete="off"
 							/>
 						</div>
@@ -524,7 +678,7 @@
 								class="flex-1 w-full rounded-lg text-sm bg-transparent outline-hidden"
 								type="text"
 								placeholder={$i18n.t('Enter proxy URL (e.g. https://user:password@host:port)')}
-								bind:value={youtubeProxyUrl}
+								bind:value={webConfig.YOUTUBE_LOADER_PROXY_URL}
 								autocomplete="off"
 							/>
 						</div>

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

@@ -52,14 +52,23 @@
 			prompts: false,
 			tools: false
 		},
+		sharing: {
+			public_models: false,
+			public_knowledge: false,
+			public_prompts: false,
+			public_tools: false
+		},
 		chat: {
 			controls: true,
 			file_upload: true,
 			delete: true,
 			edit: true,
-			temporary: true
+			multiple_models: true,
+			temporary: true,
+			temporary_enforced: false
 		},
 		features: {
+			direct_tool_servers: false,
 			web_search: true,
 			image_generation: true,
 			code_interpreter: true

src/lib/components/admin/Users/Groups/EditGroupModal.svelte

@@ -9,6 +9,7 @@
 	import Users from './Users.svelte';
 	import UserPlusSolid from '$lib/components/icons/UserPlusSolid.svelte';
 	import WrenchSolid from '$lib/components/icons/WrenchSolid.svelte';
+	import ConfirmDialog from '$lib/components/common/ConfirmDialog.svelte';
 
 	export let onSubmit: Function = () => {};
 	export let onDelete: Function = () => {};
@@ -25,6 +26,7 @@
 
 	let selectedTab = 'general';
 	let loading = false;
+	let showDeleteConfirmDialog = false;
 
 	export let name = '';
 	export let description = '';
@@ -36,6 +38,12 @@
 			prompts: false,
 			tools: false
 		},
+		sharing: {
+			public_models: false,
+			public_knowledge: false,
+			public_prompts: false,
+			public_tools: false
+		},
 		chat: {
 			controls: true,
 			file_upload: true,
@@ -44,6 +52,7 @@
 			temporary: true
 		},
 		features: {
+			direct_tool_servers: false,
 			web_search: true,
 			image_generation: true,
 			code_interpreter: true
@@ -88,6 +97,14 @@
 	});
 </script>
 
+<ConfirmDialog
+	bind:show={showDeleteConfirmDialog}
+	on:confirm={() => {
+		onDelete();
+		show = false;
+	}}
+/>
+
 <Modal size="md" bind:show>
 	<div>
 		<div class=" flex justify-between dark:text-gray-100 px-5 pt-4 mb-1.5">
@@ -263,18 +280,19 @@
 						{/if}
 					</div> -->
 
-					<div class="flex justify-end pt-3 text-sm font-medium gap-1.5">
+					<div class="flex justify-between pt-3 text-sm font-medium gap-1.5">
 						{#if edit}
 							<button
 								class="px-3.5 py-1.5 text-sm font-medium dark:bg-black dark:hover:bg-gray-900 dark:text-white bg-white text-black hover:bg-gray-100 transition rounded-full flex flex-row space-x-1 items-center"
 								type="button"
 								on:click={() => {
-									onDelete();
-									show = false;
+									showDeleteConfirmDialog = true;
 								}}
 							>
 								{$i18n.t('Delete')}
 							</button>
+						{:else}
+							<div></div>
 						{/if}
 
 						<button