Merge pull request #10459 from open-webui/dev

0.5.16
doc: changelog
2025-02-20 11:26:36 -08:00 · 2025-02-20 11:25:17 -08:00 · 2025-02-20 11:19:18 -08:00 · 2025-02-20 11:16:47 -08:00 · 2025-02-20 11:02:45 -08:00 · 2025-02-21 01:46:17 +08:00
319 changed files with 18342 additions and 7435 deletions
@@ -19,7 +19,7 @@ jobs:
        uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
-          node-version: 18
+          node-version: 22
      - uses: actions/setup-python@v5
        with:
          python-version: 3.11
@@ -5,6 +5,151 @@ 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).

+## [0.5.16] - 2025-02-20
+
+### Fixed
+
+- **🔍 Web Search Retrieval Restored**: Resolved a critical issue that broke web search retrieval by reverting deduplication changes, ensuring complete and accurate search results once again.
+
+## [0.5.15] - 2025-02-20
+
+### Added
+
+- **📄 Full Context Mode for Local Document Search (RAG)**: Toggle full context mode from Admin Settings > Documents to inject entire document content into context, improving accuracy for models with large context windows—ideal for deep context understanding.
+- **🌍 Smarter Web Search with Agentic Workflows**: Web searches now intelligently gather and refine multiple relevant terms, similar to RAG handling, delivering significantly better search results for more accurate information retrieval.
+- **🔎 Experimental Playwright Support for Web Loader**: Web content retrieval is taken to the next level with Playwright-powered scraping for enhanced accuracy in extracted web data.
+- **☁️ Experimental Azure Storage Provider**: Early-stage support for Azure Storage allows more cloud storage flexibility directly within Open WebUI.
+- **📊 Improved Jupyter Code Execution with Plots**: Interactive coding now properly displays inline plots, making data visualization more seamless inside chat interactions.
+- **⏳ Adjustable Execution Timeout for Jupyter Interpreter**: Customize execution timeout (default: 60s) for Jupyter-based code execution, allowing longer or more constrained execution based on your needs.
+- **▶️ "Running..." Indicator for Jupyter Code Execution**: A visual indicator now appears while code execution is in progress, providing real-time status updates on ongoing computations.
+- **⚙️ General Backend & Frontend Stability Enhancements**: Extensive refactoring improves reliability, performance, and overall user experience for a more seamless Open WebUI.
+- **🌍 Translation Updates**: Various international translation refinements ensure better localization and a more natural user interface experience.
+
+### Fixed
+
+- **📱 Mobile Hover Issue Resolved**: Users can now edit responses smoothly on mobile without interference, fixing a longstanding hover issue.
+- **🔄 Temporary Chat Message Duplication Fixed**: Eliminated buggy behavior where messages were being unnecessarily repeated in temporary chat mode, ensuring a smooth and consistent conversation flow.
+
+## [0.5.14] - 2025-02-17
+
+### Fixed
+
+- **🔧 Critical Import Error Resolved**: Fixed a circular import issue preventing 'override_static' from being correctly imported in 'open_webui.config', ensuring smooth system initialization and stability.
+
+## [0.5.13] - 2025-02-17
+
+### Added
+
+- **🌐 Full Context Mode for Web Search**: Enable highly accurate web searches by utilizing full context mode—ideal for models with large context windows, ensuring more precise and insightful results.
+- **⚡ Optimized Asynchronous Web Search**: Web searches now load significantly faster with optimized async support, providing users with quicker, more efficient information retrieval.
+- **🔄 Auto Text Direction for RTL Languages**: Automatic text alignment based on language input, ensuring seamless conversation flow for Arabic, Hebrew, and other right-to-left scripts.
+- **🚀 Jupyter Notebook Support for Code Execution**: The "Run" button in code blocks can now use Jupyter for execution, offering a powerful, dynamic coding experience directly in the chat.
+- **🗑️ Message Delete Confirmation Dialog**: Prevent accidental deletions with a new confirmation prompt before removing messages, adding an additional layer of security to your chat history.
+- **📥 Download Button for SVG Diagrams**: SVG diagrams generated within chat can now be downloaded instantly, making it easier to save and share complex visual data.
+- **✨ General UI/UX Improvements and Backend Stability**: A refined interface with smoother interactions, improved layouts, and backend stability enhancements for a more reliable, polished experience.
+
+### Fixed
+
+- **🛠️ Temporary Chat Message Continue Button Fixed**: The "Continue Response" button for temporary chats now works as expected, ensuring an uninterrupted conversation flow.
+
+### Changed
+
+- **📝 Prompt Variable Update**: Deprecated square bracket '[]' indicators for prompt variables; now requires double curly brackets '{{}}' for consistency and clarity.
+- **🔧 Stability Enhancements**: Error handling improved in chat history, ensuring smoother operations when reviewing previous messages.
+
+## [0.5.12] - 2025-02-13
+
+### Added
+
+- **🛠️ Multiple Tool Calls Support for Native Function Mode**: Functions now can call multiple tools within a single response, unlocking better automation and workflow flexibility when using native function calling.
+
+### Fixed
+
+- **📝 Playground Text Completion Restored**: Addressed an issue where text completion in the Playground was not functioning.
+- **🔗 Direct Connections Now Work for Regular Users**: Fixed a bug where users with the 'user' role couldn't establish direct API connections, enabling seamless model usage for all user tiers.
+- **⚡ Landing Page Input No Longer Lags with Long Text**: Improved input responsiveness on the landing page, ensuring fast and smooth typing experiences even when entering long messages.
+- **🔧 Parameter in Functions Fixed**: Fixed an issue where the reserved parameters wasn’t recognized within functions, restoring full functionality for advanced task-based automation.
+
+## [0.5.11] - 2025-02-13
+
+### Added
+
+- **🎤 Kokoro-JS TTS Support**: A new on-device, high-quality text-to-speech engine has been integrated, vastly improving voice generation quality—everything runs directly in your browser.
+- **🐍 Jupyter Notebook Support in Code Interpreter**: Now, you can configure Code Interpreter to run Python code not only via Pyodide but also through Jupyter, offering a more robust coding environment for AI-driven computations and analysis.
+- **🔗 Direct API Connections for Private & Local Inference**: You can now connect Open WebUI to your private or localhost API inference endpoints. CORS must be enabled, but this unlocks direct, on-device AI infrastructure support.
+- **🔍 Advanced Domain Filtering for Web Search**: You can now specify which domains should be included or excluded from web searches, refining results for more relevant information retrieval.
+- **🚀 Improved Image Generation Metadata Handling**: Generated images now retain metadata for better organization and future retrieval.
+- **📂 S3 Key Prefix Support**: Fine-grained control over S3 storage file structuring with configurable key prefixes.
+- **📸 Support for Image-Only Messages**: Send messages containing only images, facilitating more visual-centric interactions.
+- **🌍 Updated Translations**: German, Spanish, Traditional Chinese, and Catalan translations updated for better multilingual support.
+
+### Fixed
+
+- **🔧 OAuth Debug Logs & Username Claim Fixes**: Debug logs have been added for OAuth role and group management, with fixes ensuring proper OAuth username retrieval and claim handling.
+- **📌 Citations Formatting & Toggle Fixes**: Inline citation toggles now function correctly, and citations with more than three sources are now fully visible when expanded.
+- **📸 ComfyUI Maximum Seed Value Constraint Fixed**: The maximum allowed seed value for ComfyUI has been corrected, preventing unintended behavior.
+- **🔑 Connection Settings Stability**: Addressed connection settings issues that were causing instability when saving configurations.
+- **📂 GGUF Model Upload Stability**: Fixed upload inconsistencies for GGUF models, ensuring reliable local model handling.
+- **🔧 Web Search Configuration Bug**: Fixed issues where web search filters and settings weren't correctly applied.
+- **💾 User Settings Persistence Fix**: Ensured user-specific settings are correctly saved and applied across sessions.
+- **🔄 OpenID Username Retrieval Enhancement**: Usernames are now correctly picked up and assigned for OpenID Connect (OIDC) logins.
+
+## [0.5.10] - 2025-02-05
+
+### Fixed
+
+- **⚙️ System Prompts Now Properly Templated via API**: Resolved an issue where system prompts were not being correctly processed when used through the API, ensuring template variables now function as expected.
+- **📝 '<thinking>' Tag Display Issue Fixed**: Fixed a bug where the 'thinking' tag was disrupting content rendering, ensuring clean and accurate text display.
+- **💻 Code Interpreter Stability with Custom Functions**: Addressed failures when using the Code Interpreter with certain custom functions like Anthropic, ensuring smoother execution and better compatibility.
+
+## [0.5.9] - 2025-02-05
+
+### Fixed
+
+- **💡 "Think" Tag Display Issue**: Resolved a bug where the "Think" tag was not functioning correctly, ensuring proper visualization of the model's reasoning process before delivering responses.
+
+## [0.5.8] - 2025-02-05
+
+### Added
+
+- **🖥️ Code Interpreter**: Models can now execute code in real time to refine their answers dynamically, running securely within a sandboxed browser environment using Pyodide. Perfect for calculations, data analysis, and AI-assisted coding tasks!
+- **💬 Redesigned Chat Input UI**: Enjoy a sleeker and more intuitive message input with improved feature selection, making it easier than ever to toggle tools, enable search, and interact with AI seamlessly.
+- **🛠️ Native Tool Calling Support (Experimental)**: Supported models can now call tools natively, reducing query latency and improving contextual responses. More enhancements coming soon!
+- **🔗 Exa Search Engine Integration**: A new search provider has been added, allowing users to retrieve up-to-date and relevant information without leaving the chat interface.
+- **🌍 Localized Dates & Times**: Date and time formats now match your system locale, ensuring a more natural, region-specific experience.
+- **📎 User Headers for External Embedding APIs**: API calls to external embedding services now include user-related headers.
+- **🌍 "Always On" Web Search Toggle**: A new option under Settings > Interface allows users to enable Web Search by default—transform Open WebUI into your go-to search engine, ensuring AI-powered results with every query.
+- **🚀 General Performance & Stability**: Significant improvements across the platform for a faster, more reliable experience.
+- **🖼️ UI/UX Enhancements**: Numerous design refinements improving readability, responsiveness, and accessibility.
+- **🌍 Improved Translations**: Chinese, Korean, French, Ukrainian and Serbian translations have been updated with refined terminologies for better clarity.
+
+### Fixed
+
+- **🔄 OAuth Name Field Fallback**: Resolves OAuth login failures by using the email field as a fallback when a name is missing.
+- **🔑 Google Drive Credentials Restriction**: Ensures only authenticated users can access Google Drive credentials for enhanced security.
+- **🌐 DuckDuckGo Search Rate Limit Handling**: Fixes issues where users would encounter 202 errors due to rate limits when using DuckDuckGo for web search.
+- **📁 File Upload Permission Indicator**: Users are now notified when they lack permission to upload files, improving clarity on system restrictions.
+- **🔧 Max Tokens Issue**: Fixes cases where 'max_tokens' were not applied correctly, ensuring proper model behavior.
+- **🔍 Validation for RAG Web Search URLs**: Filters out invalid or unsupported URLs when using web-based retrieval augmentation.
+- **🖋️ Title Generation Bug**: Fixes inconsistencies in title generation, ensuring proper chat organization.
+
+### Removed
+
+- **⚡ Deprecated Non-Web Worker Pyodide Execution**: Moves entirely to browser sandboxing for better performance and security.
+
+## [0.5.7] - 2025-01-23
+
+### Added
+
+- **🌍 Enhanced Internationalization (i18n)**: Refined and expanded translations for greater global accessibility and a smoother experience for international users.
+
+### Fixed
+
+- **🔗 Connection Model ID Resolution**: Resolved an issue preventing model IDs from registering in connections.
+- **💡 Prefix ID for Ollama Connections**: Fixed a bug where prefix IDs in Ollama connections were non-functional.
+- **🔧 Ollama Model Enable/Disable Functionality**: Addressed the issue of enable/disable toggles not working for Ollama base models.
+- **🔒 RBAC Permissions for Tools and Models**: Corrected incorrect Role-Based Access Control (RBAC) permissions for tools and models, ensuring that users now only access features according to their assigned privileges, enhancing security and role clarity.
+
 ## [0.5.6] - 2025-01-22

 ### Added
@@ -13,10 +13,15 @@

 **Open WebUI is an [extensible](https://docs.openwebui.com/features/plugin/), feature-rich, and user-friendly self-hosted AI platform designed to operate entirely offline.** It supports various LLM runners like **Ollama** and **OpenAI-compatible APIs**, with **built-in inference engine** for RAG, making it a **powerful AI deployment solution**.

-For more information, be sure to check out our [Open WebUI Documentation](https://docs.openwebui.com/).
-
 ![Open WebUI Demo](./demo.gif)

+> [!TIP]  
+> **Looking for an [Enterprise Plan](https://docs.openwebui.com/enterprise)?** – **[Speak with Our Sales Team Today!](mailto:sales@openwebui.com)**
+>
+> Get **enhanced capabilities**, including **custom theming and branding**, **Service Level Agreement (SLA) support**, **Long-Term Support (LTS) versions**, and **more!**
+
+For more information, be sure to check out our [Open WebUI Documentation](https://docs.openwebui.com/).
+
 ## Key Features of Open WebUI ⭐

 - 🚀 **Effortless Setup**: Install seamlessly using Docker or Kubernetes (kubectl, kustomize or helm) for a hassle-free experience with support for both `:ollama` and `:cuda` tagged images.
@@ -174,7 +179,7 @@ docker run --rm --volume /var/run/docker.sock:/var/run/docker.sock containrrr/wa

 In the last part of the command, replace `open-webui` with your container name if it is different.

-Check our Migration Guide available in our [Open WebUI Documentation](https://docs.openwebui.com/tutorials/migration/).
+Check our Updating Guide available in our [Open WebUI Documentation](https://docs.openwebui.com/getting-started/updating).

 ### Using the Dev Branch 🌙

@@ -2,6 +2,8 @@ import json
 import logging
 import os
 import shutil
+import base64
+
 from datetime import datetime
 from pathlib import Path
 from typing import Generic, Optional, TypeVar
@@ -593,8 +595,6 @@ if frontend_favicon.exists():
        shutil.copyfile(frontend_favicon, STATIC_DIR / "favicon.png")
    except Exception as e:
        logging.error(f"An error occurred: {e}")
-else:
-    logging.warning(f"Frontend favicon not found at {frontend_favicon}")

 frontend_splash = FRONTEND_BUILD_DIR / "static" / "splash.png"

@@ -603,12 +603,18 @@ if frontend_splash.exists():
        shutil.copyfile(frontend_splash, STATIC_DIR / "splash.png")
    except Exception as e:
        logging.error(f"An error occurred: {e}")
-else:
-    logging.warning(f"Frontend splash not found at {frontend_splash}")
+
+frontend_loader = FRONTEND_BUILD_DIR / "static" / "loader.js"
+
+if frontend_loader.exists():
+    try:
+        shutil.copyfile(frontend_loader, STATIC_DIR / "loader.js")
+    except Exception as e:
+        logging.error(f"An error occurred: {e}")


 ####################################
-# CUSTOM_NAME
+# CUSTOM_NAME (Legacy)
 ####################################

 CUSTOM_NAME = os.environ.get("CUSTOM_NAME", "")
@@ -650,6 +656,16 @@ if CUSTOM_NAME:
        pass


+####################################
+# LICENSE_KEY
+####################################
+
+LICENSE_KEY = PersistentConfig(
+    "LICENSE_KEY",
+    "license.key",
+    os.environ.get("LICENSE_KEY", ""),
+)
+
 ####################################
 # STORAGE PROVIDER
 ####################################
@@ -660,6 +676,7 @@ S3_ACCESS_KEY_ID = os.environ.get("S3_ACCESS_KEY_ID", None)
 S3_SECRET_ACCESS_KEY = os.environ.get("S3_SECRET_ACCESS_KEY", None)
 S3_REGION_NAME = os.environ.get("S3_REGION_NAME", None)
 S3_BUCKET_NAME = os.environ.get("S3_BUCKET_NAME", None)
+S3_KEY_PREFIX = os.environ.get("S3_KEY_PREFIX", None)
 S3_ENDPOINT_URL = os.environ.get("S3_ENDPOINT_URL", None)

 GCS_BUCKET_NAME = os.environ.get("GCS_BUCKET_NAME", None)
@@ -667,6 +684,10 @@ GOOGLE_APPLICATION_CREDENTIALS_JSON = os.environ.get(
    "GOOGLE_APPLICATION_CREDENTIALS_JSON", None
 )

+AZURE_STORAGE_ENDPOINT = os.environ.get("AZURE_STORAGE_ENDPOINT", None)
+AZURE_STORAGE_CONTAINER_NAME = os.environ.get("AZURE_STORAGE_CONTAINER_NAME", None)
+AZURE_STORAGE_KEY = os.environ.get("AZURE_STORAGE_KEY", None)
+
 ####################################
 # File Upload DIR
 ####################################
@@ -682,6 +703,17 @@ Path(UPLOAD_DIR).mkdir(parents=True, exist_ok=True)
 CACHE_DIR = f"{DATA_DIR}/cache"
 Path(CACHE_DIR).mkdir(parents=True, exist_ok=True)

+
+####################################
+# DIRECT CONNECTIONS
+####################################
+
+ENABLE_DIRECT_CONNECTIONS = PersistentConfig(
+    "ENABLE_DIRECT_CONNECTIONS",
+    "direct.enable",
+    os.environ.get("ENABLE_DIRECT_CONNECTIONS", "True").lower() == "true",
+)
+
 ####################################
 # OLLAMA_BASE_URL
 ####################################
@@ -755,6 +787,9 @@ ENABLE_OPENAI_API = PersistentConfig(
 OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY", "")
 OPENAI_API_BASE_URL = os.environ.get("OPENAI_API_BASE_URL", "")

+GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY", "")
+GEMINI_API_BASE_URL = os.environ.get("GEMINI_API_BASE_URL", "")
+

 if OPENAI_API_BASE_URL == "":
    OPENAI_API_BASE_URL = "https://api.openai.com/v1"
@@ -927,6 +962,12 @@ USER_PERMISSIONS_FEATURES_IMAGE_GENERATION = (
    == "true"
 )

+USER_PERMISSIONS_FEATURES_CODE_INTERPRETER = (
+    os.environ.get("USER_PERMISSIONS_FEATURES_CODE_INTERPRETER", "True").lower()
+    == "true"
+)
+
+
 DEFAULT_USER_PERMISSIONS = {
    "workspace": {
        "models": USER_PERMISSIONS_WORKSPACE_MODELS_ACCESS,
@@ -944,6 +985,7 @@ DEFAULT_USER_PERMISSIONS = {
    "features": {
        "web_search": USER_PERMISSIONS_FEATURES_WEB_SEARCH,
        "image_generation": USER_PERMISSIONS_FEATURES_IMAGE_GENERATION,
+        "code_interpreter": USER_PERMISSIONS_FEATURES_CODE_INTERPRETER,
    },
 }

@@ -1094,21 +1136,27 @@ TITLE_GENERATION_PROMPT_TEMPLATE = PersistentConfig(
    os.environ.get("TITLE_GENERATION_PROMPT_TEMPLATE", ""),
 )

-DEFAULT_TITLE_GENERATION_PROMPT_TEMPLATE = """Create a concise, 3-5 word title with an emoji as a title for the chat history, in the given language. Suitable Emojis for the summary can be used to enhance understanding but avoid quotation marks or special formatting. RESPOND ONLY WITH THE TITLE TEXT.
-
-Examples of titles:
-📉 Stock Market Trends
-🍪 Perfect Chocolate Chip Recipe
-Evolution of Music Streaming
-Remote Work Productivity Tips
-Artificial Intelligence in Healthcare
-🎮 Video Game Development Insights
-
+DEFAULT_TITLE_GENERATION_PROMPT_TEMPLATE = """### Task:
+Generate a concise, 3-5 word title with an emoji summarizing the chat history.
+### Guidelines:
+- The title should clearly represent the main theme or subject of the conversation.
+- Use emojis that enhance understanding of the topic, but avoid quotation marks or special formatting.
+- Write the title in the chat's primary language; default to English if multilingual.
+- Prioritize accuracy over excessive creativity; keep it clear and simple.
+### Output:
+JSON format: { "title": "your concise title here" }
+### Examples:
+- { "title": "📉 Stock Market Trends" },
+- { "title": "🍪 Perfect Chocolate Chip Recipe" },
+- { "title": "Evolution of Music Streaming" },
+- { "title": "Remote Work Productivity Tips" },
+- { "title": "Artificial Intelligence in Healthcare" },
+- { "title": "🎮 Video Game Development Insights" }
+### Chat History:
 <chat_history>
 {{MESSAGES:END:2}}
 </chat_history>"""

-
 TAGS_GENERATION_PROMPT_TEMPLATE = PersistentConfig(
    "TAGS_GENERATION_PROMPT_TEMPLATE",
    "task.tags.prompt_template",
@@ -1165,6 +1213,12 @@ ENABLE_TAGS_GENERATION = PersistentConfig(
    os.environ.get("ENABLE_TAGS_GENERATION", "True").lower() == "true",
 )

+ENABLE_TITLE_GENERATION = PersistentConfig(
+    "ENABLE_TITLE_GENERATION",
+    "task.title.enable",
+    os.environ.get("ENABLE_TITLE_GENERATION", "True").lower() == "true",
+)
+

 ENABLE_SEARCH_QUERY_GENERATION = PersistentConfig(
    "ENABLE_SEARCH_QUERY_GENERATION",
@@ -1277,7 +1331,28 @@ TOOLS_FUNCTION_CALLING_PROMPT_TEMPLATE = PersistentConfig(
 )


-DEFAULT_TOOLS_FUNCTION_CALLING_PROMPT_TEMPLATE = """Available Tools: {{TOOLS}}\nReturn an empty string if no tools match the query. If a function tool matches, construct and return a JSON object in the format {\"name\": \"functionName\", \"parameters\": {\"requiredFunctionParamKey\": \"requiredFunctionParamValue\"}} using the appropriate tool and its parameters. Only return the object and limit the response to the JSON object without additional text."""
+DEFAULT_TOOLS_FUNCTION_CALLING_PROMPT_TEMPLATE = """Available Tools: {{TOOLS}}
+
+Your task is to choose and return the correct tool(s) from the list of available tools based on the query. Follow these guidelines:
+
+- Return only the JSON object, without any additional text or explanation.
+
+- If no tools match the query, return an empty array: 
+   {
+     "tool_calls": []
+   }
+
+- If one or more tools match the query, construct a JSON response containing a "tool_calls" array with objects that include:
+   - "name": The tool's name.
+   - "parameters": A dictionary of required parameters and their corresponding values.
+
+The format for the JSON response is strictly:
+{
+  "tool_calls": [
+    {"name": "toolName1", "parameters": {"key1": "value1"}},
+    {"name": "toolName2", "parameters": {"key2": "value2"}}
+  ]
+}"""


 DEFAULT_EMOJI_GENERATION_PROMPT_TEMPLATE = """Your task is to reflect the speaker's likely facial expression through a fitting emoji. Interpret emotions from the message and reflect their facial expression using fitting, diverse emojis (e.g., 😊, 😢, 😡, 😱).
@@ -1290,6 +1365,131 @@ Your task is to synthesize these responses into a single, high-quality response.

 Responses from models: {{responses}}"""

+
+####################################
+# Code Interpreter
+####################################
+
+
+CODE_EXECUTION_ENGINE = PersistentConfig(
+    "CODE_EXECUTION_ENGINE",
+    "code_execution.engine",
+    os.environ.get("CODE_EXECUTION_ENGINE", "pyodide"),
+)
+
+CODE_EXECUTION_JUPYTER_URL = PersistentConfig(
+    "CODE_EXECUTION_JUPYTER_URL",
+    "code_execution.jupyter.url",
+    os.environ.get("CODE_EXECUTION_JUPYTER_URL", ""),
+)
+
+CODE_EXECUTION_JUPYTER_AUTH = PersistentConfig(
+    "CODE_EXECUTION_JUPYTER_AUTH",
+    "code_execution.jupyter.auth",
+    os.environ.get("CODE_EXECUTION_JUPYTER_AUTH", ""),
+)
+
+CODE_EXECUTION_JUPYTER_AUTH_TOKEN = PersistentConfig(
+    "CODE_EXECUTION_JUPYTER_AUTH_TOKEN",
+    "code_execution.jupyter.auth_token",
+    os.environ.get("CODE_EXECUTION_JUPYTER_AUTH_TOKEN", ""),
+)
+
+
+CODE_EXECUTION_JUPYTER_AUTH_PASSWORD = PersistentConfig(
+    "CODE_EXECUTION_JUPYTER_AUTH_PASSWORD",
+    "code_execution.jupyter.auth_password",
+    os.environ.get("CODE_EXECUTION_JUPYTER_AUTH_PASSWORD", ""),
+)
+
+CODE_EXECUTION_JUPYTER_TIMEOUT = PersistentConfig(
+    "CODE_EXECUTION_JUPYTER_TIMEOUT",
+    "code_execution.jupyter.timeout",
+    int(os.environ.get("CODE_EXECUTION_JUPYTER_TIMEOUT", "60")),
+)
+
+ENABLE_CODE_INTERPRETER = PersistentConfig(
+    "ENABLE_CODE_INTERPRETER",
+    "code_interpreter.enable",
+    os.environ.get("ENABLE_CODE_INTERPRETER", "True").lower() == "true",
+)
+
+CODE_INTERPRETER_ENGINE = PersistentConfig(
+    "CODE_INTERPRETER_ENGINE",
+    "code_interpreter.engine",
+    os.environ.get("CODE_INTERPRETER_ENGINE", "pyodide"),
+)
+
+CODE_INTERPRETER_PROMPT_TEMPLATE = PersistentConfig(
+    "CODE_INTERPRETER_PROMPT_TEMPLATE",
+    "code_interpreter.prompt_template",
+    os.environ.get("CODE_INTERPRETER_PROMPT_TEMPLATE", ""),
+)
+
+CODE_INTERPRETER_JUPYTER_URL = PersistentConfig(
+    "CODE_INTERPRETER_JUPYTER_URL",
+    "code_interpreter.jupyter.url",
+    os.environ.get(
+        "CODE_INTERPRETER_JUPYTER_URL", os.environ.get("CODE_EXECUTION_JUPYTER_URL", "")
+    ),
+)
+
+CODE_INTERPRETER_JUPYTER_AUTH = PersistentConfig(
+    "CODE_INTERPRETER_JUPYTER_AUTH",
+    "code_interpreter.jupyter.auth",
+    os.environ.get(
+        "CODE_INTERPRETER_JUPYTER_AUTH",
+        os.environ.get("CODE_EXECUTION_JUPYTER_AUTH", ""),
+    ),
+)
+
+CODE_INTERPRETER_JUPYTER_AUTH_TOKEN = PersistentConfig(
+    "CODE_INTERPRETER_JUPYTER_AUTH_TOKEN",
+    "code_interpreter.jupyter.auth_token",
+    os.environ.get(
+        "CODE_INTERPRETER_JUPYTER_AUTH_TOKEN",
+        os.environ.get("CODE_EXECUTION_JUPYTER_AUTH_TOKEN", ""),
+    ),
+)
+
+
+CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD = PersistentConfig(
+    "CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD",
+    "code_interpreter.jupyter.auth_password",
+    os.environ.get(
+        "CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD",
+        os.environ.get("CODE_EXECUTION_JUPYTER_AUTH_PASSWORD", ""),
+    ),
+)
+
+CODE_INTERPRETER_JUPYTER_TIMEOUT = PersistentConfig(
+    "CODE_INTERPRETER_JUPYTER_TIMEOUT",
+    "code_interpreter.jupyter.timeout",
+    int(
+        os.environ.get(
+            "CODE_INTERPRETER_JUPYTER_TIMEOUT",
+            os.environ.get("CODE_EXECUTION_JUPYTER_TIMEOUT", "60"),
+        )
+    ),
+)
+
+
+DEFAULT_CODE_INTERPRETER_PROMPT = """
+#### Tools Available
+
+1. **Code Interpreter**: `<code_interpreter type="code" lang="python"></code_interpreter>`
+   - You have access to a Python shell that runs directly in the user's browser, enabling fast execution of code for analysis, calculations, or problem-solving.  Use it in this response.
+   - The Python code you write can incorporate a wide array of libraries, handle data manipulation or visualization, perform API calls for web-related tasks, or tackle virtually any computational challenge. Use this flexibility to **think outside the box, craft elegant solutions, and harness Python's full potential**.
+   - To use it, **you must enclose your code within `<code_interpreter type="code" lang="python">` XML tags** and stop right away. If you don't, the code won't execute. Do NOT use triple backticks.
+   - When coding, **always aim to print meaningful outputs** (e.g., results, tables, summaries, or visuals) to better interpret and verify the findings. Avoid relying on implicit outputs; prioritize explicit and clear print statements so the results are effectively communicated to the user.  
+   - After obtaining the printed output, **always provide a concise analysis, interpretation, or next steps to help the user understand the findings or refine the outcome further.**  
+   - If the results are unclear, unexpected, or require validation, refine the code and execute it again as needed. Always aim to deliver meaningful insights from the results, iterating if necessary.  
+   - **If a link to an image, audio, or any file is provided in markdown format in the output, ALWAYS regurgitate word for word, explicitly display it as part of the response to ensure the user can access it easily, do NOT change the link.**
+   - All responses should be communicated in the chat's primary language, ensuring seamless understanding. If the chat is multilingual, default to English for clarity.
+
+Ensure that the tools are effectively utilized to achieve the highest-quality analysis for the user."""
+
+
 ####################################
 # Vector Database
 ####################################
@@ -1319,6 +1519,7 @@ CHROMA_HTTP_SSL = os.environ.get("CHROMA_HTTP_SSL", "false").lower() == "true"

 MILVUS_URI = os.environ.get("MILVUS_URI", f"{DATA_DIR}/vector_db/milvus.db")
 MILVUS_DB = os.environ.get("MILVUS_DB", "default")
+MILVUS_TOKEN = os.environ.get("MILVUS_TOKEN", None)

 # Qdrant
 QDRANT_URI = os.environ.get("QDRANT_URI", None)
@@ -1393,6 +1594,12 @@ ENABLE_RAG_HYBRID_SEARCH = PersistentConfig(
    os.environ.get("ENABLE_RAG_HYBRID_SEARCH", "").lower() == "true",
 )

+RAG_FULL_CONTEXT = PersistentConfig(
+    "RAG_FULL_CONTEXT",
+    "rag.full_context",
+    os.getenv("RAG_FULL_CONTEXT", "False").lower() == "true",
+)
+
 RAG_FILE_MAX_COUNT = PersistentConfig(
    "RAG_FILE_MAX_COUNT",
    "rag.file.max_count",
@@ -1507,7 +1714,7 @@ 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] when a <source_id> tag is explicitly provided in the context.**  
+- **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.  
 - Do not use XML tags in your response.
 - Ensure citations are concise and directly related to the information provided.
@@ -1588,11 +1795,17 @@ RAG_WEB_SEARCH_ENGINE = PersistentConfig(
    os.getenv("RAG_WEB_SEARCH_ENGINE", ""),
 )

+RAG_WEB_SEARCH_FULL_CONTEXT = PersistentConfig(
+    "RAG_WEB_SEARCH_FULL_CONTEXT",
+    "rag.web.search.full_context",
+    os.getenv("RAG_WEB_SEARCH_FULL_CONTEXT", "False").lower() == "true",
+)
+
 # 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",
-    "rag.rag.web.search.domain.filter_list",
+    "rag.web.search.domain.filter_list",
    [
        # "wikipedia.com",
        # "wikimedia.org",
@@ -1637,6 +1850,12 @@ MOJEEK_SEARCH_API_KEY = PersistentConfig(
    os.getenv("MOJEEK_SEARCH_API_KEY", ""),
 )

+BOCHA_SEARCH_API_KEY = PersistentConfig(
+    "BOCHA_SEARCH_API_KEY",
+    "rag.web.search.bocha_search_api_key",
+    os.getenv("BOCHA_SEARCH_API_KEY", ""),
+)
+
 SERPSTACK_API_KEY = PersistentConfig(
    "SERPSTACK_API_KEY",
    "rag.web.search.serpstack_api_key",
@@ -1685,6 +1904,18 @@ SEARCHAPI_ENGINE = PersistentConfig(
    os.getenv("SEARCHAPI_ENGINE", ""),
 )

+SERPAPI_API_KEY = PersistentConfig(
+    "SERPAPI_API_KEY",
+    "rag.web.search.serpapi_api_key",
+    os.getenv("SERPAPI_API_KEY", ""),
+)
+
+SERPAPI_ENGINE = PersistentConfig(
+    "SERPAPI_ENGINE",
+    "rag.web.search.serpapi_engine",
+    os.getenv("SERPAPI_ENGINE", ""),
+)
+
 BING_SEARCH_V7_ENDPOINT = PersistentConfig(
    "BING_SEARCH_V7_ENDPOINT",
    "rag.web.search.bing_search_v7_endpoint",
@@ -1699,6 +1930,11 @@ BING_SEARCH_V7_SUBSCRIPTION_KEY = PersistentConfig(
    os.environ.get("BING_SEARCH_V7_SUBSCRIPTION_KEY", ""),
 )

+EXA_API_KEY = PersistentConfig(
+    "EXA_API_KEY",
+    "rag.web.search.exa_api_key",
+    os.getenv("EXA_API_KEY", ""),
+)

 RAG_WEB_SEARCH_RESULT_COUNT = PersistentConfig(
    "RAG_WEB_SEARCH_RESULT_COUNT",
@@ -1712,6 +1948,35 @@ RAG_WEB_SEARCH_CONCURRENT_REQUESTS = PersistentConfig(
    int(os.getenv("RAG_WEB_SEARCH_CONCURRENT_REQUESTS", "10")),
 )

+RAG_WEB_LOADER_ENGINE = PersistentConfig(
+    "RAG_WEB_LOADER_ENGINE",
+    "rag.web.loader.engine",
+    os.environ.get("RAG_WEB_LOADER_ENGINE", "safe_web"),
+)
+
+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",
+)
+
+PLAYWRIGHT_WS_URI = PersistentConfig(
+    "PLAYWRIGHT_WS_URI",
+    "rag.web.loader.engine.playwright.ws.uri",
+    os.environ.get("PLAYWRIGHT_WS_URI", None),
+)
+
+FIRECRAWL_API_KEY = PersistentConfig(
+    "FIRECRAWL_API_KEY",
+    "firecrawl.api_key",
+    os.environ.get("FIRECRAWL_API_KEY", ""),
+)
+
+FIRECRAWL_API_BASE_URL = PersistentConfig(
+    "FIRECRAWL_API_BASE_URL",
+    "firecrawl.api_url",
+    os.environ.get("FIRECRAWL_API_BASE_URL", "https://api.firecrawl.dev"),
+)

 ####################################
 # Images
@@ -1923,6 +2188,17 @@ IMAGES_OPENAI_API_KEY = PersistentConfig(
    os.getenv("IMAGES_OPENAI_API_KEY", OPENAI_API_KEY),
 )

+IMAGES_GEMINI_API_BASE_URL = PersistentConfig(
+    "IMAGES_GEMINI_API_BASE_URL",
+    "image_generation.gemini.api_base_url",
+    os.getenv("IMAGES_GEMINI_API_BASE_URL", GEMINI_API_BASE_URL),
+)
+IMAGES_GEMINI_API_KEY = PersistentConfig(
+    "IMAGES_GEMINI_API_KEY",
+    "image_generation.gemini.api_key",
+    os.getenv("IMAGES_GEMINI_API_KEY", GEMINI_API_KEY),
+)
+
 IMAGE_SIZE = PersistentConfig(
    "IMAGE_SIZE", "image_generation.size", os.getenv("IMAGE_SIZE", "512x512")
 )
@@ -1954,6 +2230,12 @@ WHISPER_MODEL_AUTO_UPDATE = (
    and os.environ.get("WHISPER_MODEL_AUTO_UPDATE", "").lower() == "true"
 )

+# Add Deepgram configuration
+DEEPGRAM_API_KEY = PersistentConfig(
+    "DEEPGRAM_API_KEY",
+    "audio.stt.deepgram.api_key",
+    os.getenv("DEEPGRAM_API_KEY", ""),
+)

 AUDIO_STT_OPENAI_API_BASE_URL = PersistentConfig(
    "AUDIO_STT_OPENAI_API_BASE_URL",
@@ -57,7 +57,7 @@ class ERROR_MESSAGES(str, Enum):
    )

    FILE_NOT_SENT = "FILE_NOT_SENT"
-    FILE_NOT_SUPPORTED = "Oops! It seems like the file format you're trying to upload is not supported. Please upload a file with a supported format (e.g., JPG, PNG, PDF, TXT) and try again."
+    FILE_NOT_SUPPORTED = "Oops! It seems like the file format you're trying to upload is not supported. Please upload a file with a supported format and try again."

    NOT_FOUND = "We could not find what you're looking for :/"
    USER_NOT_FOUND = "We could not find what you're looking for :/"
@@ -92,6 +92,7 @@ log_sources = [
    "RAG",
    "WEBHOOK",
    "SOCKET",
+    "OAUTH",
 ]

 SRC_LOG_LEVELS = {}
@@ -112,6 +113,7 @@ if WEBUI_NAME != "Open WebUI":

 WEBUI_FAVICON_URL = "https://openwebui.com/favicon.png"

+TRUSTED_SIGNATURE_KEY = os.environ.get("TRUSTED_SIGNATURE_KEY", "")

 ####################################
 # ENV (dev,test,prod)
@@ -356,14 +358,22 @@ WEBUI_SECRET_KEY = os.environ.get(
    ),  # DEPRECATED: remove at next major version
 )

-WEBUI_SESSION_COOKIE_SAME_SITE = os.environ.get(
-    "WEBUI_SESSION_COOKIE_SAME_SITE",
-    os.environ.get("WEBUI_SESSION_COOKIE_SAME_SITE", "lax"),
+WEBUI_SESSION_COOKIE_SAME_SITE = os.environ.get("WEBUI_SESSION_COOKIE_SAME_SITE", "lax")
+
+WEBUI_SESSION_COOKIE_SECURE = (
+    os.environ.get("WEBUI_SESSION_COOKIE_SECURE", "false").lower() == "true"
 )

-WEBUI_SESSION_COOKIE_SECURE = os.environ.get(
-    "WEBUI_SESSION_COOKIE_SECURE",
-    os.environ.get("WEBUI_SESSION_COOKIE_SECURE", "false").lower() == "true",
+WEBUI_AUTH_COOKIE_SAME_SITE = os.environ.get(
+    "WEBUI_AUTH_COOKIE_SAME_SITE", WEBUI_SESSION_COOKIE_SAME_SITE
+)
+
+WEBUI_AUTH_COOKIE_SECURE = (
+    os.environ.get(
+        "WEBUI_AUTH_COOKIE_SECURE",
+        os.environ.get("WEBUI_SESSION_COOKIE_SECURE", "false"),
+    ).lower()
+    == "true"
 )

 if WEBUI_AUTH and WEBUI_SECRET_KEY == "":
@@ -250,7 +250,7 @@ async def generate_function_chat_completion(

        params = model_info.params.model_dump()
        form_data = apply_model_params_to_body_openai(params, form_data)
-        form_data = apply_model_system_prompt_to_body(params, form_data, user)
+        form_data = apply_model_system_prompt_to_body(params, form_data, metadata, user)

    pipe_id = get_pipe_id(form_data)
    function_module = get_function_module_by_id(request, pipe_id)
@@ -88,6 +88,7 @@ from open_webui.models.models import Models
 from open_webui.models.users import UserModel, Users

 from open_webui.config import (
+    LICENSE_KEY,
    # Ollama
    ENABLE_OLLAMA_API,
    OLLAMA_BASE_URLS,
@@ -97,6 +98,23 @@ from open_webui.config import (
    OPENAI_API_BASE_URLS,
    OPENAI_API_KEYS,
    OPENAI_API_CONFIGS,
+    # Direct Connections
+    ENABLE_DIRECT_CONNECTIONS,
+    # Code Execution
+    CODE_EXECUTION_ENGINE,
+    CODE_EXECUTION_JUPYTER_URL,
+    CODE_EXECUTION_JUPYTER_AUTH,
+    CODE_EXECUTION_JUPYTER_AUTH_TOKEN,
+    CODE_EXECUTION_JUPYTER_AUTH_PASSWORD,
+    CODE_EXECUTION_JUPYTER_TIMEOUT,
+    ENABLE_CODE_INTERPRETER,
+    CODE_INTERPRETER_ENGINE,
+    CODE_INTERPRETER_PROMPT_TEMPLATE,
+    CODE_INTERPRETER_JUPYTER_URL,
+    CODE_INTERPRETER_JUPYTER_AUTH,
+    CODE_INTERPRETER_JUPYTER_AUTH_TOKEN,
+    CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD,
+    CODE_INTERPRETER_JUPYTER_TIMEOUT,
    # Image
    AUTOMATIC1111_API_AUTH,
    AUTOMATIC1111_BASE_URL,
@@ -115,6 +133,8 @@ from open_webui.config import (
    IMAGE_STEPS,
    IMAGES_OPENAI_API_BASE_URL,
    IMAGES_OPENAI_API_KEY,
+    IMAGES_GEMINI_API_BASE_URL,
+    IMAGES_GEMINI_API_KEY,
    # Audio
    AUDIO_STT_ENGINE,
    AUDIO_STT_MODEL,
@@ -129,12 +149,18 @@ from open_webui.config import (
    AUDIO_TTS_VOICE,
    AUDIO_TTS_AZURE_SPEECH_REGION,
    AUDIO_TTS_AZURE_SPEECH_OUTPUT_FORMAT,
+    PLAYWRIGHT_WS_URI,
+    FIRECRAWL_API_BASE_URL,
+    FIRECRAWL_API_KEY,
+    RAG_WEB_LOADER_ENGINE,
    WHISPER_MODEL,
+    DEEPGRAM_API_KEY,
    WHISPER_MODEL_AUTO_UPDATE,
    WHISPER_MODEL_DIR,
    # Retrieval
    RAG_TEMPLATE,
    DEFAULT_RAG_TEMPLATE,
+    RAG_FULL_CONTEXT,
    RAG_EMBEDDING_MODEL,
    RAG_EMBEDDING_MODEL_AUTO_UPDATE,
    RAG_EMBEDDING_MODEL_TRUST_REMOTE_CODE,
@@ -162,12 +188,16 @@ from open_webui.config import (
    YOUTUBE_LOADER_PROXY_URL,
    # Retrieval (Web Search)
    RAG_WEB_SEARCH_ENGINE,
+    RAG_WEB_SEARCH_FULL_CONTEXT,
    RAG_WEB_SEARCH_RESULT_COUNT,
    RAG_WEB_SEARCH_CONCURRENT_REQUESTS,
+    RAG_WEB_SEARCH_TRUST_ENV,
    RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
    JINA_API_KEY,
    SEARCHAPI_API_KEY,
    SEARCHAPI_ENGINE,
+    SERPAPI_API_KEY,
+    SERPAPI_ENGINE,
    SEARXNG_QUERY_URL,
    SERPER_API_KEY,
    SERPLY_API_KEY,
@@ -177,8 +207,10 @@ from open_webui.config import (
    BING_SEARCH_V7_ENDPOINT,
    BING_SEARCH_V7_SUBSCRIPTION_KEY,
    BRAVE_SEARCH_API_KEY,
+    EXA_API_KEY,
    KAGI_SEARCH_API_KEY,
    MOJEEK_SEARCH_API_KEY,
+    BOCHA_SEARCH_API_KEY,
    GOOGLE_PSE_API_KEY,
    GOOGLE_PSE_ENGINE_ID,
    GOOGLE_DRIVE_CLIENT_ID,
@@ -251,6 +283,7 @@ from open_webui.config import (
    TASK_MODEL,
    TASK_MODEL_EXTERNAL,
    ENABLE_TAGS_GENERATION,
+    ENABLE_TITLE_GENERATION,
    ENABLE_SEARCH_QUERY_GENERATION,
    ENABLE_RETRIEVAL_QUERY_GENERATION,
    ENABLE_AUTOCOMPLETE_GENERATION,
@@ -297,15 +330,17 @@ from open_webui.utils.middleware import process_chat_payload, process_chat_respo
 from open_webui.utils.access_control import has_access

 from open_webui.utils.auth import (
+    get_license_data,
    decode_token,
    get_admin_user,
    get_verified_user,
 )
-from open_webui.utils.oauth import oauth_manager
+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

+
 if SAFE_MODE:
    print("SAFE MODE ENABLED")
    Functions.deactivate_all_functions()
@@ -321,19 +356,23 @@ class SPAStaticFiles(StaticFiles):
            return await super().get_response(path, scope)
        except (HTTPException, StarletteHTTPException) as ex:
            if ex.status_code == 404:
-                return await super().get_response("index.html", scope)
+                if path.endswith(".js"):
+                    # Return 404 for javascript files
+                    raise ex
+                else:
+                    return await super().get_response("index.html", scope)
            else:
                raise ex


 print(
    rf"""
-  ___                    __        __   _     _   _ ___
- / _ \ _ __   ___ _ __   \ \      / /__| |__ | | | |_ _|
-| | | | '_ \ / _ \ '_ \   \ \ /\ / / _ \ '_ \| | | || |
-| |_| | |_) |  __/ | | |   \ V  V /  __/ |_) | |_| || |
- \___/| .__/ \___|_| |_|    \_/\_/ \___|_.__/ \___/|___|
-      |_|
+ ██████╗ ██████╗ ███████╗███╗   ██╗    ██╗    ██╗███████╗██████╗ ██╗   ██╗██╗
+██╔═══██╗██╔══██╗██╔════╝████╗  ██║    ██║    ██║██╔════╝██╔══██╗██║   ██║██║
+██║   ██║██████╔╝█████╗  ██╔██╗ ██║    ██║ █╗ ██║█████╗  ██████╔╝██║   ██║██║
+██║   ██║██╔═══╝ ██╔══╝  ██║╚██╗██║    ██║███╗██║██╔══╝  ██╔══██╗██║   ██║██║
+╚██████╔╝██║     ███████╗██║ ╚████║    ╚███╔███╔╝███████╗██████╔╝╚██████╔╝██║
+ ╚═════╝ ╚═╝     ╚══════╝╚═╝  ╚═══╝     ╚══╝╚══╝ ╚══════╝╚═════╝  ╚═════╝ ╚═╝


 v{VERSION} - building the best open-source AI user interface.
@@ -348,6 +387,9 @@ async def lifespan(app: FastAPI):
    if RESET_CONFIG_ON_START:
        reset_config()

+    if app.state.config.LICENSE_KEY:
+        get_license_data(app, app.state.config.LICENSE_KEY)
+
    asyncio.create_task(periodic_usage_pool_cleanup())
    yield

@@ -359,8 +401,12 @@ app = FastAPI(
    lifespan=lifespan,
 )

+oauth_manager = OAuthManager(app)
+
 app.state.config = AppConfig()

+app.state.WEBUI_NAME = WEBUI_NAME
+app.state.config.LICENSE_KEY = LICENSE_KEY

 ########################################
 #
@@ -388,6 +434,14 @@ app.state.config.OPENAI_API_CONFIGS = OPENAI_API_CONFIGS

 app.state.OPENAI_MODELS = {}

+########################################
+#
+# DIRECT CONNECTIONS
+#
+########################################
+
+app.state.config.ENABLE_DIRECT_CONNECTIONS = ENABLE_DIRECT_CONNECTIONS
+
 ########################################
 #
 # WEBUI
@@ -454,10 +508,10 @@ 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.USER_COUNT = None
 app.state.TOOLS = {}
 app.state.FUNCTIONS = {}

-
 ########################################
 #
 # RETRIEVAL
@@ -470,6 +524,8 @@ 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

+
+app.state.config.RAG_FULL_CONTEXT = RAG_FULL_CONTEXT
 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
@@ -504,6 +560,7 @@ 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.RAG_WEB_SEARCH_FULL_CONTEXT = RAG_WEB_SEARCH_FULL_CONTEXT
 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
@@ -513,6 +570,7 @@ app.state.config.GOOGLE_PSE_ENGINE_ID = GOOGLE_PSE_ENGINE_ID
 app.state.config.BRAVE_SEARCH_API_KEY = BRAVE_SEARCH_API_KEY
 app.state.config.KAGI_SEARCH_API_KEY = KAGI_SEARCH_API_KEY
 app.state.config.MOJEEK_SEARCH_API_KEY = MOJEEK_SEARCH_API_KEY
+app.state.config.BOCHA_SEARCH_API_KEY = BOCHA_SEARCH_API_KEY
 app.state.config.SERPSTACK_API_KEY = SERPSTACK_API_KEY
 app.state.config.SERPSTACK_HTTPS = SERPSTACK_HTTPS
 app.state.config.SERPER_API_KEY = SERPER_API_KEY
@@ -520,12 +578,20 @@ app.state.config.SERPLY_API_KEY = SERPLY_API_KEY
 app.state.config.TAVILY_API_KEY = TAVILY_API_KEY
 app.state.config.SEARCHAPI_API_KEY = SEARCHAPI_API_KEY
 app.state.config.SEARCHAPI_ENGINE = SEARCHAPI_ENGINE
+app.state.config.SERPAPI_API_KEY = SERPAPI_API_KEY
+app.state.config.SERPAPI_ENGINE = SERPAPI_ENGINE
 app.state.config.JINA_API_KEY = JINA_API_KEY
 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.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.FIRECRAWL_API_BASE_URL = FIRECRAWL_API_BASE_URL
+app.state.config.FIRECRAWL_API_KEY = FIRECRAWL_API_KEY

 app.state.EMBEDDING_FUNCTION = None
 app.state.ef = None
@@ -567,6 +633,34 @@ app.state.EMBEDDING_FUNCTION = get_embedding_function(
    app.state.config.RAG_EMBEDDING_BATCH_SIZE,
 )

+########################################
+#
+# CODE EXECUTION
+#
+########################################
+
+app.state.config.CODE_EXECUTION_ENGINE = CODE_EXECUTION_ENGINE
+app.state.config.CODE_EXECUTION_JUPYTER_URL = CODE_EXECUTION_JUPYTER_URL
+app.state.config.CODE_EXECUTION_JUPYTER_AUTH = CODE_EXECUTION_JUPYTER_AUTH
+app.state.config.CODE_EXECUTION_JUPYTER_AUTH_TOKEN = CODE_EXECUTION_JUPYTER_AUTH_TOKEN
+app.state.config.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD = (
+    CODE_EXECUTION_JUPYTER_AUTH_PASSWORD
+)
+app.state.config.CODE_EXECUTION_JUPYTER_TIMEOUT = CODE_EXECUTION_JUPYTER_TIMEOUT
+
+app.state.config.ENABLE_CODE_INTERPRETER = ENABLE_CODE_INTERPRETER
+app.state.config.CODE_INTERPRETER_ENGINE = CODE_INTERPRETER_ENGINE
+app.state.config.CODE_INTERPRETER_PROMPT_TEMPLATE = CODE_INTERPRETER_PROMPT_TEMPLATE
+
+app.state.config.CODE_INTERPRETER_JUPYTER_URL = CODE_INTERPRETER_JUPYTER_URL
+app.state.config.CODE_INTERPRETER_JUPYTER_AUTH = CODE_INTERPRETER_JUPYTER_AUTH
+app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_TOKEN = (
+    CODE_INTERPRETER_JUPYTER_AUTH_TOKEN
+)
+app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD = (
+    CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD
+)
+app.state.config.CODE_INTERPRETER_JUPYTER_TIMEOUT = CODE_INTERPRETER_JUPYTER_TIMEOUT

 ########################################
 #
@@ -581,6 +675,9 @@ app.state.config.ENABLE_IMAGE_PROMPT_GENERATION = ENABLE_IMAGE_PROMPT_GENERATION
 app.state.config.IMAGES_OPENAI_API_BASE_URL = IMAGES_OPENAI_API_BASE_URL
 app.state.config.IMAGES_OPENAI_API_KEY = IMAGES_OPENAI_API_KEY

+app.state.config.IMAGES_GEMINI_API_BASE_URL = IMAGES_GEMINI_API_BASE_URL
+app.state.config.IMAGES_GEMINI_API_KEY = IMAGES_GEMINI_API_KEY
+
 app.state.config.IMAGE_GENERATION_MODEL = IMAGE_GENERATION_MODEL

 app.state.config.AUTOMATIC1111_BASE_URL = AUTOMATIC1111_BASE_URL
@@ -609,6 +706,7 @@ app.state.config.STT_ENGINE = AUDIO_STT_ENGINE
 app.state.config.STT_MODEL = AUDIO_STT_MODEL

 app.state.config.WHISPER_MODEL = WHISPER_MODEL
+app.state.config.DEEPGRAM_API_KEY = DEEPGRAM_API_KEY

 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
@@ -643,6 +741,7 @@ app.state.config.ENABLE_SEARCH_QUERY_GENERATION = ENABLE_SEARCH_QUERY_GENERATION
 app.state.config.ENABLE_RETRIEVAL_QUERY_GENERATION = ENABLE_RETRIEVAL_QUERY_GENERATION
 app.state.config.ENABLE_AUTOCOMPLETE_GENERATION = ENABLE_AUTOCOMPLETE_GENERATION
 app.state.config.ENABLE_TAGS_GENERATION = ENABLE_TAGS_GENERATION
+app.state.config.ENABLE_TITLE_GENERATION = ENABLE_TITLE_GENERATION


 app.state.config.TITLE_GENERATION_PROMPT_TEMPLATE = TITLE_GENERATION_PROMPT_TEMPLATE
@@ -751,6 +850,7 @@ app.include_router(openai.router, prefix="/openai", tags=["openai"])
 app.include_router(pipelines.router, prefix="/api/v1/pipelines", tags=["pipelines"])
 app.include_router(tasks.router, prefix="/api/v1/tasks", tags=["tasks"])
 app.include_router(images.router, prefix="/api/v1/images", tags=["images"])
+
 app.include_router(audio.router, prefix="/api/v1/audio", tags=["audio"])
 app.include_router(retrieval.router, prefix="/api/v1/retrieval", tags=["retrieval"])

@@ -853,19 +953,30 @@ async def chat_completion(
    if not request.app.state.MODELS:
        await get_all_models(request)

+    model_item = form_data.pop("model_item", {})
    tasks = form_data.pop("background_tasks", None)
-    try:
-        model_id = form_data.get("model", None)
-        if model_id not in request.app.state.MODELS:
-            raise Exception("Model not found")
-        model = request.app.state.MODELS[model_id]

-        # Check if user has access to the model
-        if not BYPASS_MODEL_ACCESS_CONTROL and user.role == "user":
-            try:
-                check_model_access(user, model)
-            except Exception as e:
-                raise e
+    try:
+        if not model_item.get("direct", False):
+            model_id = form_data.get("model", None)
+            if model_id not in request.app.state.MODELS:
+                raise Exception("Model not found")
+
+            model = request.app.state.MODELS[model_id]
+            model_info = Models.get_model_by_id(model_id)
+
+            # Check if user has access to the model
+            if not BYPASS_MODEL_ACCESS_CONTROL and user.role == "user":
+                try:
+                    check_model_access(user, model)
+                except Exception as e:
+                    raise e
+        else:
+            model = model_item
+            model_info = None
+
+            request.state.direct = True
+            request.state.model = model

        metadata = {
            "user_id": user.id,
@@ -875,13 +986,30 @@ async def chat_completion(
            "tool_ids": form_data.get("tool_ids", None),
            "files": form_data.get("files", None),
            "features": form_data.get("features", None),
+            "variables": form_data.get("variables", None),
+            "model": model_info.model_dump() if model_info else model,
+            "direct": model_item.get("direct", False),
+            **(
+                {"function_calling": "native"}
+                if form_data.get("params", {}).get("function_calling") == "native"
+                or (
+                    model_info
+                    and model_info.params.model_dump().get("function_calling")
+                    == "native"
+                )
+                else {}
+            ),
        }
+
+        request.state.metadata = metadata
        form_data["metadata"] = metadata

-        form_data, events = await process_chat_payload(
+        form_data, metadata, events = await process_chat_payload(
            request, form_data, metadata, user, model
        )
+
    except Exception as e:
+        log.debug(f"Error processing chat payload: {e}")
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=str(e),
@@ -889,6 +1017,7 @@ async def chat_completion(

    try:
        response = await chat_completion_handler(request, form_data, user)
+
        return await process_chat_response(
            request, response, form_data, user, events, metadata, tasks
        )
@@ -909,6 +1038,12 @@ async def chat_completed(
    request: Request, form_data: dict, user=Depends(get_verified_user)
 ):
    try:
+        model_item = form_data.pop("model_item", {})
+
+        if model_item.get("direct", False):
+            request.state.direct = True
+            request.state.model = model_item
+
        return await chat_completed_handler(request, form_data, user)
    except Exception as e:
        raise HTTPException(
@@ -922,6 +1057,12 @@ async def chat_action(
    request: Request, action_id: str, form_data: dict, user=Depends(get_verified_user)
 ):
    try:
+        model_item = form_data.pop("model_item", {})
+
+        if model_item.get("direct", False):
+            request.state.direct = True
+            request.state.model = model_item
+
        return await chat_action_handler(request, action_id, form_data, user)
    except Exception as e:
        raise HTTPException(
@@ -975,7 +1116,7 @@ async def get_app_config(request: Request):
    return {
        **({"onboarding": True} if onboarding else {}),
        "status": True,
-        "name": WEBUI_NAME,
+        "name": app.state.WEBUI_NAME,
        "version": VERSION,
        "default_locale": str(DEFAULT_LOCALE),
        "oauth": {
@@ -994,27 +1135,29 @@ async def get_app_config(request: Request):
            "enable_websocket": ENABLE_WEBSOCKET_SUPPORT,
            **(
                {
+                    "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_google_drive_integration": app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
+                    "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_admin_export": ENABLE_ADMIN_EXPORT,
                    "enable_admin_chat_access": ENABLE_ADMIN_CHAT_ACCESS,
+                    "enable_google_drive_integration": app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
                }
                if user is not None
                else {}
            ),
        },
-        "google_drive": {
-            "client_id": GOOGLE_DRIVE_CLIENT_ID.value,
-            "api_key": GOOGLE_DRIVE_API_KEY.value,
-        },
        **(
            {
                "default_models": app.state.config.DEFAULT_MODELS,
                "default_prompt_suggestions": app.state.config.DEFAULT_PROMPT_SUGGESTIONS,
+                "code": {
+                    "engine": app.state.config.CODE_EXECUTION_ENGINE,
+                },
                "audio": {
                    "tts": {
                        "engine": app.state.config.TTS_ENGINE,
@@ -1030,6 +1173,10 @@ async def get_app_config(request: Request):
                    "max_count": app.state.config.FILE_MAX_COUNT,
                },
                "permissions": {**app.state.config.USER_PERMISSIONS},
+                "google_drive": {
+                    "client_id": GOOGLE_DRIVE_CLIENT_ID.value,
+                    "api_key": GOOGLE_DRIVE_API_KEY.value,
+                },
            }
            if user is not None
            else {}
@@ -1063,7 +1210,7 @@ async def get_app_version():


@app.get("/api/version/updates")
-async def get_app_latest_release_version():
+async def get_app_latest_release_version(user=Depends(get_verified_user)):
    if OFFLINE_MODE:
        log.debug(
            f"Offline mode is enabled, returning current version as latest version"
@@ -1107,7 +1254,7 @@ if len(OAUTH_PROVIDERS) > 0:

@app.get("/oauth/{provider}/login")
 async def oauth_login(provider: str, request: Request):
-    return await oauth_manager.handle_login(provider, request)
+    return await oauth_manager.handle_login(request, provider)


 # OAuth login logic is as follows:
@@ -1118,14 +1265,14 @@ async def oauth_login(provider: str, request: Request):
 #    - Email addresses are considered unique, so we fail registration if the email address is already taken
@app.get("/oauth/{provider}/callback")
 async def oauth_callback(provider: str, request: Request, response: Response):
-    return await oauth_manager.handle_callback(provider, request, response)
+    return await oauth_manager.handle_callback(request, provider, response)


@app.get("/manifest.json")
 async def get_manifest_json():
    return {
-        "name": WEBUI_NAME,
-        "short_name": WEBUI_NAME,
+        "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",
@@ -1152,8 +1299,8 @@ async def get_manifest_json():
 async def get_opensearch_xml():
    xml_content = rf"""
    <OpenSearchDescription xmlns="http://a9.com/-/spec/opensearch/1.1/" xmlns:moz="http://www.mozilla.org/2006/browser/search/">
-    <ShortName>{WEBUI_NAME}</ShortName>
-    <Description>Search {WEBUI_NAME}</Description>
+    <ShortName>{app.state.WEBUI_NAME}</ShortName>
+    <Description>Search {app.state.WEBUI_NAME}</Description>
    <InputEncoding>UTF-8</InputEncoding>
    <Image width="16" height="16" type="image/x-icon">{app.state.config.WEBUI_URL}/static/favicon.png</Image>
    <Url type="text/html" method="get" template="{app.state.config.WEBUI_URL}/?q={"{searchTerms}"}"/>
@@ -470,7 +470,7 @@ class ChatTable:
        try:
            with get_db() as db:
                # it is possible that the shared link was deleted. hence,
-                # we check if the chat is still shared by checkng if a chat with the share_id exists
+                # we check if the chat is still shared by checking if a chat with the share_id exists
                chat = db.query(Chat).filter_by(share_id=id).first()

                if chat:
@@ -271,6 +271,24 @@ class UsersTable:
        except Exception:
            return None

+    def update_user_settings_by_id(self, id: str, updated: dict) -> Optional[UserModel]:
+        try:
+            with get_db() as db:
+                user_settings = db.query(User).filter_by(id=id).first().settings
+
+                if user_settings is None:
+                    user_settings = {}
+
+                user_settings.update(updated)
+
+                db.query(User).filter_by(id=id).update({"settings": user_settings})
+                db.commit()
+
+                user = db.query(User).filter_by(id=id).first()
+                return UserModel.model_validate(user)
+        except Exception:
+            return None
+
    def delete_user_by_id(self, id: str) -> bool:
        try:
            # Remove User from Groups
@@ -14,9 +14,15 @@ 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
+from open_webui.utils.misc import get_last_user_message, calculate_sha256_string

-from open_webui.env import SRC_LOG_LEVELS, OFFLINE_MODE
+from open_webui.models.users import UserModel
+
+from open_webui.env import (
+    SRC_LOG_LEVELS,
+    OFFLINE_MODE,
+    ENABLE_FORWARD_USER_INFO_HEADERS,
+)

 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
@@ -61,9 +67,7 @@ class VectorSearchRetriever(BaseRetriever):


 def query_doc(
-    collection_name: str,
-    query_embedding: list[float],
-    k: int,
+    collection_name: str, query_embedding: list[float], k: int, user: UserModel = None
 ):
    try:
        result = VECTOR_DB_CLIENT.search(
@@ -81,6 +85,19 @@ def query_doc(
        raise e


+def get_doc(collection_name: str, user: UserModel = None):
+    try:
+        result = VECTOR_DB_CLIENT.get(collection_name=collection_name)
+
+        if result:
+            log.info(f"query_doc:result {result.ids} {result.metadatas}")
+
+        return result
+    except Exception as e:
+        print(e)
+        raise e
+
+
 def query_doc_with_hybrid_search(
    collection_name: str,
    query: str,
@@ -134,6 +151,27 @@ def query_doc_with_hybrid_search(
        raise e


+def merge_get_results(get_results: list[dict]) -> dict:
+    # Initialize lists to store combined data
+    combined_documents = []
+    combined_metadatas = []
+    combined_ids = []
+
+    for data in get_results:
+        combined_documents.extend(data["documents"][0])
+        combined_metadatas.extend(data["metadatas"][0])
+        combined_ids.extend(data["ids"][0])
+
+    # Create the output dictionary
+    result = {
+        "documents": [combined_documents],
+        "metadatas": [combined_metadatas],
+        "ids": [combined_ids],
+    }
+
+    return result
+
+
 def merge_and_sort_query_results(
    query_results: list[dict], k: int, reverse: bool = False
 ) -> list[dict]:
@@ -177,6 +215,23 @@ def merge_and_sort_query_results(
    return result


+def get_all_items_from_collections(collection_names: list[str]) -> dict:
+    results = []
+
+    for collection_name in collection_names:
+        if collection_name:
+            try:
+                result = get_doc(collection_name=collection_name)
+                if result is not None:
+                    results.append(result.model_dump())
+            except Exception as e:
+                log.exception(f"Error when querying the collection: {e}")
+        else:
+            pass
+
+    return merge_get_results(results)
+
+
 def query_collection(
    collection_names: list[str],
    queries: list[str],
@@ -259,26 +314,31 @@ def get_embedding_function(
    embedding_batch_size,
 ):
    if embedding_engine == "":
-        return lambda query: embedding_function.encode(query).tolist()
+        return lambda query, user=None: embedding_function.encode(query).tolist()
    elif embedding_engine in ["ollama", "openai"]:
-        func = lambda query: generate_embeddings(
+        func = lambda query, user=None: generate_embeddings(
            engine=embedding_engine,
            model=embedding_model,
            text=query,
            url=url,
            key=key,
+            user=user,
        )

-        def generate_multiple(query, func):
+        def generate_multiple(query, 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]))
+                    embeddings.extend(
+                        func(query[i : i + embedding_batch_size], user=user)
+                    )
                return embeddings
            else:
-                return func(query)
+                return func(query, user)

-        return lambda query: generate_multiple(query, func)
+        return lambda query, user=None: generate_multiple(query, user, func)
+    else:
+        raise ValueError(f"Unknown embedding engine: {embedding_engine}")


 def get_sources_from_files(
@@ -289,14 +349,22 @@ def get_sources_from_files(
    reranking_function,
    r,
    hybrid_search,
+    full_context=False,
 ):
-    log.debug(f"files: {files} {queries} {embedding_function} {reranking_function}")
+    log.debug(
+        f"files: {files} {queries} {embedding_function} {reranking_function} {full_context}"
+    )

    extracted_collections = []
    relevant_contexts = []

    for file in files:
-        if file.get("context") == "full":
+        if file.get("docs"):
+            context = {
+                "documents": [[doc.get("content") for doc in file.get("docs")]],
+                "metadatas": [[doc.get("metadata") for doc in file.get("docs")]],
+            }
+        elif file.get("context") == "full":
            context = {
                "documents": [[file.get("file").get("data", {}).get("content")]],
                "metadatas": [[{"file_id": file.get("id"), "name": file.get("name")}]],
@@ -323,36 +391,43 @@ def get_sources_from_files(
                log.debug(f"skipping {file} as it has already been extracted")
                continue

-            try:
-                context = None
-                if file.get("type") == "text":
-                    context = file["content"]
-                else:
-                    if hybrid_search:
-                        try:
-                            context = query_collection_with_hybrid_search(
+            if full_context:
+                try:
+                    context = get_all_items_from_collections(collection_names)
+                except Exception as e:
+                    log.exception(e)
+
+            else:
+                try:
+                    context = None
+                    if file.get("type") == "text":
+                        context = file["content"]
+                    else:
+                        if hybrid_search:
+                            try:
+                                context = query_collection_with_hybrid_search(
+                                    collection_names=collection_names,
+                                    queries=queries,
+                                    embedding_function=embedding_function,
+                                    k=k,
+                                    reranking_function=reranking_function,
+                                    r=r,
+                                )
+                            except Exception as e:
+                                log.debug(
+                                    "Error when using hybrid search, using"
+                                    " non hybrid search as fallback."
+                                )
+
+                        if (not hybrid_search) or (context is None):
+                            context = query_collection(
                                collection_names=collection_names,
                                queries=queries,
                                embedding_function=embedding_function,
                                k=k,
-                                reranking_function=reranking_function,
-                                r=r,
                            )
-                        except Exception as e:
-                            log.debug(
-                                "Error when using hybrid search, using"
-                                " non hybrid search as fallback."
-                            )
-
-                    if (not hybrid_search) or (context is None):
-                        context = query_collection(
-                            collection_names=collection_names,
-                            queries=queries,
-                            embedding_function=embedding_function,
-                            k=k,
-                        )
-            except Exception as e:
-                log.exception(e)
+                except Exception as e:
+                    log.exception(e)

            extracted_collections.extend(collection_names)

@@ -423,7 +498,11 @@ def get_model_path(model: str, update_model: bool = False):


 def generate_openai_batch_embeddings(
-    model: str, texts: list[str], url: str = "https://api.openai.com/v1", key: str = ""
+    model: str,
+    texts: list[str],
+    url: str = "https://api.openai.com/v1",
+    key: str = "",
+    user: UserModel = None,
 ) -> Optional[list[list[float]]]:
    try:
        r = requests.post(
@@ -431,6 +510,16 @@ def generate_openai_batch_embeddings(
            headers={
                "Content-Type": "application/json",
                "Authorization": f"Bearer {key}",
+                **(
+                    {
+                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Id": user.id,
+                        "X-OpenWebUI-User-Email": user.email,
+                        "X-OpenWebUI-User-Role": user.role,
+                    }
+                    if ENABLE_FORWARD_USER_INFO_HEADERS and user
+                    else {}
+                ),
            },
            json={"input": texts, "model": model},
        )
@@ -446,7 +535,7 @@ def generate_openai_batch_embeddings(


 def generate_ollama_batch_embeddings(
-    model: str, texts: list[str], url: str, key: str = ""
+    model: str, texts: list[str], url: str, key: str = "", user: UserModel = None
 ) -> Optional[list[list[float]]]:
    try:
        r = requests.post(
@@ -454,6 +543,16 @@ def generate_ollama_batch_embeddings(
            headers={
                "Content-Type": "application/json",
                "Authorization": f"Bearer {key}",
+                **(
+                    {
+                        "X-OpenWebUI-User-Name": user.name,
+                        "X-OpenWebUI-User-Id": user.id,
+                        "X-OpenWebUI-User-Email": user.email,
+                        "X-OpenWebUI-User-Role": user.role,
+                    }
+                    if ENABLE_FORWARD_USER_INFO_HEADERS
+                    else {}
+                ),
            },
            json={"input": texts, "model": model},
        )
@@ -472,22 +571,29 @@ def generate_ollama_batch_embeddings(
 def generate_embeddings(engine: str, model: str, text: Union[str, list[str]], **kwargs):
    url = kwargs.get("url", "")
    key = kwargs.get("key", "")
+    user = kwargs.get("user")

    if engine == "ollama":
        if isinstance(text, list):
            embeddings = generate_ollama_batch_embeddings(
-                **{"model": model, "texts": text, "url": url, "key": key}
+                **{"model": model, "texts": text, "url": url, "key": key, "user": user}
            )
        else:
            embeddings = generate_ollama_batch_embeddings(
-                **{"model": model, "texts": [text], "url": url, "key": key}
+                **{
+                    "model": model,
+                    "texts": [text],
+                    "url": url,
+                    "key": key,
+                    "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)
+            embeddings = generate_openai_batch_embeddings(model, text, url, key, user)
        else:
-            embeddings = generate_openai_batch_embeddings(model, [text], url, key)
+            embeddings = generate_openai_batch_embeddings(model, [text], url, key, user)

        return embeddings[0] if isinstance(text, str) else embeddings

@@ -8,13 +8,17 @@ from open_webui.retrieval.vector.main import VectorItem, SearchResult, GetResult
 from open_webui.config import (
    MILVUS_URI,
    MILVUS_DB,
+    MILVUS_TOKEN,
 )


 class MilvusClient:
    def __init__(self):
        self.collection_prefix = "open_webui"
-        self.client = Client(uri=MILVUS_URI, database=MILVUS_DB)
+        if MILVUS_TOKEN is None:
+            self.client = Client(uri=MILVUS_URI, database=MILVUS_DB)
+        else:
+            self.client = Client(uri=MILVUS_URI, database=MILVUS_DB, token=MILVUS_TOKEN)

    def _result_to_get_result(self, result) -> GetResult:
        ids = []
@@ -113,6 +113,34 @@ class OpenSearchClient:

        return self._result_to_search_result(result)

+    def query(
+        self, collection_name: str, filter: dict, limit: Optional[int] = None
+    ) -> Optional[GetResult]:
+        if not self.has_collection(collection_name):
+            return None
+
+        query_body = {
+            "query": {"bool": {"filter": []}},
+            "_source": ["text", "metadata"],
+        }
+
+        for field, value in filter.items():
+            query_body["query"]["bool"]["filter"].append({"term": {field: value}})
+
+        size = limit if limit else 10
+
+        try:
+            result = self.client.search(
+                index=f"{self.index_prefix}_{collection_name}",
+                body=query_body,
+                size=size,
+            )
+
+            return self._result_to_get_result(result)
+
+        except Exception as e:
+            return None
+
    def get_or_create_index(self, index_name: str, dimension: int):
        if not self.has_index(index_name):
            self._create_index(index_name, dimension)
@@ -0,0 +1,65 @@
+import logging
+from typing import Optional
+
+import requests
+import json
+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 _parse_response(response):
+    result = {}
+    if "data" in response:
+        data = response["data"]
+        if "webPages" in data:
+            webPages = data["webPages"]
+            if "value" in webPages:
+                result["webpage"] = [
+                    {
+                        "id": item.get("id", ""),
+                        "name": item.get("name", ""),
+                        "url": item.get("url", ""),
+                        "snippet": item.get("snippet", ""),
+                        "summary": item.get("summary", ""),
+                        "siteName": item.get("siteName", ""),
+                        "siteIcon": item.get("siteIcon", ""),
+                        "datePublished": item.get("datePublished", "")
+                        or item.get("dateLastCrawled", ""),
+                    }
+                    for item in webPages["value"]
+                ]
+    return result
+
+
+def search_bocha(
+    api_key: str, query: str, count: int, filter_list: Optional[list[str]] = None
+) -> list[SearchResult]:
+    """Search using Bocha's Search API and return the results as a list of SearchResult objects.
+
+    Args:
+        api_key (str): A Bocha Search API key
+        query (str): The query to search for
+    """
+    url = "https://api.bochaai.com/v1/web-search?utm_source=ollama"
+    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
+
+    payload = json.dumps(
+        {"query": query, "summary": True, "freshness": "noLimit", "count": count}
+    )
+
+    response = requests.post(url, headers=headers, data=payload, timeout=5)
+    response.raise_for_status()
+    results = _parse_response(response.json())
+    print(results)
+    if filter_list:
+        results = get_filtered_results(results, filter_list)
+
+    return [
+        SearchResult(
+            link=result["url"], title=result.get("name"), snippet=result.get("summary")
+        )
+        for result in results.get("webpage", [])[:count]
+    ]
@@ -32,19 +32,15 @@ def search_duckduckgo(
            # Convert the search results into a list
            search_results = [r for r in ddgs_gen]

-    # Create an empty list to store the SearchResult objects
-    results = []
-    # Iterate over each search result
-    for result in search_results:
-        # Create a SearchResult object and append it to the results list
-        results.append(
-            SearchResult(
-                link=result["href"],
-                title=result.get("title"),
-                snippet=result.get("body"),
-            )
-        )
    if filter_list:
-        results = get_filtered_results(results, filter_list)
+        search_results = get_filtered_results(search_results, filter_list)
+
    # Return the list of search results
-    return results
+    return [
+        SearchResult(
+            link=result["href"],
+            title=result.get("title"),
+            snippet=result.get("body"),
+        )
+        for result in search_results
+    ]
@@ -0,0 +1,76 @@
+import logging
+from dataclasses import dataclass
+from typing import Optional
+
+import requests
+from open_webui.env import SRC_LOG_LEVELS
+from open_webui.retrieval.web.main import SearchResult
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+EXA_API_BASE = "https://api.exa.ai"
+
+
+@dataclass
+class ExaResult:
+    url: str
+    title: str
+    text: str
+
+
+def search_exa(
+    api_key: str,
+    query: str,
+    count: int,
+    filter_list: Optional[list[str]] = None,
+) -> list[SearchResult]:
+    """Search using Exa Search API and return the results as a list of SearchResult objects.
+
+    Args:
+        api_key (str): A Exa Search API key
+        query (str): The query to search for
+        count (int): Number of results to return
+        filter_list (Optional[list[str]]): List of domains to filter results by
+    """
+    log.info(f"Searching with Exa for query: {query}")
+
+    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
+
+    payload = {
+        "query": query,
+        "numResults": count or 5,
+        "includeDomains": filter_list,
+        "contents": {"text": True, "highlights": True},
+        "type": "auto",  # Use the auto search type (keyword or neural)
+    }
+
+    try:
+        response = requests.post(
+            f"{EXA_API_BASE}/search", headers=headers, json=payload
+        )
+        response.raise_for_status()
+        data = response.json()
+
+        results = []
+        for result in data["results"]:
+            results.append(
+                ExaResult(
+                    url=result["url"],
+                    title=result["title"],
+                    text=result["text"],
+                )
+            )
+
+        log.info(f"Found {len(results)} results")
+        return [
+            SearchResult(
+                link=result.url,
+                title=result.title,
+                snippet=result.text,
+            )
+            for result in results
+        ]
+    except Exception as e:
+        log.error(f"Error searching Exa: {e}")
+        return []
@@ -17,34 +17,53 @@ def search_google_pse(
    filter_list: Optional[list[str]] = None,
 ) -> list[SearchResult]:
    """Search using Google's Programmable Search Engine API and return the results as a list of SearchResult objects.
+    Handles pagination for counts greater than 10.

    Args:
        api_key (str): A Programmable Search Engine API key
        search_engine_id (str): A Programmable Search Engine ID
        query (str): The query to search for
+        count (int): The number of results to return (max 100, as PSE max results per query is 10 and max page is 10)
+        filter_list (Optional[list[str]], optional): A list of keywords to filter out from results. Defaults to None.
+
+    Returns:
+        list[SearchResult]: A list of SearchResult objects.
    """
    url = "https://www.googleapis.com/customsearch/v1"
-
    headers = {"Content-Type": "application/json"}
-    params = {
-        "cx": search_engine_id,
-        "q": query,
-        "key": api_key,
-        "num": count,
-    }
+    all_results = []
+    start_index = 1  # Google PSE start parameter is 1-based

-    response = requests.request("GET", url, headers=headers, params=params)
-    response.raise_for_status()
+    while count > 0:
+        num_results_this_page = min(count, 10)  # Google PSE max results per page is 10
+        params = {
+            "cx": search_engine_id,
+            "q": query,
+            "key": api_key,
+            "num": num_results_this_page,
+            "start": start_index,
+        }
+        response = requests.request("GET", url, headers=headers, params=params)
+        response.raise_for_status()
+        json_response = response.json()
+        results = json_response.get("items", [])
+        if results:  # check if results are returned. If not, no more pages to fetch.
+            all_results.extend(results)
+            count -= len(
+                results
+            )  # Decrement count by the number of results fetched in this page.
+            start_index += 10  # Increment start index for the next page
+        else:
+            break  # No more results from Google PSE, break the loop

-    json_response = response.json()
-    results = json_response.get("items", [])
    if filter_list:
-        results = get_filtered_results(results, filter_list)
+        all_results = get_filtered_results(all_results, filter_list)
+
    return [
        SearchResult(
            link=result["link"],
            title=result.get("title"),
            snippet=result.get("snippet"),
        )
-        for result in results
+        for result in all_results
    ]
@@ -20,14 +20,23 @@ def search_jina(api_key: str, query: str, count: int) -> list[SearchResult]:
        list[SearchResult]: A list of search results
    """
    jina_search_endpoint = "https://s.jina.ai/"
-    headers = {"Accept": "application/json", "Authorization": f"Bearer {api_key}"}
-    url = str(URL(jina_search_endpoint + query))
-    response = requests.get(url, headers=headers)
+
+    headers = {
+        "Accept": "application/json",
+        "Content-Type": "application/json",
+        "Authorization": api_key,
+        "X-Retain-Images": "none",
+    }
+
+    payload = {"q": query, "count": count if count <= 10 else 10}
+
+    url = str(URL(jina_search_endpoint))
+    response = requests.post(url, headers=headers, json=payload)
    response.raise_for_status()
    data = response.json()

    results = []
-    for result in data["data"][:count]:
+    for result in data["data"]:
        results.append(
            SearchResult(
                link=result["url"],
@@ -1,3 +1,5 @@
+import validators
+
 from typing import Optional
 from urllib.parse import urlparse

@@ -10,6 +12,8 @@ def get_filtered_results(results, filter_list):
    filtered_results = []
    for result in results:
        url = result.get("url") or result.get("link", "")
+        if not validators.url(url):
+            continue
        domain = urlparse(url).netloc
        if any(domain.endswith(filtered_domain) for filtered_domain in filter_list):
            filtered_results.append(result)
@@ -0,0 +1,48 @@
+import logging
+from typing import Optional
+from urllib.parse import urlencode
+
+import requests
+from open_webui.retrieval.web.main import SearchResult, get_filtered_results
+from open_webui.env import SRC_LOG_LEVELS
+
+log = logging.getLogger(__name__)
+log.setLevel(SRC_LOG_LEVELS["RAG"])
+
+
+def search_serpapi(
+    api_key: str,
+    engine: str,
+    query: str,
+    count: int,
+    filter_list: Optional[list[str]] = None,
+) -> list[SearchResult]:
+    """Search using serpapi.com's API and return the results as a list of SearchResult objects.
+
+    Args:
+      api_key (str): A serpapi.com API key
+      query (str): The query to search for
+    """
+    url = "https://serpapi.com/search"
+
+    engine = engine or "google"
+
+    payload = {"engine": engine, "q": query, "api_key": api_key}
+
+    url = f"{url}?{urlencode(payload)}"
+    response = requests.request("GET", url)
+
+    json_response = response.json()
+    log.info(f"results from serpapi search: {json_response}")
+
+    results = sorted(
+        json_response.get("organic_results", []), key=lambda x: x.get("position", 0)
+    )
+    if filter_list:
+        results = get_filtered_results(results, filter_list)
+    return [
+        SearchResult(
+            link=result["link"], title=result["title"], snippet=result["snippet"]
+        )
+        for result in results[:count]
+    ]
@@ -1,4 +1,5 @@
 import logging
+from typing import Optional

 import requests
 from open_webui.retrieval.web.main import SearchResult
@@ -8,7 +9,13 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])


-def search_tavily(api_key: str, query: str, count: int) -> list[SearchResult]:
+def search_tavily(
+    api_key: str,
+    query: str,
+    count: int,
+    filter_list: Optional[list[str]] = None,
+    # **kwargs,
+) -> list[SearchResult]:
    """Search using Tavily's Search API and return the results as a list of SearchResult objects.

    Args:
@@ -20,7 +27,6 @@ def search_tavily(api_key: str, query: str, count: int) -> list[SearchResult]:
    """
    url = "https://api.tavily.com/search"
    data = {"query": query, "api_key": api_key}
-
    response = requests.post(url, json=data)
    response.raise_for_status()

@@ -1,19 +1,38 @@
-import socket
-import urllib.parse
-import validators
-from typing import Union, Sequence, Iterator
-
-from langchain_community.document_loaders import (
-    WebBaseLoader,
-)
-from langchain_core.documents import Document
-
-
-from open_webui.constants import ERROR_MESSAGES
-from open_webui.config import ENABLE_RAG_LOCAL_WEB_FETCH
-from open_webui.env import SRC_LOG_LEVELS
-
+import asyncio
 import logging
+import socket
+import ssl
+import urllib.parse
+import urllib.request
+from collections import defaultdict
+from datetime import datetime, time, timedelta
+from typing import (
+    Any,
+    AsyncIterator,
+    Dict,
+    Iterator,
+    List,
+    Optional,
+    Sequence,
+    Union,
+    Literal,
+)
+import aiohttp
+import certifi
+import validators
+from langchain_community.document_loaders import PlaywrightURLLoader, WebBaseLoader
+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.constants import ERROR_MESSAGES
+from open_webui.config import (
+    ENABLE_RAG_LOCAL_WEB_FETCH,
+    PLAYWRIGHT_WS_URI,
+    RAG_WEB_LOADER_ENGINE,
+    FIRECRAWL_API_BASE_URL,
+    FIRECRAWL_API_KEY,
+)
+from open_webui.env import SRC_LOG_LEVELS

 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["RAG"])
@@ -43,6 +62,17 @@ def validate_url(url: Union[str, Sequence[str]]):
        return False


+def safe_validate_urls(url: Sequence[str]) -> Sequence[str]:
+    valid_urls = []
+    for u in url:
+        try:
+            if validate_url(u):
+                valid_urls.append(u)
+        except ValueError:
+            continue
+    return valid_urls
+
+
 def resolve_hostname(hostname):
    # Get address information
    addr_info = socket.getaddrinfo(hostname, None)
@@ -54,9 +84,381 @@ def resolve_hostname(hostname):
    return ipv4_addresses, ipv6_addresses


+def extract_metadata(soup, url):
+    metadata = {"source": url}
+    if title := soup.find("title"):
+        metadata["title"] = title.get_text()
+    if description := soup.find("meta", attrs={"name": "description"}):
+        metadata["description"] = description.get("content", "No description found.")
+    if html := soup.find("html"):
+        metadata["language"] = html.get("lang", "No language found.")
+    return metadata
+
+
+def verify_ssl_cert(url: str) -> bool:
+    """Verify SSL certificate for the given URL."""
+    if not url.startswith("https://"):
+        return True
+
+    try:
+        hostname = url.split("://")[-1].split("/")[0]
+        context = ssl.create_default_context(cafile=certifi.where())
+        with context.wrap_socket(ssl.socket(), server_hostname=hostname) as s:
+            s.connect((hostname, 443))
+        return True
+    except ssl.SSLError:
+        return False
+    except Exception as e:
+        log.warning(f"SSL verification failed for {url}: {str(e)}")
+        return False
+
+
+class SafeFireCrawlLoader(BaseLoader):
+    def __init__(
+        self,
+        web_paths,
+        verify_ssl: bool = True,
+        trust_env: bool = False,
+        requests_per_second: Optional[float] = None,
+        continue_on_failure: bool = True,
+        api_key: Optional[str] = None,
+        api_url: Optional[str] = None,
+        mode: Literal["crawl", "scrape", "map"] = "crawl",
+        proxy: Optional[Dict[str, str]] = None,
+        params: Optional[Dict] = None,
+    ):
+        """Concurrent document loader for FireCrawl operations.
+
+        Executes multiple FireCrawlLoader instances concurrently using thread pooling
+        to improve bulk processing efficiency.
+        Args:
+            web_paths: List of URLs/paths to process.
+            verify_ssl: If True, verify SSL certificates.
+            trust_env: If True, use proxy settings from environment variables.
+            requests_per_second: Number of requests per second to limit to.
+            continue_on_failure (bool): If True, continue loading other URLs on failure.
+            api_key: API key for FireCrawl service. Defaults to None
+                (uses FIRE_CRAWL_API_KEY environment variable if not provided).
+            api_url: Base URL for FireCrawl API. Defaults to official API endpoint.
+            mode: Operation mode selection:
+                - 'crawl': Website crawling mode (default)
+                - 'scrape': Direct page scraping
+                - 'map': Site map generation
+            proxy: Proxy override settings for the FireCrawl API.
+            params: The parameters to pass to the Firecrawl API.
+                Examples include crawlerOptions.
+                For more details, visit: https://github.com/mendableai/firecrawl-py
+        """
+        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}
+        self.web_paths = web_paths
+        self.verify_ssl = verify_ssl
+        self.requests_per_second = requests_per_second
+        self.last_request_time = None
+        self.trust_env = trust_env
+        self.continue_on_failure = continue_on_failure
+        self.api_key = api_key
+        self.api_url = api_url
+        self.mode = mode
+        self.params = params
+
+    def lazy_load(self) -> Iterator[Document]:
+        """Load documents concurrently using FireCrawl."""
+        for url in self.web_paths:
+            try:
+                self._safe_process_url_sync(url)
+                loader = FireCrawlLoader(
+                    url=url,
+                    api_key=self.api_key,
+                    api_url=self.api_url,
+                    mode=self.mode,
+                    params=self.params,
+                )
+                yield from loader.lazy_load()
+            except Exception as e:
+                if self.continue_on_failure:
+                    log.exception(e, "Error loading %s", url)
+                    continue
+                raise e
+
+    async def alazy_load(self):
+        """Async version of lazy_load."""
+        for url in self.web_paths:
+            try:
+                await self._safe_process_url(url)
+                loader = FireCrawlLoader(
+                    url=url,
+                    api_key=self.api_key,
+                    api_url=self.api_url,
+                    mode=self.mode,
+                    params=self.params,
+                )
+                async for document in loader.alazy_load():
+                    yield document
+            except Exception as e:
+                if self.continue_on_failure:
+                    log.exception(e, "Error loading %s", url)
+                    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()
+
+    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 SafePlaywrightURLLoader(PlaywrightURLLoader):
+    """Load HTML pages safely with Playwright, supporting SSL verification, rate limiting, and remote browser connection.
+
+    Attributes:
+        web_paths (List[str]): List of URLs to load.
+        verify_ssl (bool): If True, verify SSL certificates.
+        trust_env (bool): If True, use proxy settings from environment variables.
+        requests_per_second (Optional[float]): Number of requests per second to limit to.
+        continue_on_failure (bool): If True, continue loading other URLs on failure.
+        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.
+    """
+
+    def __init__(
+        self,
+        web_paths: List[str],
+        verify_ssl: bool = True,
+        trust_env: bool = False,
+        requests_per_second: Optional[float] = None,
+        continue_on_failure: bool = True,
+        headless: bool = True,
+        remove_selectors: Optional[List[str]] = None,
+        proxy: Optional[Dict[str, str]] = None,
+        playwright_ws_url: Optional[str] = None,
+    ):
+        """Initialize with additional safety parameters and remote browser support."""
+
+        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}
+
+        # We'll set headless to False if using playwright_ws_url since it's handled by the remote browser
+        super().__init__(
+            urls=web_paths,
+            continue_on_failure=continue_on_failure,
+            headless=headless if playwright_ws_url is None else False,
+            remove_selectors=remove_selectors,
+            proxy=proxy,
+        )
+        self.verify_ssl = verify_ssl
+        self.requests_per_second = requests_per_second
+        self.last_request_time = None
+        self.playwright_ws_url = playwright_ws_url
+        self.trust_env = trust_env
+
+    def lazy_load(self) -> Iterator[Document]:
+        """Safely load URLs synchronously with support for remote browser."""
+        from playwright.sync_api import sync_playwright
+
+        with sync_playwright() as p:
+            # Use remote browser if ws_endpoint is provided, otherwise use local browser
+            if self.playwright_ws_url:
+                browser = p.chromium.connect(self.playwright_ws_url)
+            else:
+                browser = p.chromium.launch(headless=self.headless, proxy=self.proxy)
+
+            for url in self.urls:
+                try:
+                    self._safe_process_url_sync(url)
+                    page = browser.new_page()
+                    response = page.goto(url)
+                    if response is None:
+                        raise ValueError(f"page.goto() returned None for url {url}")
+
+                    text = self.evaluator.evaluate(page, browser, response)
+                    metadata = {"source": url}
+                    yield Document(page_content=text, metadata=metadata)
+                except Exception as e:
+                    if self.continue_on_failure:
+                        log.exception(e, "Error loading %s", url)
+                        continue
+                    raise e
+            browser.close()
+
+    async def alazy_load(self) -> AsyncIterator[Document]:
+        """Safely load URLs asynchronously with support for remote browser."""
+        from playwright.async_api import async_playwright
+
+        async with async_playwright() as p:
+            # Use remote browser if ws_endpoint is provided, otherwise use local browser
+            if self.playwright_ws_url:
+                browser = await p.chromium.connect(self.playwright_ws_url)
+            else:
+                browser = await p.chromium.launch(
+                    headless=self.headless, proxy=self.proxy
+                )
+
+            for url in self.urls:
+                try:
+                    await self._safe_process_url(url)
+                    page = await browser.new_page()
+                    response = await page.goto(url)
+                    if response is None:
+                        raise ValueError(f"page.goto() returned None for url {url}")
+
+                    text = await self.evaluator.evaluate_async(page, browser, response)
+                    metadata = {"source": url}
+                    yield Document(page_content=text, metadata=metadata)
+                except Exception as e:
+                    if self.continue_on_failure:
+                        log.exception(e, "Error loading %s", url)
+                        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."""

+    def __init__(self, trust_env: bool = False, *args, **kwargs):
+        """Initialize SafeWebBaseLoader
+        Args:
+            trust_env (bool, optional): set to True if using proxy to make web requests, for example
+                using http(s)_proxy environment variables. Defaults to False.
+        """
+        super().__init__(*args, **kwargs)
+        self.trust_env = trust_env
+
+    async def _fetch(
+        self, url: str, retries: int = 3, cooldown: int = 2, backoff: float = 1.5
+    ) -> str:
+        async with aiohttp.ClientSession(trust_env=self.trust_env) as session:
+            for i in range(retries):
+                try:
+                    kwargs: Dict = dict(
+                        headers=self.session.headers,
+                        cookies=self.session.cookies.get_dict(),
+                    )
+                    if not self.session.verify:
+                        kwargs["ssl"] = False
+
+                    async with session.get(
+                        url, **(self.requests_kwargs | kwargs)
+                    ) as response:
+                        if self.raise_for_status:
+                            response.raise_for_status()
+                        return await response.text()
+                except aiohttp.ClientConnectionError as e:
+                    if i == retries - 1:
+                        raise
+                    else:
+                        log.warning(
+                            f"Error fetching {url} with attempt "
+                            f"{i + 1}/{retries}: {e}. Retrying..."
+                        )
+                        await asyncio.sleep(cooldown * backoff**i)
+        raise ValueError("retry count exceeded")
+
+    def _unpack_fetch_results(
+        self, results: Any, urls: List[str], parser: Union[str, None] = None
+    ) -> List[Any]:
+        """Unpack fetch results into BeautifulSoup objects."""
+        from bs4 import BeautifulSoup
+
+        final_results = []
+        for i, result in enumerate(results):
+            url = urls[i]
+            if parser is None:
+                if url.endswith(".xml"):
+                    parser = "xml"
+                else:
+                    parser = self.default_parser
+                self._check_parser(parser)
+            final_results.append(BeautifulSoup(result, parser, **self.bs_kwargs))
+        return final_results
+
+    async def ascrape_all(
+        self, urls: List[str], parser: Union[str, None] = None
+    ) -> List[Any]:
+        """Async fetch all urls, then return soups for all results."""
+        results = await self.fetch_all(urls)
+        return self._unpack_fetch_results(results, urls, parser=parser)
+
    def lazy_load(self) -> Iterator[Document]:
        """Lazy load text from the url(s) in web_path with error handling."""
        for path in self.web_paths:
@@ -65,33 +467,72 @@ class SafeWebBaseLoader(WebBaseLoader):
                text = soup.get_text(**self.bs_get_text_kwargs)

                # Build metadata
-                metadata = {"source": path}
-                if title := soup.find("title"):
-                    metadata["title"] = title.get_text()
-                if description := soup.find("meta", attrs={"name": "description"}):
-                    metadata["description"] = description.get(
-                        "content", "No description found."
-                    )
-                if html := soup.find("html"):
-                    metadata["language"] = html.get("lang", "No language found.")
+                metadata = extract_metadata(soup, path)

                yield Document(page_content=text, metadata=metadata)
            except Exception as e:
                # Log the error and continue with the next URL
-                log.error(f"Error loading {path}: {e}")
+                log.exception(e, "Error loading %s", path)
+
+    async def alazy_load(self) -> AsyncIterator[Document]:
+        """Async lazy load text from the url(s) in web_path."""
+        results = await self.ascrape_all(self.web_paths)
+        for path, soup in zip(self.web_paths, results):
+            text = soup.get_text(**self.bs_get_text_kwargs)
+            metadata = {"source": path}
+            if title := soup.find("title"):
+                metadata["title"] = title.get_text()
+            if description := soup.find("meta", attrs={"name": "description"}):
+                metadata["description"] = description.get(
+                    "content", "No description found."
+                )
+            if html := soup.find("html"):
+                metadata["language"] = html.get("lang", "No language found.")
+            yield Document(page_content=text, metadata=metadata)
+
+    async def aload(self) -> list[Document]:
+        """Load data into Document objects."""
+        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,
    requests_per_second: int = 2,
+    trust_env: bool = False,
 ):
-    # Check if the URL is valid
-    if not validate_url(urls):
-        raise ValueError(ERROR_MESSAGES.INVALID_URL)
-    return SafeWebBaseLoader(
-        urls,
-        verify_ssl=verify_ssl,
-        requests_per_second=requests_per_second,
-        continue_on_failure=True,
+    # Check if the URLs are valid
+    safe_urls = safe_validate_urls([urls] if isinstance(urls, str) else urls)
+
+    web_loader_args = {
+        "web_paths": safe_urls,
+        "verify_ssl": verify_ssl,
+        "requests_per_second": requests_per_second,
+        "continue_on_failure": True,
+        "trust_env": trust_env,
+    }
+
+    if PLAYWRIGHT_WS_URI.value:
+        web_loader_args["playwright_ws_url"] = PLAYWRIGHT_WS_URI.value
+
+    if RAG_WEB_LOADER_ENGINE.value == "firecrawl":
+        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)
+
+    log.debug(
+        "Using RAG_WEB_LOADER_ENGINE %s for %s URLs",
+        web_loader.__class__.__name__,
+        len(safe_urls),
    )
+
+    return web_loader
@@ -11,6 +11,7 @@ from pydub.silence import split_on_silence
 import aiohttp
 import aiofiles
 import requests
+import mimetypes

 from fastapi import (
    Depends,
@@ -36,6 +37,7 @@ from open_webui.config import (

 from open_webui.constants import ERROR_MESSAGES
 from open_webui.env import (
+    AIOHTTP_CLIENT_TIMEOUT,
    ENV,
    SRC_LOG_LEVELS,
    DEVICE_TYPE,
@@ -138,6 +140,7 @@ class STTConfigForm(BaseModel):
    ENGINE: str
    MODEL: str
    WHISPER_MODEL: str
+    DEEPGRAM_API_KEY: str


 class AudioConfigUpdateForm(BaseModel):
@@ -165,6 +168,7 @@ async def get_audio_config(request: Request, user=Depends(get_admin_user)):
            "ENGINE": request.app.state.config.STT_ENGINE,
            "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,
        },
    }

@@ -190,6 +194,7 @@ async def update_audio_config(
    request.app.state.config.STT_ENGINE = form_data.stt.ENGINE
    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

    if request.app.state.config.STT_ENGINE == "":
        request.app.state.faster_whisper_model = set_faster_whisper_model(
@@ -214,6 +219,7 @@ async def update_audio_config(
            "ENGINE": request.app.state.config.STT_ENGINE,
            "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,
        },
    }

@@ -261,7 +267,10 @@ async def speech(request: Request, user=Depends(get_verified_user)):

        try:
            # print(payload)
-            async with aiohttp.ClientSession() as session:
+            timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT)
+            async with aiohttp.ClientSession(
+                timeout=timeout, trust_env=True
+            ) as session:
                async with session.post(
                    url=f"{request.app.state.config.TTS_OPENAI_API_BASE_URL}/audio/speech",
                    json=payload,
@@ -318,7 +327,10 @@ async def speech(request: Request, user=Depends(get_verified_user)):
            )

        try:
-            async with aiohttp.ClientSession() as session:
+            timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT)
+            async with aiohttp.ClientSession(
+                timeout=timeout, trust_env=True
+            ) as session:
                async with session.post(
                    f"https://api.elevenlabs.io/v1/text-to-speech/{voice_id}",
                    json={
@@ -375,7 +387,10 @@ async def speech(request: Request, user=Depends(get_verified_user)):
            data = f"""<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xml:lang="{locale}">
                <voice name="{language}">{payload["input"]}</voice>
            </speak>"""
-            async with aiohttp.ClientSession() as session:
+            timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT)
+            async with aiohttp.ClientSession(
+                timeout=timeout, trust_env=True
+            ) as session:
                async with session.post(
                    f"https://{region}.tts.speech.microsoft.com/cognitiveservices/v1",
                    headers={
@@ -521,6 +536,69 @@ def transcribe(request: Request, file_path):

            raise Exception(detail if detail else "Open WebUI: Server Connection Error")

+    elif request.app.state.config.STT_ENGINE == "deepgram":
+        try:
+            # Determine the MIME type of the file
+            mime, _ = mimetypes.guess_type(file_path)
+            if not mime:
+                mime = "audio/wav"  # fallback to wav if undetectable
+
+            # Read the audio file
+            with open(file_path, "rb") as f:
+                file_data = f.read()
+
+            # Build headers and parameters
+            headers = {
+                "Authorization": f"Token {request.app.state.config.DEEPGRAM_API_KEY}",
+                "Content-Type": mime,
+            }
+
+            # Add model if specified
+            params = {}
+            if request.app.state.config.STT_MODEL:
+                params["model"] = request.app.state.config.STT_MODEL
+
+            # Make request to Deepgram API
+            r = requests.post(
+                "https://api.deepgram.com/v1/listen",
+                headers=headers,
+                params=params,
+                data=file_data,
+            )
+            r.raise_for_status()
+            response_data = r.json()
+
+            # Extract transcript from Deepgram response
+            try:
+                transcript = response_data["results"]["channels"][0]["alternatives"][
+                    0
+                ].get("transcript", "")
+            except (KeyError, IndexError) as e:
+                log.error(f"Malformed response from Deepgram: {str(e)}")
+                raise Exception(
+                    "Failed to parse Deepgram response - unexpected response format"
+                )
+            data = {"text": transcript.strip()}
+
+            # Save transcript
+            transcript_file = f"{file_dir}/{id}.json"
+            with open(transcript_file, "w") as f:
+                json.dump(data, f)
+
+            return data
+
+        except Exception as e:
+            log.exception(e)
+            detail = None
+            if r is not None:
+                try:
+                    res = r.json()
+                    if "error" in res:
+                        detail = f"External: {res['error'].get('message', '')}"
+                except Exception:
+                    detail = f"External: {e}"
+            raise Exception(detail if detail else "Open WebUI: Server Connection Error")
+

 def compress_audio(file_path):
    if os.path.getsize(file_path) > MAX_FILE_SIZE:
@@ -25,8 +25,8 @@ from open_webui.env import (
    WEBUI_AUTH,
    WEBUI_AUTH_TRUSTED_EMAIL_HEADER,
    WEBUI_AUTH_TRUSTED_NAME_HEADER,
-    WEBUI_SESSION_COOKIE_SAME_SITE,
-    WEBUI_SESSION_COOKIE_SECURE,
+    WEBUI_AUTH_COOKIE_SAME_SITE,
+    WEBUI_AUTH_COOKIE_SECURE,
    SRC_LOG_LEVELS,
 )
 from fastapi import APIRouter, Depends, HTTPException, Request, status
@@ -95,8 +95,8 @@ async def get_session_user(
        value=token,
        expires=datetime_expires_at,
        httponly=True,  # Ensures the cookie is not accessible via JavaScript
-        samesite=WEBUI_SESSION_COOKIE_SAME_SITE,
-        secure=WEBUI_SESSION_COOKIE_SECURE,
+        samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+        secure=WEBUI_AUTH_COOKIE_SECURE,
    )

    user_permissions = get_permissions(
@@ -164,7 +164,7 @@ async def update_password(
 ############################
 # LDAP Authentication
 ############################
-@router.post("/ldap", response_model=SigninResponse)
+@router.post("/ldap", response_model=SessionUserResponse)
 async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
    ENABLE_LDAP = request.app.state.config.ENABLE_LDAP
    LDAP_SERVER_LABEL = request.app.state.config.LDAP_SERVER_LABEL
@@ -251,9 +251,19 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
            user = Users.get_user_by_email(mail)
            if not user:
                try:
+                    user_count = Users.get_num_users()
+                    if (
+                        request.app.state.USER_COUNT
+                        and user_count >= request.app.state.USER_COUNT
+                    ):
+                        raise HTTPException(
+                            status.HTTP_403_FORBIDDEN,
+                            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+                        )
+
                    role = (
                        "admin"
-                        if Users.get_num_users() == 0
+                        if user_count == 0
                        else request.app.state.config.DEFAULT_USER_ROLE
                    )

@@ -288,6 +298,10 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                    httponly=True,  # Ensures the cookie is not accessible via JavaScript
                )

+                user_permissions = get_permissions(
+                    user.id, request.app.state.config.USER_PERMISSIONS
+                )
+
                return {
                    "token": token,
                    "token_type": "Bearer",
@@ -296,6 +310,7 @@ async def ldap_auth(request: Request, response: Response, form_data: LdapForm):
                    "name": user.name,
                    "role": user.role,
                    "profile_image_url": user.profile_image_url,
+                    "permissions": user_permissions,
                }
            else:
                raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
@@ -378,8 +393,8 @@ async def signin(request: Request, response: Response, form_data: SigninForm):
            value=token,
            expires=datetime_expires_at,
            httponly=True,  # Ensures the cookie is not accessible via JavaScript
-            samesite=WEBUI_SESSION_COOKIE_SAME_SITE,
-            secure=WEBUI_SESSION_COOKIE_SECURE,
+            samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+            secure=WEBUI_AUTH_COOKIE_SECURE,
        )

        user_permissions = get_permissions(
@@ -408,6 +423,7 @@ async def signin(request: Request, response: Response, form_data: SigninForm):

@router.post("/signup", response_model=SessionUserResponse)
 async def signup(request: Request, response: Response, form_data: SignupForm):
+
    if WEBUI_AUTH:
        if (
            not request.app.state.config.ENABLE_SIGNUP
@@ -422,6 +438,12 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
                status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.ACCESS_PROHIBITED
            )

+    user_count = Users.get_num_users()
+    if request.app.state.USER_COUNT and user_count >= request.app.state.USER_COUNT:
+        raise HTTPException(
+            status.HTTP_403_FORBIDDEN, detail=ERROR_MESSAGES.ACCESS_PROHIBITED
+        )
+
    if not validate_email_format(form_data.email.lower()):
        raise HTTPException(
            status.HTTP_400_BAD_REQUEST, detail=ERROR_MESSAGES.INVALID_EMAIL_FORMAT
@@ -432,12 +454,10 @@ async def signup(request: Request, response: Response, form_data: SignupForm):

    try:
        role = (
-            "admin"
-            if Users.get_num_users() == 0
-            else request.app.state.config.DEFAULT_USER_ROLE
+            "admin" if user_count == 0 else request.app.state.config.DEFAULT_USER_ROLE
        )

-        if Users.get_num_users() == 0:
+        if user_count == 0:
            # Disable signup after the first user is created
            request.app.state.config.ENABLE_SIGNUP = False

@@ -473,12 +493,13 @@ async def signup(request: Request, response: Response, form_data: SignupForm):
                value=token,
                expires=datetime_expires_at,
                httponly=True,  # Ensures the cookie is not accessible via JavaScript
-                samesite=WEBUI_SESSION_COOKIE_SAME_SITE,
-                secure=WEBUI_SESSION_COOKIE_SECURE,
+                samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+                secure=WEBUI_AUTH_COOKIE_SECURE,
            )

            if request.app.state.config.WEBHOOK_URL:
                post_webhook(
+                    request.app.state.WEBUI_NAME,
                    request.app.state.config.WEBHOOK_URL,
                    WEBHOOK_MESSAGES.USER_SIGNUP(user.name),
                    {
@@ -525,7 +546,8 @@ async def signout(request: Request, response: Response):
                            if logout_url:
                                response.delete_cookie("oauth_id_token")
                                return RedirectResponse(
-                                    url=f"{logout_url}?id_token_hint={oauth_id_token}"
+                                    headers=response.headers,
+                                    url=f"{logout_url}?id_token_hint={oauth_id_token}",
                                )
                        else:
                            raise HTTPException(
@@ -192,7 +192,7 @@ async def get_channel_messages(
 ############################


-async def send_notification(webui_url, channel, message, active_user_ids):
+async def send_notification(name, webui_url, channel, message, active_user_ids):
    users = get_users_with_access("read", channel.access_control)

    for user in users:
@@ -206,6 +206,7 @@ async def send_notification(webui_url, channel, message, active_user_ids):

                if webhook_url:
                    post_webhook(
+                        name,
                        webhook_url,
                        f"#{channel.name} - {webui_url}/channels/{channel.id}\n\n{message.content}",
                        {
@@ -302,6 +303,7 @@ async def post_new_message(

            background_tasks.add_task(
                send_notification,
+                request.app.state.WEBUI_NAME,
                request.app.state.config.WEBUI_URL,
                channel,
                message,
@@ -444,15 +444,21 @@ async def pin_chat_by_id(id: str, user=Depends(get_verified_user)):
 ############################


+class CloneForm(BaseModel):
+    title: Optional[str] = None
+
+
@router.post("/{id}/clone", response_model=Optional[ChatResponse])
-async def clone_chat_by_id(id: str, user=Depends(get_verified_user)):
+async def clone_chat_by_id(
+    form_data: CloneForm, id: str, user=Depends(get_verified_user)
+):
    chat = Chats.get_chat_by_id_and_user_id(id, user.id)
    if chat:
        updated_chat = {
            **chat.chat,
            "originalChatId": chat.id,
            "branchPointMessageId": chat.chat["history"]["currentId"],
-            "title": f"Clone of {chat.title}",
+            "title": form_data.title if form_data.title else f"Clone of {chat.title}",
        }

        chat = Chats.insert_new_chat(user.id, ChatForm(**{"chat": updated_chat}))
@@ -36,6 +36,140 @@ async def export_config(user=Depends(get_admin_user)):
    return get_config()


+############################
+# Direct Connections Config
+############################
+
+
+class DirectConnectionsConfigForm(BaseModel):
+    ENABLE_DIRECT_CONNECTIONS: bool
+
+
+@router.get("/direct_connections", response_model=DirectConnectionsConfigForm)
+async def get_direct_connections_config(request: Request, user=Depends(get_admin_user)):
+    return {
+        "ENABLE_DIRECT_CONNECTIONS": request.app.state.config.ENABLE_DIRECT_CONNECTIONS,
+    }
+
+
+@router.post("/direct_connections", response_model=DirectConnectionsConfigForm)
+async def set_direct_connections_config(
+    request: Request,
+    form_data: DirectConnectionsConfigForm,
+    user=Depends(get_admin_user),
+):
+    request.app.state.config.ENABLE_DIRECT_CONNECTIONS = (
+        form_data.ENABLE_DIRECT_CONNECTIONS
+    )
+    return {
+        "ENABLE_DIRECT_CONNECTIONS": request.app.state.config.ENABLE_DIRECT_CONNECTIONS,
+    }
+
+
+############################
+# CodeInterpreterConfig
+############################
+class CodeInterpreterConfigForm(BaseModel):
+    CODE_EXECUTION_ENGINE: str
+    CODE_EXECUTION_JUPYTER_URL: Optional[str]
+    CODE_EXECUTION_JUPYTER_AUTH: Optional[str]
+    CODE_EXECUTION_JUPYTER_AUTH_TOKEN: Optional[str]
+    CODE_EXECUTION_JUPYTER_AUTH_PASSWORD: Optional[str]
+    CODE_EXECUTION_JUPYTER_TIMEOUT: Optional[int]
+    ENABLE_CODE_INTERPRETER: bool
+    CODE_INTERPRETER_ENGINE: str
+    CODE_INTERPRETER_PROMPT_TEMPLATE: Optional[str]
+    CODE_INTERPRETER_JUPYTER_URL: Optional[str]
+    CODE_INTERPRETER_JUPYTER_AUTH: Optional[str]
+    CODE_INTERPRETER_JUPYTER_AUTH_TOKEN: Optional[str]
+    CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD: Optional[str]
+    CODE_INTERPRETER_JUPYTER_TIMEOUT: Optional[int]
+
+
+@router.get("/code_execution", response_model=CodeInterpreterConfigForm)
+async def get_code_execution_config(request: Request, user=Depends(get_admin_user)):
+    return {
+        "CODE_EXECUTION_ENGINE": request.app.state.config.CODE_EXECUTION_ENGINE,
+        "CODE_EXECUTION_JUPYTER_URL": request.app.state.config.CODE_EXECUTION_JUPYTER_URL,
+        "CODE_EXECUTION_JUPYTER_AUTH": request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH,
+        "CODE_EXECUTION_JUPYTER_AUTH_TOKEN": request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_TOKEN,
+        "CODE_EXECUTION_JUPYTER_AUTH_PASSWORD": request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD,
+        "CODE_EXECUTION_JUPYTER_TIMEOUT": request.app.state.config.CODE_EXECUTION_JUPYTER_TIMEOUT,
+        "ENABLE_CODE_INTERPRETER": request.app.state.config.ENABLE_CODE_INTERPRETER,
+        "CODE_INTERPRETER_ENGINE": request.app.state.config.CODE_INTERPRETER_ENGINE,
+        "CODE_INTERPRETER_PROMPT_TEMPLATE": request.app.state.config.CODE_INTERPRETER_PROMPT_TEMPLATE,
+        "CODE_INTERPRETER_JUPYTER_URL": request.app.state.config.CODE_INTERPRETER_JUPYTER_URL,
+        "CODE_INTERPRETER_JUPYTER_AUTH": request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH,
+        "CODE_INTERPRETER_JUPYTER_AUTH_TOKEN": request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_TOKEN,
+        "CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD": request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD,
+        "CODE_INTERPRETER_JUPYTER_TIMEOUT": request.app.state.config.CODE_INTERPRETER_JUPYTER_TIMEOUT,
+    }
+
+
+@router.post("/code_execution", response_model=CodeInterpreterConfigForm)
+async def set_code_execution_config(
+    request: Request, form_data: CodeInterpreterConfigForm, user=Depends(get_admin_user)
+):
+
+    request.app.state.config.CODE_EXECUTION_ENGINE = form_data.CODE_EXECUTION_ENGINE
+    request.app.state.config.CODE_EXECUTION_JUPYTER_URL = (
+        form_data.CODE_EXECUTION_JUPYTER_URL
+    )
+    request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH = (
+        form_data.CODE_EXECUTION_JUPYTER_AUTH
+    )
+    request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_TOKEN = (
+        form_data.CODE_EXECUTION_JUPYTER_AUTH_TOKEN
+    )
+    request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD = (
+        form_data.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD
+    )
+    request.app.state.config.CODE_EXECUTION_JUPYTER_TIMEOUT = (
+        form_data.CODE_EXECUTION_JUPYTER_TIMEOUT
+    )
+
+    request.app.state.config.ENABLE_CODE_INTERPRETER = form_data.ENABLE_CODE_INTERPRETER
+    request.app.state.config.CODE_INTERPRETER_ENGINE = form_data.CODE_INTERPRETER_ENGINE
+    request.app.state.config.CODE_INTERPRETER_PROMPT_TEMPLATE = (
+        form_data.CODE_INTERPRETER_PROMPT_TEMPLATE
+    )
+
+    request.app.state.config.CODE_INTERPRETER_JUPYTER_URL = (
+        form_data.CODE_INTERPRETER_JUPYTER_URL
+    )
+
+    request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH = (
+        form_data.CODE_INTERPRETER_JUPYTER_AUTH
+    )
+
+    request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_TOKEN = (
+        form_data.CODE_INTERPRETER_JUPYTER_AUTH_TOKEN
+    )
+    request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD = (
+        form_data.CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD
+    )
+    request.app.state.config.CODE_INTERPRETER_JUPYTER_TIMEOUT = (
+        form_data.CODE_INTERPRETER_JUPYTER_TIMEOUT
+    )
+
+    return {
+        "CODE_EXECUTION_ENGINE": request.app.state.config.CODE_EXECUTION_ENGINE,
+        "CODE_EXECUTION_JUPYTER_URL": request.app.state.config.CODE_EXECUTION_JUPYTER_URL,
+        "CODE_EXECUTION_JUPYTER_AUTH": request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH,
+        "CODE_EXECUTION_JUPYTER_AUTH_TOKEN": request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_TOKEN,
+        "CODE_EXECUTION_JUPYTER_AUTH_PASSWORD": request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD,
+        "CODE_EXECUTION_JUPYTER_TIMEOUT": request.app.state.config.CODE_EXECUTION_JUPYTER_TIMEOUT,
+        "ENABLE_CODE_INTERPRETER": request.app.state.config.ENABLE_CODE_INTERPRETER,
+        "CODE_INTERPRETER_ENGINE": request.app.state.config.CODE_INTERPRETER_ENGINE,
+        "CODE_INTERPRETER_PROMPT_TEMPLATE": request.app.state.config.CODE_INTERPRETER_PROMPT_TEMPLATE,
+        "CODE_INTERPRETER_JUPYTER_URL": request.app.state.config.CODE_INTERPRETER_JUPYTER_URL,
+        "CODE_INTERPRETER_JUPYTER_AUTH": request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH,
+        "CODE_INTERPRETER_JUPYTER_AUTH_TOKEN": request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_TOKEN,
+        "CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD": request.app.state.config.CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD,
+        "CODE_INTERPRETER_JUPYTER_TIMEOUT": request.app.state.config.CODE_INTERPRETER_JUPYTER_TIMEOUT,
+    }
+
+
 ############################
 # SetDefaultModels
 ############################
@@ -3,30 +3,22 @@ import os
 import uuid
 from pathlib import Path
 from typing import Optional
-from pydantic import BaseModel
-import mimetypes
 from urllib.parse import quote

-from open_webui.storage.provider import Storage
-
+from fastapi import APIRouter, Depends, File, HTTPException, Request, UploadFile, status
+from fastapi.responses import FileResponse, StreamingResponse
+from open_webui.constants import ERROR_MESSAGES
+from open_webui.env import SRC_LOG_LEVELS
 from open_webui.models.files import (
    FileForm,
    FileModel,
    FileModelResponse,
    Files,
 )
-from open_webui.routers.retrieval import process_file, ProcessFileForm
-
-from open_webui.config import UPLOAD_DIR
-from open_webui.env import SRC_LOG_LEVELS
-from open_webui.constants import ERROR_MESSAGES
-
-
-from fastapi import APIRouter, Depends, File, HTTPException, UploadFile, status, Request
-from fastapi.responses import FileResponse, StreamingResponse
-
-
+from open_webui.routers.retrieval import ProcessFileForm, process_file
+from open_webui.storage.provider import Storage
 from open_webui.utils.auth import get_admin_user, get_verified_user
+from pydantic import BaseModel

 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MODELS"])
@@ -41,7 +33,10 @@ router = APIRouter()

@router.post("/", response_model=FileModelResponse)
 def upload_file(
-    request: Request, file: UploadFile = File(...), user=Depends(get_verified_user)
+    request: Request,
+    file: UploadFile = File(...),
+    user=Depends(get_verified_user),
+    file_metadata: dict = {},
 ):
    log.info(f"file.content_type: {file.content_type}")
    try:
@@ -65,13 +60,14 @@ def upload_file(
                        "name": name,
                        "content_type": file.content_type,
                        "size": len(contents),
+                        "data": file_metadata,
                    },
                }
            ),
        )

        try:
-            process_file(request, ProcessFileForm(file_id=id))
+            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)
@@ -126,7 +122,7 @@ async def delete_all_files(user=Depends(get_admin_user)):
            Storage.delete_all_files()
        except Exception as e:
            log.exception(e)
-            log.error(f"Error deleting files")
+            log.error("Error deleting files")
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=ERROR_MESSAGES.DEFAULT("Error deleting files"),
@@ -193,7 +189,9 @@ async def update_file_data_content_by_id(
    if file and (file.user_id == user.id or user.role == "admin"):
        try:
            process_file(
-                request, ProcessFileForm(file_id=id, content=form_data.content)
+                request,
+                ProcessFileForm(file_id=id, content=form_data.content),
+                user=user,
            )
            file = Files.get_file_by_id(id=id)
        except Exception as e:
@@ -227,17 +225,24 @@ async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
                filename = file.meta.get("name", file.filename)
                encoded_filename = quote(filename)  # RFC5987 encoding

+                content_type = file.meta.get("content_type")
+                filename = file.meta.get("name", file.filename)
+                encoded_filename = quote(filename)
                headers = {}
-                if file.meta.get("content_type") not in [
-                    "application/pdf",
-                    "text/plain",
-                ]:
-                    headers = {
-                        **headers,
-                        "Content-Disposition": f"attachment; filename*=UTF-8''{encoded_filename}",
-                    }

-                return FileResponse(file_path, headers=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":
+                    headers["Content-Disposition"] = (
+                        f"attachment; filename*=UTF-8''{encoded_filename}"
+                    )
+
+                return FileResponse(file_path, headers=headers, media_type=content_type)

            else:
                raise HTTPException(
@@ -246,7 +251,7 @@ async def get_file_content_by_id(id: str, user=Depends(get_verified_user)):
                )
        except Exception as e:
            log.exception(e)
-            log.error(f"Error getting file content")
+            log.error("Error getting file content")
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=ERROR_MESSAGES.DEFAULT("Error getting file content"),
@@ -277,7 +282,7 @@ async def get_html_file_content_by_id(id: str, user=Depends(get_verified_user)):
                )
        except Exception as e:
            log.exception(e)
-            log.error(f"Error getting file content")
+            log.error("Error getting file content")
            raise HTTPException(
                status_code=status.HTTP_400_BAD_REQUEST,
                detail=ERROR_MESSAGES.DEFAULT("Error getting file content"),
@@ -353,7 +358,7 @@ async def delete_file_by_id(id: str, user=Depends(get_verified_user)):
                Storage.delete_file(file.path)
            except Exception as e:
                log.exception(e)
-                log.error(f"Error deleting files")
+                log.error("Error deleting files")
                raise HTTPException(
                    status_code=status.HTTP_400_BAD_REQUEST,
                    detail=ERROR_MESSAGES.DEFAULT("Error deleting files"),
@@ -1,32 +1,26 @@
 import asyncio
 import base64
+import io
 import json
 import logging
 import mimetypes
 import re
-import uuid
 from pathlib import Path
 from typing import Optional

 import requests
-
-
-from fastapi import Depends, FastAPI, HTTPException, Request, APIRouter
-from fastapi.middleware.cors import CORSMiddleware
-from pydantic import BaseModel
-
-
+from fastapi import APIRouter, Depends, HTTPException, Request, UploadFile
 from open_webui.config import CACHE_DIR
 from open_webui.constants import ERROR_MESSAGES
-from open_webui.env import ENV, SRC_LOG_LEVELS, ENABLE_FORWARD_USER_INFO_HEADERS
-
+from open_webui.env import ENABLE_FORWARD_USER_INFO_HEADERS, SRC_LOG_LEVELS
+from open_webui.routers.files import upload_file
 from open_webui.utils.auth import get_admin_user, get_verified_user
 from open_webui.utils.images.comfyui import (
    ComfyUIGenerateImageForm,
    ComfyUIWorkflow,
    comfyui_generate_image,
 )
-
+from pydantic import BaseModel

 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["IMAGES"])
@@ -61,6 +55,10 @@ async def get_config(request: Request, user=Depends(get_admin_user)):
            "COMFYUI_WORKFLOW": request.app.state.config.COMFYUI_WORKFLOW,
            "COMFYUI_WORKFLOW_NODES": request.app.state.config.COMFYUI_WORKFLOW_NODES,
        },
+        "gemini": {
+            "GEMINI_API_BASE_URL": request.app.state.config.IMAGES_GEMINI_API_BASE_URL,
+            "GEMINI_API_KEY": request.app.state.config.IMAGES_GEMINI_API_KEY,
+        },
    }


@@ -84,6 +82,11 @@ class ComfyUIConfigForm(BaseModel):
    COMFYUI_WORKFLOW_NODES: list[dict]


+class GeminiConfigForm(BaseModel):
+    GEMINI_API_BASE_URL: str
+    GEMINI_API_KEY: str
+
+
 class ConfigForm(BaseModel):
    enabled: bool
    engine: str
@@ -91,6 +94,7 @@ class ConfigForm(BaseModel):
    openai: OpenAIConfigForm
    automatic1111: Automatic1111ConfigForm
    comfyui: ComfyUIConfigForm
+    gemini: GeminiConfigForm


@router.post("/config/update")
@@ -109,6 +113,11 @@ async def update_config(
    )
    request.app.state.config.IMAGES_OPENAI_API_KEY = form_data.openai.OPENAI_API_KEY

+    request.app.state.config.IMAGES_GEMINI_API_BASE_URL = (
+        form_data.gemini.GEMINI_API_BASE_URL
+    )
+    request.app.state.config.IMAGES_GEMINI_API_KEY = form_data.gemini.GEMINI_API_KEY
+
    request.app.state.config.AUTOMATIC1111_BASE_URL = (
        form_data.automatic1111.AUTOMATIC1111_BASE_URL
    )
@@ -161,6 +170,10 @@ async def update_config(
            "COMFYUI_WORKFLOW": request.app.state.config.COMFYUI_WORKFLOW,
            "COMFYUI_WORKFLOW_NODES": request.app.state.config.COMFYUI_WORKFLOW_NODES,
        },
+        "gemini": {
+            "GEMINI_API_BASE_URL": request.app.state.config.IMAGES_GEMINI_API_BASE_URL,
+            "GEMINI_API_KEY": request.app.state.config.IMAGES_GEMINI_API_KEY,
+        },
    }


@@ -230,6 +243,12 @@ def get_image_model(request):
            if request.app.state.config.IMAGE_GENERATION_MODEL
            else "dall-e-2"
        )
+    elif request.app.state.config.IMAGE_GENERATION_ENGINE == "gemini":
+        return (
+            request.app.state.config.IMAGE_GENERATION_MODEL
+            if request.app.state.config.IMAGE_GENERATION_MODEL
+            else "imagen-3.0-generate-002"
+        )
    elif request.app.state.config.IMAGE_GENERATION_ENGINE == "comfyui":
        return (
            request.app.state.config.IMAGE_GENERATION_MODEL
@@ -271,7 +290,6 @@ async def get_image_config(request: Request, user=Depends(get_admin_user)):
 async def update_image_config(
    request: Request, form_data: ImageConfigForm, user=Depends(get_admin_user)
 ):
-
    set_image_model(request, form_data.MODEL)

    pattern = r"^\d+x\d+$"
@@ -306,6 +324,10 @@ def get_models(request: Request, user=Depends(get_verified_user)):
                {"id": "dall-e-2", "name": "DALL·E 2"},
                {"id": "dall-e-3", "name": "DALL·E 3"},
            ]
+        elif request.app.state.config.IMAGE_GENERATION_ENGINE == "gemini":
+            return [
+                {"id": "imagen-3-0-generate-002", "name": "imagen-3.0 generate-002"},
+            ]
        elif request.app.state.config.IMAGE_GENERATION_ENGINE == "comfyui":
            # TODO - get models from comfyui
            headers = {
@@ -383,40 +405,22 @@ class GenerateImageForm(BaseModel):
    negative_prompt: Optional[str] = None


-def save_b64_image(b64_str):
+def load_b64_image_data(b64_str):
    try:
-        image_id = str(uuid.uuid4())
-
        if "," in b64_str:
            header, encoded = b64_str.split(",", 1)
            mime_type = header.split(";")[0]
-
            img_data = base64.b64decode(encoded)
-            image_format = mimetypes.guess_extension(mime_type)
-
-            image_filename = f"{image_id}{image_format}"
-            file_path = IMAGE_CACHE_DIR / f"{image_filename}"
-            with open(file_path, "wb") as f:
-                f.write(img_data)
-            return image_filename
        else:
-            image_filename = f"{image_id}.png"
-            file_path = IMAGE_CACHE_DIR.joinpath(image_filename)
-
+            mime_type = "image/png"
            img_data = base64.b64decode(b64_str)
-
-            # Write the image data to a file
-            with open(file_path, "wb") as f:
-                f.write(img_data)
-            return image_filename
-
+        return img_data, mime_type
    except Exception as e:
-        log.exception(f"Error saving image: {e}")
+        log.exception(f"Error loading image data: {e}")
        return None


-def save_url_image(url, headers=None):
-    image_id = str(uuid.uuid4())
+def load_url_image_data(url, headers=None):
    try:
        if headers:
            r = requests.get(url, headers=headers)
@@ -426,18 +430,7 @@ def save_url_image(url, headers=None):
        r.raise_for_status()
        if r.headers["content-type"].split("/")[0] == "image":
            mime_type = r.headers["content-type"]
-            image_format = mimetypes.guess_extension(mime_type)
-
-            if not image_format:
-                raise ValueError("Could not determine image type from MIME type")
-
-            image_filename = f"{image_id}{image_format}"
-
-            file_path = IMAGE_CACHE_DIR.joinpath(f"{image_filename}")
-            with open(file_path, "wb") as image_file:
-                for chunk in r.iter_content(chunk_size=8192):
-                    image_file.write(chunk)
-            return image_filename
+            return r.content, mime_type
        else:
            log.error("Url does not point to an image.")
            return None
@@ -447,6 +440,20 @@ def save_url_image(url, headers=None):
        return None


+def upload_image(request, image_metadata, image_data, content_type, user):
+    image_format = mimetypes.guess_extension(content_type)
+    file = UploadFile(
+        file=io.BytesIO(image_data),
+        filename=f"generated-image{image_format}",  # will be converted to a unique ID on upload_file
+        headers={
+            "content-type": content_type,
+        },
+    )
+    file_item = upload_file(request, file, user, file_metadata=image_metadata)
+    url = request.app.url_path_for("get_file_content_by_id", id=file_item.id)
+    return url
+
+
@router.post("/generations")
 async def image_generations(
    request: Request,
@@ -500,12 +507,43 @@ async def image_generations(
            images = []

            for image in res["data"]:
-                image_filename = save_b64_image(image["b64_json"])
-                images.append({"url": f"/cache/image/generations/{image_filename}"})
-                file_body_path = IMAGE_CACHE_DIR.joinpath(f"{image_filename}.json")
+                image_data, content_type = load_b64_image_data(image["b64_json"])
+                url = upload_image(request, data, image_data, content_type, user)
+                images.append({"url": url})
+            return images

-                with open(file_body_path, "w") as f:
-                    json.dump(data, f)
+        elif request.app.state.config.IMAGE_GENERATION_ENGINE == "gemini":
+            headers = {}
+            headers["Content-Type"] = "application/json"
+            headers["x-goog-api-key"] = request.app.state.config.IMAGES_GEMINI_API_KEY
+
+            model = get_image_model(request)
+            data = {
+                "instances": {"prompt": form_data.prompt},
+                "parameters": {
+                    "sampleCount": form_data.n,
+                    "outputOptions": {"mimeType": "image/png"},
+                },
+            }
+
+            # Use asyncio.to_thread for the requests.post call
+            r = await asyncio.to_thread(
+                requests.post,
+                url=f"{request.app.state.config.IMAGES_GEMINI_API_BASE_URL}/models/{model}:predict",
+                json=data,
+                headers=headers,
+            )
+
+            r.raise_for_status()
+            res = r.json()
+
+            images = []
+            for image in res["predictions"]:
+                image_data, content_type = load_b64_image_data(
+                    image["bytesBase64Encoded"]
+                )
+                url = upload_image(request, data, image_data, content_type, user)
+                images.append({"url": url})

            return images

@@ -552,14 +590,15 @@ async def image_generations(
                        "Authorization": f"Bearer {request.app.state.config.COMFYUI_API_KEY}"
                    }

-                image_filename = save_url_image(image["url"], headers)
-                images.append({"url": f"/cache/image/generations/{image_filename}"})
-                file_body_path = IMAGE_CACHE_DIR.joinpath(f"{image_filename}.json")
-
-                with open(file_body_path, "w") as f:
-                    json.dump(form_data.model_dump(exclude_none=True), f)
-
-            log.debug(f"images: {images}")
+                image_data, content_type = load_url_image_data(image["url"], headers)
+                url = upload_image(
+                    request,
+                    form_data.model_dump(exclude_none=True),
+                    image_data,
+                    content_type,
+                    user,
+                )
+                images.append({"url": url})
            return images
        elif (
            request.app.state.config.IMAGE_GENERATION_ENGINE == "automatic1111"
@@ -604,13 +643,15 @@ async def image_generations(
            images = []

            for image in res["images"]:
-                image_filename = save_b64_image(image)
-                images.append({"url": f"/cache/image/generations/{image_filename}"})
-                file_body_path = IMAGE_CACHE_DIR.joinpath(f"{image_filename}.json")
-
-                with open(file_body_path, "w") as f:
-                    json.dump({**data, "info": res["info"]}, f)
-
+                image_data, content_type = load_b64_image_data(image)
+                url = upload_image(
+                    request,
+                    {**data, "info": res["info"]},
+                    image_data,
+                    content_type,
+                    user,
+                )
+                images.append({"url": url})
            return images
    except Exception as e:
        error = e
@@ -264,7 +264,11 @@ def add_file_to_knowledge_by_id(
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if knowledge.user_id != user.id and user.role != "admin":
+    if (
+        knowledge.user_id != user.id
+        and not has_access(user.id, "write", knowledge.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -285,7 +289,9 @@ def add_file_to_knowledge_by_id(
    # Add content to the vector database
    try:
        process_file(
-            request, ProcessFileForm(file_id=form_data.file_id, collection_name=id)
+            request,
+            ProcessFileForm(file_id=form_data.file_id, collection_name=id),
+            user=user,
        )
    except Exception as e:
        log.debug(e)
@@ -342,7 +348,12 @@ def update_file_from_knowledge_by_id(
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if knowledge.user_id != user.id and user.role != "admin":
+    if (
+        knowledge.user_id != user.id
+        and not has_access(user.id, "write", knowledge.access_control)
+        and user.role != "admin"
+    ):
+
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -363,7 +374,9 @@ def update_file_from_knowledge_by_id(
    # Add content to the vector database
    try:
        process_file(
-            request, ProcessFileForm(file_id=form_data.file_id, collection_name=id)
+            request,
+            ProcessFileForm(file_id=form_data.file_id, collection_name=id),
+            user=user,
        )
    except Exception as e:
        raise HTTPException(
@@ -406,7 +419,11 @@ def remove_file_from_knowledge_by_id(
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if knowledge.user_id != user.id and user.role != "admin":
+    if (
+        knowledge.user_id != user.id
+        and not has_access(user.id, "write", knowledge.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -429,10 +446,6 @@ def remove_file_from_knowledge_by_id(
    if VECTOR_DB_CLIENT.has_collection(collection_name=file_collection):
        VECTOR_DB_CLIENT.delete_collection(collection_name=file_collection)

-    # Delete physical file
-    if file.path:
-        Storage.delete_file(file.path)
-
    # Delete file from database
    Files.delete_file_by_id(form_data.file_id)

@@ -484,7 +497,11 @@ async def delete_knowledge_by_id(id: str, user=Depends(get_verified_user)):
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if knowledge.user_id != user.id and user.role != "admin":
+    if (
+        knowledge.user_id != user.id
+        and not has_access(user.id, "write", knowledge.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -543,7 +560,11 @@ async def reset_knowledge_by_id(id: str, user=Depends(get_verified_user)):
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if knowledge.user_id != user.id and user.role != "admin":
+    if (
+        knowledge.user_id != user.id
+        and not has_access(user.id, "write", knowledge.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -582,7 +603,11 @@ def add_files_to_knowledge_batch(
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if knowledge.user_id != user.id and user.role != "admin":
+    if (
+        knowledge.user_id != user.id
+        and not has_access(user.id, "write", knowledge.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_400_BAD_REQUEST,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -57,7 +57,7 @@ async def add_memory(
            {
                "id": memory.id,
                "text": memory.content,
-                "vector": request.app.state.EMBEDDING_FUNCTION(memory.content),
+                "vector": request.app.state.EMBEDDING_FUNCTION(memory.content, user),
                "metadata": {"created_at": memory.created_at},
            }
        ],
@@ -82,7 +82,7 @@ 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)],
+        vectors=[request.app.state.EMBEDDING_FUNCTION(form_data.content, user)],
        limit=form_data.k,
    )

@@ -105,7 +105,7 @@ async def reset_memory_from_vector_db(
            {
                "id": memory.id,
                "text": memory.content,
-                "vector": request.app.state.EMBEDDING_FUNCTION(memory.content),
+                "vector": request.app.state.EMBEDDING_FUNCTION(memory.content, user),
                "metadata": {
                    "created_at": memory.created_at,
                    "updated_at": memory.updated_at,
@@ -160,7 +160,9 @@ async def update_memory_by_id(
                {
                    "id": memory.id,
                    "text": memory.content,
-                    "vector": request.app.state.EMBEDDING_FUNCTION(memory.content),
+                    "vector": request.app.state.EMBEDDING_FUNCTION(
+                        memory.content, user
+                    ),
                    "metadata": {
                        "created_at": memory.created_at,
                        "updated_at": memory.updated_at,
@@ -155,6 +155,16 @@ async def update_model_by_id(
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

+    if (
+        model.user_id != user.id
+        and not has_access(user.id, "write", model.access_control)
+        and user.role != "admin"
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
    model = Models.update_model_by_id(id, form_data)
    return model

@@ -173,7 +183,11 @@ async def delete_model_by_id(id: str, user=Depends(get_verified_user)):
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if model.user_id != user.id and user.role != "admin":
+    if (
+        user.role != "admin"
+        and model.user_id != user.id
+        and not has_access(user.id, "write", model.access_control)
+    ):
        raise HTTPException(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.UNAUTHORIZED,
@@ -11,10 +11,8 @@ import re
 import time
 from typing import Optional, Union
 from urllib.parse import urlparse
-
 import aiohttp
 from aiocache import cached
-
 import requests

 from fastapi import (
@@ -28,7 +26,7 @@ from fastapi import (
 )
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import StreamingResponse
-from pydantic import BaseModel, ConfigDict
+from pydantic import BaseModel, ConfigDict, validator
 from starlette.background import BackgroundTask


@@ -261,7 +259,7 @@ async def get_all_models(request: Request):
    if request.app.state.config.ENABLE_OLLAMA_API:
        request_tasks = []
        for idx, url in enumerate(request.app.state.config.OLLAMA_BASE_URLS):
-            if (str(idx) not in request.app.state.config.OLLAMA_API_CONFIGS) or (
+            if (str(idx) not in request.app.state.config.OLLAMA_API_CONFIGS) and (
                url not in request.app.state.config.OLLAMA_API_CONFIGS  # Legacy support
            ):
                request_tasks.append(send_get_request(f"{url}/api/tags"))
@@ -395,7 +393,7 @@ async def get_ollama_tags(
            )

    if user.role == "user" and not BYPASS_MODEL_ACCESS_CONTROL:
-        models["models"] = get_filtered_models(models, user)
+        models["models"] = await get_filtered_models(models, user)

    return models

@@ -938,18 +936,33 @@ async def generate_completion(

 class ChatMessage(BaseModel):
    role: str
-    content: str
+    content: Optional[str] = None
+    tool_calls: Optional[list[dict]] = None
    images: Optional[list[str]] = None

+    @validator("content", pre=True)
+    @classmethod
+    def check_at_least_one_field(cls, field_value, values, **kwargs):
+        # Raise an error if both 'content' and 'tool_calls' are None
+        if field_value is None and (
+            "tool_calls" not in values or values["tool_calls"] is None
+        ):
+            raise ValueError(
+                "At least one of 'content' or 'tool_calls' must be provided"
+            )
+
+        return field_value
+

 class GenerateChatCompletionForm(BaseModel):
    model: str
    messages: list[ChatMessage]
-    format: Optional[dict] = None
+    format: Optional[Union[dict, str]] = None
    options: Optional[dict] = None
    template: Optional[str] = None
    stream: Optional[bool] = True
    keep_alive: Optional[Union[int, str]] = None
+    tools: Optional[list[dict]] = None


 async def get_ollama_url(request: Request, model: str, url_idx: Optional[int] = None):
@@ -962,7 +975,7 @@ async def get_ollama_url(request: Request, model: str, url_idx: Optional[int] =
            )
        url_idx = random.choice(models[model].get("urls", []))
    url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
-    return url
+    return url, url_idx


@router.post("/api/chat")
@@ -977,6 +990,7 @@ async def generate_chat_completion(
    if BYPASS_MODEL_ACCESS_CONTROL:
        bypass_filter = True

+    metadata = form_data.pop("metadata", None)
    try:
        form_data = GenerateChatCompletionForm(**form_data)
    except Exception as e:
@@ -1006,7 +1020,7 @@ async def generate_chat_completion(
            payload["options"] = apply_model_params_to_body_ollama(
                params, payload["options"]
            )
-            payload = apply_model_system_prompt_to_body(params, payload, user)
+            payload = apply_model_system_prompt_to_body(params, payload, metadata, user)

        # Check if user has access to the model
        if not bypass_filter and user.role == "user":
@@ -1030,7 +1044,7 @@ async def generate_chat_completion(
    if ":" not in payload["model"]:
        payload["model"] = f"{payload['model']}:latest"

-    url = await get_ollama_url(request, payload["model"], url_idx)
+    url, url_idx = await get_ollama_url(request, payload["model"], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
@@ -1132,7 +1146,7 @@ async def generate_openai_completion(
    if ":" not in payload["model"]:
        payload["model"] = f"{payload['model']}:latest"

-    url = await get_ollama_url(request, payload["model"], url_idx)
+    url, url_idx = await get_ollama_url(request, payload["model"], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
@@ -1159,6 +1173,8 @@ async def generate_openai_chat_completion(
    url_idx: Optional[int] = None,
    user=Depends(get_verified_user),
 ):
+    metadata = form_data.pop("metadata", None)
+
    try:
        completion_form = OpenAIChatCompletionForm(**form_data)
    except Exception as e:
@@ -1185,7 +1201,7 @@ async def generate_openai_chat_completion(

        if params:
            payload = apply_model_params_to_body_openai(params, payload)
-            payload = apply_model_system_prompt_to_body(params, payload, user)
+            payload = apply_model_system_prompt_to_body(params, payload, metadata, user)

        # Check if user has access to the model
        if user.role == "user":
@@ -1209,7 +1225,7 @@ async def generate_openai_chat_completion(
    if ":" not in payload["model"]:
        payload["model"] = f"{payload['model']}:latest"

-    url = await get_ollama_url(request, payload["model"], url_idx)
+    url, url_idx = await get_ollama_url(request, payload["model"], url_idx)
    api_config = request.app.state.config.OLLAMA_API_CONFIGS.get(
        str(url_idx),
        request.app.state.config.OLLAMA_API_CONFIGS.get(url, {}),  # Legacy support
@@ -1405,9 +1421,10 @@ async def download_model(
        return None


+# TODO: Progress bar does not reflect size & duration of upload.
@router.post("/models/upload")
@router.post("/models/upload/{url_idx}")
-def upload_model(
+async def upload_model(
    request: Request,
    file: UploadFile = File(...),
    url_idx: Optional[int] = None,
@@ -1416,59 +1433,85 @@ def upload_model(
    if url_idx is None:
        url_idx = 0
    ollama_url = request.app.state.config.OLLAMA_BASE_URLS[url_idx]
+    file_path = os.path.join(UPLOAD_DIR, file.filename)
+    os.makedirs(UPLOAD_DIR, exist_ok=True)

-    file_path = f"{UPLOAD_DIR}/{file.filename}"
+    # --- P1: save file locally ---
+    chunk_size = 1024 * 1024 * 2  # 2 MB chunks
+    with open(file_path, "wb") as out_f:
+        while True:
+            chunk = file.file.read(chunk_size)
+            # log.info(f"Chunk: {str(chunk)}") # DEBUG
+            if not chunk:
+                break
+            out_f.write(chunk)

-    # Save file in chunks
-    with open(file_path, "wb+") as f:
-        for chunk in file.file:
-            f.write(chunk)
-
-    def file_process_stream():
+    async def file_process_stream():
        nonlocal ollama_url
        total_size = os.path.getsize(file_path)
-        chunk_size = 1024 * 1024
+        log.info(f"Total Model Size: {str(total_size)}")  # DEBUG
+
+        # --- P2: SSE progress + calculate sha256 hash ---
+        file_hash = calculate_sha256(file_path, chunk_size)
+        log.info(f"Model Hash: {str(file_hash)}")  # DEBUG
        try:
            with open(file_path, "rb") as f:
-                total = 0
-                done = False
-
-                while not done:
-                    chunk = f.read(chunk_size)
-                    if not chunk:
-                        done = True
-                        continue
-
-                    total += len(chunk)
-                    progress = round((total / total_size) * 100, 2)
-
-                    res = {
+                bytes_read = 0
+                while chunk := f.read(chunk_size):
+                    bytes_read += len(chunk)
+                    progress = round(bytes_read / total_size * 100, 2)
+                    data_msg = {
                        "progress": progress,
                        "total": total_size,
-                        "completed": total,
+                        "completed": bytes_read,
                    }
-                    yield f"data: {json.dumps(res)}\n\n"
+                    yield f"data: {json.dumps(data_msg)}\n\n"

-                if done:
-                    f.seek(0)
-                    hashed = calculate_sha256(f)
-                    f.seek(0)
+            # --- P3: Upload to ollama /api/blobs ---
+            with open(file_path, "rb") as f:
+                url = f"{ollama_url}/api/blobs/sha256:{file_hash}"
+                response = requests.post(url, data=f)

-                    url = f"{ollama_url}/api/blobs/sha256:{hashed}"
-                    response = requests.post(url, data=f)
+            if response.ok:
+                log.info(f"Uploaded to /api/blobs")  # DEBUG
+                # Remove local file
+                os.remove(file_path)

-                    if response.ok:
-                        res = {
-                            "done": done,
-                            "blob": f"sha256:{hashed}",
-                            "name": file.filename,
-                        }
-                        os.remove(file_path)
-                        yield f"data: {json.dumps(res)}\n\n"
-                    else:
-                        raise Exception(
-                            "Ollama: Could not create blob, Please try again."
-                        )
+                # Create model in ollama
+                model_name, ext = os.path.splitext(file.filename)
+                log.info(f"Created Model: {model_name}")  # DEBUG
+
+                create_payload = {
+                    "model": model_name,
+                    # Reference the file by its original name => the uploaded blob's digest
+                    "files": {file.filename: f"sha256:{file_hash}"},
+                }
+                log.info(f"Model Payload: {create_payload}")  # DEBUG
+
+                # Call ollama /api/create
+                # https://github.com/ollama/ollama/blob/main/docs/api.md#create-a-model
+                create_resp = requests.post(
+                    url=f"{ollama_url}/api/create",
+                    headers={"Content-Type": "application/json"},
+                    data=json.dumps(create_payload),
+                )
+
+                if create_resp.ok:
+                    log.info(f"API SUCCESS!")  # DEBUG
+                    done_msg = {
+                        "done": True,
+                        "blob": f"sha256:{file_hash}",
+                        "name": file.filename,
+                        "model_created": model_name,
+                    }
+                    yield f"data: {json.dumps(done_msg)}\n\n"
+                else:
+                    raise Exception(
+                        f"Failed to create model in Ollama. {create_resp.text}"
+                    )
+
+            else:
+                raise Exception("Ollama: Could not create blob, Please try again.")

        except Exception as e:
            res = {"error": str(e)}
@@ -75,9 +75,9 @@ async def cleanup_response(
        await session.close()


-def openai_o1_handler(payload):
+def openai_o1_o3_handler(payload):
    """
-    Handle O1 specific parameters
+    Handle o1, o3 specific parameters
    """
    if "max_tokens" in payload:
        # Remove "max_tokens" from the payload
@@ -266,7 +266,7 @@ async def get_all_models_responses(request: Request) -> list:

    request_tasks = []
    for idx, url in enumerate(request.app.state.config.OPENAI_API_BASE_URLS):
-        if (str(idx) not in request.app.state.config.OPENAI_API_CONFIGS) or (
+        if (str(idx) not in request.app.state.config.OPENAI_API_CONFIGS) and (
            url not in request.app.state.config.OPENAI_API_CONFIGS  # Legacy support
        ):
            request_tasks.append(
@@ -489,7 +489,7 @@ async def get_models(
                raise HTTPException(status_code=500, detail=error_detail)

    if user.role == "user" and not BYPASS_MODEL_ACCESS_CONTROL:
-        models["data"] = get_filtered_models(models, user)
+        models["data"] = await get_filtered_models(models, user)

    return models

@@ -551,9 +551,9 @@ async def generate_chat_completion(
        bypass_filter = True

    idx = 0
+
    payload = {**form_data}
-    if "metadata" in payload:
-        del payload["metadata"]
+    metadata = payload.pop("metadata", None)

    model_id = form_data.get("model")
    model_info = Models.get_model_by_id(model_id)
@@ -566,7 +566,7 @@ async def generate_chat_completion(

        params = model_info.params.model_dump()
        payload = apply_model_params_to_body_openai(params, payload)
-        payload = apply_model_system_prompt_to_body(params, payload, user)
+        payload = apply_model_system_prompt_to_body(params, payload, metadata, user)

        # Check if user has access to the model
        if not bypass_filter and user.role == "user":
@@ -621,10 +621,10 @@ async def generate_chat_completion(
    url = request.app.state.config.OPENAI_API_BASE_URLS[idx]
    key = request.app.state.config.OPENAI_API_KEYS[idx]

-    # Fix: O1 does not support the "max_tokens" parameter, Modify "max_tokens" to "max_completion_tokens"
-    is_o1 = payload["model"].lower().startswith("o1-")
-    if is_o1:
-        payload = openai_o1_handler(payload)
+    # Fix: o1,o3 does not support the "max_tokens" parameter, Modify "max_tokens" to "max_completion_tokens"
+    is_o1_o3 = payload["model"].lower().startswith(("o1", "o3-"))
+    if is_o1_o3:
+        payload = openai_o1_o3_handler(payload)
    elif "api.openai.com" not in url:
        # Remove "max_completion_tokens" from the payload for backward compatibility
        if "max_completion_tokens" in payload:
@@ -9,6 +9,7 @@ from fastapi import (
    status,
    APIRouter,
 )
+import aiohttp
 import os
 import logging
 import shutil
@@ -56,96 +57,103 @@ def get_sorted_filters(model_id, models):
    return sorted_filters


-def process_pipeline_inlet_filter(request, payload, user, models):
+async def process_pipeline_inlet_filter(request, payload, user, models):
    user = {"id": user.id, "email": user.email, "name": user.name, "role": user.role}
    model_id = payload["model"]
-
    sorted_filters = get_sorted_filters(model_id, models)
    model = models[model_id]

    if "pipeline" in model:
        sorted_filters.append(model)

-    for filter in sorted_filters:
-        r = None
-        try:
-            urlIdx = filter["urlIdx"]
+    async with aiohttp.ClientSession() as session:
+        for filter in sorted_filters:
+            urlIdx = filter.get("urlIdx")
+            if urlIdx is None:
+                continue

            url = request.app.state.config.OPENAI_API_BASE_URLS[urlIdx]
            key = request.app.state.config.OPENAI_API_KEYS[urlIdx]

-            if key == "":
+            if not key:
                continue

            headers = {"Authorization": f"Bearer {key}"}
-            r = requests.post(
-                f"{url}/{filter['id']}/filter/inlet",
-                headers=headers,
-                json={
-                    "user": user,
-                    "body": payload,
-                },
-            )
+            request_data = {
+                "user": user,
+                "body": payload,
+            }

-            r.raise_for_status()
-            payload = r.json()
-        except Exception as e:
-            # Handle connection error here
-            print(f"Connection error: {e}")
-
-            if r is not None:
-                res = r.json()
+            try:
+                async with session.post(
+                    f"{url}/{filter['id']}/filter/inlet",
+                    headers=headers,
+                    json=request_data,
+                ) as response:
+                    response.raise_for_status()
+                    payload = await response.json()
+            except aiohttp.ClientResponseError as e:
+                res = (
+                    await response.json()
+                    if response.content_type == "application/json"
+                    else {}
+                )
                if "detail" in res:
-                    raise Exception(r.status_code, res["detail"])
+                    raise Exception(response.status, res["detail"])
+            except Exception as e:
+                print(f"Connection error: {e}")

    return payload


-def process_pipeline_outlet_filter(request, payload, user, models):
+async def process_pipeline_outlet_filter(request, payload, user, models):
    user = {"id": user.id, "email": user.email, "name": user.name, "role": user.role}
    model_id = payload["model"]
-
    sorted_filters = get_sorted_filters(model_id, models)
    model = models[model_id]

    if "pipeline" in model:
        sorted_filters = [model] + sorted_filters

-    for filter in sorted_filters:
-        r = None
-        try:
-            urlIdx = filter["urlIdx"]
+    async with aiohttp.ClientSession() as session:
+        for filter in sorted_filters:
+            urlIdx = filter.get("urlIdx")
+            if urlIdx is None:
+                continue

            url = request.app.state.config.OPENAI_API_BASE_URLS[urlIdx]
            key = request.app.state.config.OPENAI_API_KEYS[urlIdx]

-            if key != "":
-                r = requests.post(
+            if not key:
+                continue
+
+            headers = {"Authorization": f"Bearer {key}"}
+            request_data = {
+                "user": user,
+                "body": payload,
+            }
+
+            try:
+                async with session.post(
                    f"{url}/{filter['id']}/filter/outlet",
-                    headers={"Authorization": f"Bearer {key}"},
-                    json={
-                        "user": user,
-                        "body": payload,
-                    },
-                )
-
-                r.raise_for_status()
-                data = r.json()
-                payload = data
-        except Exception as e:
-            # Handle connection error here
-            print(f"Connection error: {e}")
-
-            if r is not None:
+                    headers=headers,
+                    json=request_data,
+                ) as response:
+                    response.raise_for_status()
+                    payload = await response.json()
+            except aiohttp.ClientResponseError as e:
                try:
-                    res = r.json()
+                    res = (
+                        await response.json()
+                        if "application/json" in response.content_type
+                        else {}
+                    )
                    if "detail" in res:
-                        return Exception(r.status_code, res)
+                        raise Exception(response.status, res)
                except Exception:
                    pass
-
-            else:
-                pass
+            except Exception as e:
+                print(f"Connection error: {e}")

    return payload

@@ -147,7 +147,11 @@ async def delete_prompt_by_command(command: str, user=Depends(get_verified_user)
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if prompt.user_id != user.id and user.role != "admin":
+    if (
+        prompt.user_id != user.id
+        and not has_access(user.id, "write", prompt.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
@@ -21,6 +21,7 @@ from fastapi import (
    APIRouter,
 )
 from fastapi.middleware.cors import CORSMiddleware
+from fastapi.concurrency import run_in_threadpool
 from pydantic import BaseModel
 import tiktoken

@@ -45,16 +46,19 @@ from open_webui.retrieval.web.utils import get_web_loader
 from open_webui.retrieval.web.brave import search_brave
 from open_webui.retrieval.web.kagi import search_kagi
 from open_webui.retrieval.web.mojeek import search_mojeek
+from open_webui.retrieval.web.bocha import search_bocha
 from open_webui.retrieval.web.duckduckgo import search_duckduckgo
 from open_webui.retrieval.web.google_pse import search_google_pse
 from open_webui.retrieval.web.jina_search import search_jina
 from open_webui.retrieval.web.searchapi import search_searchapi
+from open_webui.retrieval.web.serpapi import search_serpapi
 from open_webui.retrieval.web.searxng import search_searxng
 from open_webui.retrieval.web.serper import search_serper
 from open_webui.retrieval.web.serply import search_serply
 from open_webui.retrieval.web.serpstack import search_serpstack
 from open_webui.retrieval.web.tavily import search_tavily
 from open_webui.retrieval.web.bing import search_bing
+from open_webui.retrieval.web.exa import search_exa


 from open_webui.retrieval.utils import (
@@ -347,6 +351,7 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
    return {
        "status": True,
        "pdf_extract_images": request.app.state.config.PDF_EXTRACT_IMAGES,
+        "RAG_FULL_CONTEXT": request.app.state.config.RAG_FULL_CONTEXT,
        "enable_google_drive_integration": request.app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
        "content_extraction": {
            "engine": request.app.state.config.CONTENT_EXTRACTION_ENGINE,
@@ -367,7 +372,8 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
            "proxy_url": request.app.state.config.YOUTUBE_LOADER_PROXY_URL,
        },
        "web": {
-            "web_loader_ssl_verification": request.app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
+            "ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION": request.app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
+            "RAG_WEB_SEARCH_FULL_CONTEXT": request.app.state.config.RAG_WEB_SEARCH_FULL_CONTEXT,
            "search": {
                "enabled": request.app.state.config.ENABLE_RAG_WEB_SEARCH,
                "drive": request.app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION,
@@ -378,6 +384,7 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
                "brave_search_api_key": request.app.state.config.BRAVE_SEARCH_API_KEY,
                "kagi_search_api_key": request.app.state.config.KAGI_SEARCH_API_KEY,
                "mojeek_search_api_key": request.app.state.config.MOJEEK_SEARCH_API_KEY,
+                "bocha_search_api_key": request.app.state.config.BOCHA_SEARCH_API_KEY,
                "serpstack_api_key": request.app.state.config.SERPSTACK_API_KEY,
                "serpstack_https": request.app.state.config.SERPSTACK_HTTPS,
                "serper_api_key": request.app.state.config.SERPER_API_KEY,
@@ -385,11 +392,15 @@ async def get_rag_config(request: Request, user=Depends(get_admin_user)):
                "tavily_api_key": request.app.state.config.TAVILY_API_KEY,
                "searchapi_api_key": request.app.state.config.SEARCHAPI_API_KEY,
                "searchapi_engine": request.app.state.config.SEARCHAPI_ENGINE,
+                "serpapi_api_key": request.app.state.config.SERPAPI_API_KEY,
+                "serpapi_engine": request.app.state.config.SERPAPI_ENGINE,
                "jina_api_key": request.app.state.config.JINA_API_KEY,
                "bing_search_v7_endpoint": request.app.state.config.BING_SEARCH_V7_ENDPOINT,
                "bing_search_v7_subscription_key": request.app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY,
+                "exa_api_key": request.app.state.config.EXA_API_KEY,
                "result_count": request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
                "concurrent_requests": request.app.state.config.RAG_WEB_SEARCH_CONCURRENT_REQUESTS,
+                "domain_filter_list": request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
            },
        },
    }
@@ -426,6 +437,7 @@ class WebSearchConfig(BaseModel):
    brave_search_api_key: Optional[str] = None
    kagi_search_api_key: Optional[str] = None
    mojeek_search_api_key: Optional[str] = None
+    bocha_search_api_key: Optional[str] = None
    serpstack_api_key: Optional[str] = None
    serpstack_https: Optional[bool] = None
    serper_api_key: Optional[str] = None
@@ -433,19 +445,26 @@ class WebSearchConfig(BaseModel):
    tavily_api_key: Optional[str] = None
    searchapi_api_key: Optional[str] = None
    searchapi_engine: Optional[str] = None
+    serpapi_api_key: Optional[str] = None
+    serpapi_engine: Optional[str] = None
    jina_api_key: Optional[str] = None
    bing_search_v7_endpoint: Optional[str] = None
    bing_search_v7_subscription_key: Optional[str] = None
+    exa_api_key: Optional[str] = None
    result_count: Optional[int] = None
    concurrent_requests: Optional[int] = None
+    trust_env: Optional[bool] = None
+    domain_filter_list: Optional[List[str]] = []


 class WebConfig(BaseModel):
    search: WebSearchConfig
-    web_loader_ssl_verification: Optional[bool] = None
+    ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION: Optional[bool] = None
+    RAG_WEB_SEARCH_FULL_CONTEXT: Optional[bool] = None


 class ConfigUpdateForm(BaseModel):
+    RAG_FULL_CONTEXT: Optional[bool] = None
    pdf_extract_images: Optional[bool] = None
    enable_google_drive_integration: Optional[bool] = None
    file: Optional[FileConfig] = None
@@ -465,6 +484,12 @@ async def update_rag_config(
        else request.app.state.config.PDF_EXTRACT_IMAGES
    )

+    request.app.state.config.RAG_FULL_CONTEXT = (
+        form_data.RAG_FULL_CONTEXT
+        if form_data.RAG_FULL_CONTEXT is not None
+        else request.app.state.config.RAG_FULL_CONTEXT
+    )
+
    request.app.state.config.ENABLE_GOOGLE_DRIVE_INTEGRATION = (
        form_data.enable_google_drive_integration
        if form_data.enable_google_drive_integration is not None
@@ -497,11 +522,16 @@ async def update_rag_config(
    if form_data.web is not None:
        request.app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION = (
            # Note: When UI "Bypass SSL verification for Websites"=True then ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION=False
-            form_data.web.web_loader_ssl_verification
+            form_data.web.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION
        )

        request.app.state.config.ENABLE_RAG_WEB_SEARCH = form_data.web.search.enabled
        request.app.state.config.RAG_WEB_SEARCH_ENGINE = form_data.web.search.engine
+
+        request.app.state.config.RAG_WEB_SEARCH_FULL_CONTEXT = (
+            form_data.web.RAG_WEB_SEARCH_FULL_CONTEXT
+        )
+
        request.app.state.config.SEARXNG_QUERY_URL = (
            form_data.web.search.searxng_query_url
        )
@@ -520,6 +550,9 @@ async def update_rag_config(
        request.app.state.config.MOJEEK_SEARCH_API_KEY = (
            form_data.web.search.mojeek_search_api_key
        )
+        request.app.state.config.BOCHA_SEARCH_API_KEY = (
+            form_data.web.search.bocha_search_api_key
+        )
        request.app.state.config.SERPSTACK_API_KEY = (
            form_data.web.search.serpstack_api_key
        )
@@ -534,6 +567,9 @@ async def update_rag_config(
            form_data.web.search.searchapi_engine
        )

+        request.app.state.config.SERPAPI_API_KEY = form_data.web.search.serpapi_api_key
+        request.app.state.config.SERPAPI_ENGINE = form_data.web.search.serpapi_engine
+
        request.app.state.config.JINA_API_KEY = form_data.web.search.jina_api_key
        request.app.state.config.BING_SEARCH_V7_ENDPOINT = (
            form_data.web.search.bing_search_v7_endpoint
@@ -542,16 +578,25 @@ async def update_rag_config(
            form_data.web.search.bing_search_v7_subscription_key
        )

+        request.app.state.config.EXA_API_KEY = form_data.web.search.exa_api_key
+
        request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT = (
            form_data.web.search.result_count
        )
        request.app.state.config.RAG_WEB_SEARCH_CONCURRENT_REQUESTS = (
            form_data.web.search.concurrent_requests
        )
+        request.app.state.config.RAG_WEB_SEARCH_TRUST_ENV = (
+            form_data.web.search.trust_env
+        )
+        request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST = (
+            form_data.web.search.domain_filter_list
+        )

    return {
        "status": True,
        "pdf_extract_images": request.app.state.config.PDF_EXTRACT_IMAGES,
+        "RAG_FULL_CONTEXT": request.app.state.config.RAG_FULL_CONTEXT,
        "file": {
            "max_size": request.app.state.config.FILE_MAX_SIZE,
            "max_count": request.app.state.config.FILE_MAX_COUNT,
@@ -571,7 +616,8 @@ async def update_rag_config(
            "translation": request.app.state.YOUTUBE_LOADER_TRANSLATION,
        },
        "web": {
-            "web_loader_ssl_verification": request.app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
+            "ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION": request.app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
+            "RAG_WEB_SEARCH_FULL_CONTEXT": request.app.state.config.RAG_WEB_SEARCH_FULL_CONTEXT,
            "search": {
                "enabled": request.app.state.config.ENABLE_RAG_WEB_SEARCH,
                "engine": request.app.state.config.RAG_WEB_SEARCH_ENGINE,
@@ -581,18 +627,24 @@ async def update_rag_config(
                "brave_search_api_key": request.app.state.config.BRAVE_SEARCH_API_KEY,
                "kagi_search_api_key": request.app.state.config.KAGI_SEARCH_API_KEY,
                "mojeek_search_api_key": request.app.state.config.MOJEEK_SEARCH_API_KEY,
+                "bocha_search_api_key": request.app.state.config.BOCHA_SEARCH_API_KEY,
                "serpstack_api_key": request.app.state.config.SERPSTACK_API_KEY,
                "serpstack_https": request.app.state.config.SERPSTACK_HTTPS,
                "serper_api_key": request.app.state.config.SERPER_API_KEY,
                "serply_api_key": request.app.state.config.SERPLY_API_KEY,
                "serachapi_api_key": request.app.state.config.SEARCHAPI_API_KEY,
                "searchapi_engine": request.app.state.config.SEARCHAPI_ENGINE,
+                "serpapi_api_key": request.app.state.config.SERPAPI_API_KEY,
+                "serpapi_engine": request.app.state.config.SERPAPI_ENGINE,
                "tavily_api_key": request.app.state.config.TAVILY_API_KEY,
                "jina_api_key": request.app.state.config.JINA_API_KEY,
                "bing_search_v7_endpoint": request.app.state.config.BING_SEARCH_V7_ENDPOINT,
                "bing_search_v7_subscription_key": request.app.state.config.BING_SEARCH_V7_SUBSCRIPTION_KEY,
+                "exa_api_key": request.app.state.config.EXA_API_KEY,
                "result_count": request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
                "concurrent_requests": request.app.state.config.RAG_WEB_SEARCH_CONCURRENT_REQUESTS,
+                "trust_env": request.app.state.config.RAG_WEB_SEARCH_TRUST_ENV,
+                "domain_filter_list": request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
            },
        },
    }
@@ -660,6 +712,7 @@ def save_docs_to_vector_db(
    overwrite: bool = False,
    split: bool = True,
    add: bool = False,
+    user=None,
 ) -> bool:
    def _get_docs_info(docs: list[Document]) -> str:
        docs_info = set()
@@ -740,7 +793,11 @@ def save_docs_to_vector_db(
    # for meta-data so convert them to string.
    for metadata in metadatas:
        for key, value in metadata.items():
-            if isinstance(value, datetime):
+            if (
+                isinstance(value, datetime)
+                or isinstance(value, list)
+                or isinstance(value, dict)
+            ):
                metadata[key] = str(value)

    try:
@@ -775,7 +832,7 @@ def save_docs_to_vector_db(
        )

        embeddings = embedding_function(
-            list(map(lambda x: x.replace("\n", " "), texts))
+            list(map(lambda x: x.replace("\n", " "), texts)), user=user
        )

        items = [
@@ -933,6 +990,7 @@ def process_file(
                    "hash": hash,
                },
                add=(True if form_data.collection_name else False),
+                user=user,
            )

            if result:
@@ -990,7 +1048,7 @@ def process_text(
    text_content = form_data.content
    log.debug(f"text_content: {text_content}")

-    result = save_docs_to_vector_db(request, docs, collection_name)
+    result = save_docs_to_vector_db(request, docs, collection_name, user=user)
    if result:
        return {
            "status": True,
@@ -1023,7 +1081,9 @@ def process_youtube_video(
        content = " ".join([doc.page_content for doc in docs])
        log.debug(f"text_content: {content}")

-        save_docs_to_vector_db(request, docs, collection_name, overwrite=True)
+        save_docs_to_vector_db(
+            request, docs, collection_name, overwrite=True, user=user
+        )

        return {
            "status": True,
@@ -1064,7 +1124,9 @@ def process_web(
        content = " ".join([doc.page_content for doc in docs])

        log.debug(f"text_content: {content}")
-        save_docs_to_vector_db(request, docs, collection_name, overwrite=True)
+        save_docs_to_vector_db(
+            request, docs, collection_name, overwrite=True, user=user
+        )

        return {
            "status": True,
@@ -1095,11 +1157,14 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
    - BRAVE_SEARCH_API_KEY
    - KAGI_SEARCH_API_KEY
    - MOJEEK_SEARCH_API_KEY
+    - BOCHA_SEARCH_API_KEY
    - SERPSTACK_API_KEY
    - SERPER_API_KEY
    - SERPLY_API_KEY
    - TAVILY_API_KEY
+    - EXA_API_KEY
    - SEARCHAPI_API_KEY + SEARCHAPI_ENGINE (by default `google`)
+    - SERPAPI_API_KEY + SERPAPI_ENGINE (by default `google`)
    Args:
        query (str): The query to search for
    """
@@ -1161,6 +1226,16 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
            )
        else:
            raise Exception("No MOJEEK_SEARCH_API_KEY found in environment variables")
+    elif engine == "bocha":
+        if request.app.state.config.BOCHA_SEARCH_API_KEY:
+            return search_bocha(
+                request.app.state.config.BOCHA_SEARCH_API_KEY,
+                query,
+                request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
+                request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
+            )
+        else:
+            raise Exception("No BOCHA_SEARCH_API_KEY found in environment variables")
    elif engine == "serpstack":
        if request.app.state.config.SERPSTACK_API_KEY:
            return search_serpstack(
@@ -1204,6 +1279,7 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
                request.app.state.config.TAVILY_API_KEY,
                query,
                request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
+                request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
            )
        else:
            raise Exception("No TAVILY_API_KEY found in environment variables")
@@ -1218,6 +1294,17 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
            )
        else:
            raise Exception("No SEARCHAPI_API_KEY found in environment variables")
+    elif engine == "serpapi":
+        if request.app.state.config.SERPAPI_API_KEY:
+            return search_serpapi(
+                request.app.state.config.SERPAPI_API_KEY,
+                request.app.state.config.SERPAPI_ENGINE,
+                query,
+                request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
+                request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
+            )
+        else:
+            raise Exception("No SERPAPI_API_KEY found in environment variables")
    elif engine == "jina":
        return search_jina(
            request.app.state.config.JINA_API_KEY,
@@ -1233,12 +1320,19 @@ def search_web(request: Request, engine: str, query: str) -> list[SearchResult]:
            request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
            request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
        )
+    elif engine == "exa":
+        return search_exa(
+            request.app.state.config.EXA_API_KEY,
+            query,
+            request.app.state.config.RAG_WEB_SEARCH_RESULT_COUNT,
+            request.app.state.config.RAG_WEB_SEARCH_DOMAIN_FILTER_LIST,
+        )
    else:
        raise Exception("No search engine API key found in environment variables")


@router.post("/process/web/search")
-def process_web_search(
+async def process_web_search(
    request: Request, form_data: SearchForm, user=Depends(get_verified_user)
 ):
    try:
@@ -1270,15 +1364,39 @@ def process_web_search(
            urls,
            verify_ssl=request.app.state.config.ENABLE_RAG_WEB_LOADER_SSL_VERIFICATION,
            requests_per_second=request.app.state.config.RAG_WEB_SEARCH_CONCURRENT_REQUESTS,
+            trust_env=request.app.state.config.RAG_WEB_SEARCH_TRUST_ENV,
        )
-        docs = loader.load()
-        save_docs_to_vector_db(request, docs, collection_name, overwrite=True)
+        docs = await loader.aload()

-        return {
-            "status": True,
-            "collection_name": collection_name,
-            "filenames": urls,
-        }
+        if request.app.state.config.RAG_WEB_SEARCH_FULL_CONTEXT:
+            return {
+                "status": True,
+                "docs": [
+                    {
+                        "content": doc.page_content,
+                        "metadata": doc.metadata,
+                    }
+                    for doc in docs
+                ],
+                "filenames": urls,
+                "loaded_count": len(docs),
+            }
+        else:
+            await run_in_threadpool(
+                save_docs_to_vector_db,
+                request,
+                docs,
+                collection_name,
+                overwrite=True,
+                user=user,
+            )
+
+            return {
+                "status": True,
+                "collection_name": collection_name,
+                "filenames": urls,
+                "loaded_count": len(docs),
+            }
    except Exception as e:
        log.exception(e)
        raise HTTPException(
@@ -1306,7 +1424,9 @@ def query_doc_handler(
            return query_doc_with_hybrid_search(
                collection_name=form_data.collection_name,
                query=form_data.query,
-                embedding_function=request.app.state.EMBEDDING_FUNCTION,
+                embedding_function=lambda query: request.app.state.EMBEDDING_FUNCTION(
+                    query, user=user
+                ),
                k=form_data.k if form_data.k else request.app.state.config.TOP_K,
                reranking_function=request.app.state.rf,
                r=(
@@ -1314,12 +1434,16 @@ def query_doc_handler(
                    if form_data.r
                    else request.app.state.config.RELEVANCE_THRESHOLD
                ),
+                user=user,
            )
        else:
            return query_doc(
                collection_name=form_data.collection_name,
-                query_embedding=request.app.state.EMBEDDING_FUNCTION(form_data.query),
+                query_embedding=request.app.state.EMBEDDING_FUNCTION(
+                    form_data.query, user=user
+                ),
                k=form_data.k if form_data.k else request.app.state.config.TOP_K,
+                user=user,
            )
    except Exception as e:
        log.exception(e)
@@ -1348,7 +1472,9 @@ def query_collection_handler(
            return query_collection_with_hybrid_search(
                collection_names=form_data.collection_names,
                queries=[form_data.query],
-                embedding_function=request.app.state.EMBEDDING_FUNCTION,
+                embedding_function=lambda query: request.app.state.EMBEDDING_FUNCTION(
+                    query, user=user
+                ),
                k=form_data.k if form_data.k else request.app.state.config.TOP_K,
                reranking_function=request.app.state.rf,
                r=(
@@ -1361,7 +1487,9 @@ def query_collection_handler(
            return query_collection(
                collection_names=form_data.collection_names,
                queries=[form_data.query],
-                embedding_function=request.app.state.EMBEDDING_FUNCTION,
+                embedding_function=lambda query: request.app.state.EMBEDDING_FUNCTION(
+                    query, user=user
+                ),
                k=form_data.k if form_data.k else request.app.state.config.TOP_K,
            )

@@ -1509,6 +1637,7 @@ def process_files_batch(
                docs=all_docs,
                collection_name=collection_name,
                add=True,
+                user=user,
            )

            # Update all files with collection name
@@ -4,6 +4,7 @@ from fastapi.responses import JSONResponse, RedirectResponse
 from pydantic import BaseModel
 from typing import Optional
 import logging
+import re

 from open_webui.utils.chat import generate_chat_completion
 from open_webui.utils.task import (
@@ -57,6 +58,7 @@ async def get_task_config(request: Request, user=Depends(get_verified_user)):
        "AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH": request.app.state.config.AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH,
        "TAGS_GENERATION_PROMPT_TEMPLATE": request.app.state.config.TAGS_GENERATION_PROMPT_TEMPLATE,
        "ENABLE_TAGS_GENERATION": request.app.state.config.ENABLE_TAGS_GENERATION,
+        "ENABLE_TITLE_GENERATION": request.app.state.config.ENABLE_TITLE_GENERATION,
        "ENABLE_SEARCH_QUERY_GENERATION": request.app.state.config.ENABLE_SEARCH_QUERY_GENERATION,
        "ENABLE_RETRIEVAL_QUERY_GENERATION": request.app.state.config.ENABLE_RETRIEVAL_QUERY_GENERATION,
        "QUERY_GENERATION_PROMPT_TEMPLATE": request.app.state.config.QUERY_GENERATION_PROMPT_TEMPLATE,
@@ -67,6 +69,7 @@ async def get_task_config(request: Request, user=Depends(get_verified_user)):
 class TaskConfigForm(BaseModel):
    TASK_MODEL: Optional[str]
    TASK_MODEL_EXTERNAL: Optional[str]
+    ENABLE_TITLE_GENERATION: bool
    TITLE_GENERATION_PROMPT_TEMPLATE: str
    IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE: str
    ENABLE_AUTOCOMPLETE_GENERATION: bool
@@ -85,10 +88,15 @@ async def update_task_config(
 ):
    request.app.state.config.TASK_MODEL = form_data.TASK_MODEL
    request.app.state.config.TASK_MODEL_EXTERNAL = form_data.TASK_MODEL_EXTERNAL
+    request.app.state.config.ENABLE_TITLE_GENERATION = form_data.ENABLE_TITLE_GENERATION
    request.app.state.config.TITLE_GENERATION_PROMPT_TEMPLATE = (
        form_data.TITLE_GENERATION_PROMPT_TEMPLATE
    )

+    request.app.state.config.IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE = (
+        form_data.IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE
+    )
+
    request.app.state.config.ENABLE_AUTOCOMPLETE_GENERATION = (
        form_data.ENABLE_AUTOCOMPLETE_GENERATION
    )
@@ -117,6 +125,7 @@ async def update_task_config(
    return {
        "TASK_MODEL": request.app.state.config.TASK_MODEL,
        "TASK_MODEL_EXTERNAL": request.app.state.config.TASK_MODEL_EXTERNAL,
+        "ENABLE_TITLE_GENERATION": request.app.state.config.ENABLE_TITLE_GENERATION,
        "TITLE_GENERATION_PROMPT_TEMPLATE": request.app.state.config.TITLE_GENERATION_PROMPT_TEMPLATE,
        "IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE": request.app.state.config.IMAGE_PROMPT_GENERATION_PROMPT_TEMPLATE,
        "ENABLE_AUTOCOMPLETE_GENERATION": request.app.state.config.ENABLE_AUTOCOMPLETE_GENERATION,
@@ -134,7 +143,19 @@ async def update_task_config(
 async def generate_title(
    request: Request, form_data: dict, user=Depends(get_verified_user)
 ):
-    models = request.app.state.MODELS
+
+    if not request.app.state.config.ENABLE_TITLE_GENERATION:
+        return JSONResponse(
+            status_code=status.HTTP_200_OK,
+            content={"detail": "Title generation is disabled"},
+        )
+
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
@@ -161,9 +182,20 @@ async def generate_title(
    else:
        template = DEFAULT_TITLE_GENERATION_PROMPT_TEMPLATE

+    messages = form_data["messages"]
+
+    # Remove reasoning details from the messages
+    for message in messages:
+        message["content"] = re.sub(
+            r"<details\s+type=\"reasoning\"[^>]*>.*?<\/details>",
+            "",
+            message["content"],
+            flags=re.S,
+        ).strip()
+
    content = title_generation_template(
        template,
-        form_data["messages"],
+        messages,
        {
            "name": user.name,
            "location": user.info.get("location") if user.info else None,
@@ -175,13 +207,14 @@ async def generate_title(
        "messages": [{"role": "user", "content": content}],
        "stream": False,
        **(
-            {"max_tokens": 50}
-            if models[task_model_id]["owned_by"] == "ollama"
+            {"max_tokens": 1000}
+            if models[task_model_id].get("owned_by") == "ollama"
            else {
-                "max_completion_tokens": 50,
+                "max_completion_tokens": 1000,
            }
        ),
        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
            "task": str(TASKS.TITLE_GENERATION),
            "task_body": form_data,
            "chat_id": form_data.get("chat_id", None),
@@ -209,7 +242,12 @@ async def generate_chat_tags(
            content={"detail": "Tags generation is disabled"},
        )

-    models = request.app.state.MODELS
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
@@ -245,6 +283,7 @@ async def generate_chat_tags(
        "messages": [{"role": "user", "content": content}],
        "stream": False,
        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
            "task": str(TASKS.TAGS_GENERATION),
            "task_body": form_data,
            "chat_id": form_data.get("chat_id", None),
@@ -265,7 +304,12 @@ async def generate_chat_tags(
 async def generate_image_prompt(
    request: Request, form_data: dict, user=Depends(get_verified_user)
 ):
-    models = request.app.state.MODELS
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
@@ -305,6 +349,7 @@ async def generate_image_prompt(
        "messages": [{"role": "user", "content": content}],
        "stream": False,
        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
            "task": str(TASKS.IMAGE_PROMPT_GENERATION),
            "task_body": form_data,
            "chat_id": form_data.get("chat_id", None),
@@ -340,7 +385,12 @@ async def generate_queries(
                detail=f"Query generation is disabled",
            )

-    models = request.app.state.MODELS
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
@@ -376,6 +426,7 @@ async def generate_queries(
        "messages": [{"role": "user", "content": content}],
        "stream": False,
        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
            "task": str(TASKS.QUERY_GENERATION),
            "task_body": form_data,
            "chat_id": form_data.get("chat_id", None),
@@ -415,7 +466,12 @@ async def generate_autocompletion(
                detail=f"Input prompt exceeds maximum length of {request.app.state.config.AUTOCOMPLETE_GENERATION_INPUT_MAX_LENGTH}",
            )

-    models = request.app.state.MODELS
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
@@ -451,6 +507,7 @@ async def generate_autocompletion(
        "messages": [{"role": "user", "content": content}],
        "stream": False,
        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
            "task": str(TASKS.AUTOCOMPLETE_GENERATION),
            "task_body": form_data,
            "chat_id": form_data.get("chat_id", None),
@@ -472,7 +529,12 @@ async def generate_emoji(
    request: Request, form_data: dict, user=Depends(get_verified_user)
 ):

-    models = request.app.state.MODELS
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
@@ -509,13 +571,17 @@ async def generate_emoji(
        "stream": False,
        **(
            {"max_tokens": 4}
-            if models[task_model_id]["owned_by"] == "ollama"
+            if models[task_model_id].get("owned_by") == "ollama"
            else {
                "max_completion_tokens": 4,
            }
        ),
        "chat_id": form_data.get("chat_id", None),
-        "metadata": {"task": str(TASKS.EMOJI_GENERATION), "task_body": form_data},
+        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
+            "task": str(TASKS.EMOJI_GENERATION),
+            "task_body": form_data,
+        },
    }

    try:
@@ -532,7 +598,13 @@ async def generate_moa_response(
    request: Request, form_data: dict, user=Depends(get_verified_user)
 ):

-    models = request.app.state.MODELS
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS
+
    model_id = form_data["model"]

    if model_id not in models:
@@ -565,6 +637,7 @@ async def generate_moa_response(
        "messages": [{"role": "user", "content": content}],
        "stream": form_data.get("stream", False),
        "metadata": {
+            **(request.state.metadata if hasattr(request.state, "metadata") else {}),
            "chat_id": form_data.get("chat_id", None),
            "task": str(TASKS.MOA_RESPONSE_GENERATION),
            "task_body": form_data,
@@ -227,7 +227,11 @@ async def delete_tools_by_id(
            detail=ERROR_MESSAGES.NOT_FOUND,
        )

-    if tools.user_id != user.id and user.role != "admin":
+    if (
+        tools.user_id != user.id
+        and not has_access(user.id, "write", tools.access_control)
+        and user.role != "admin"
+    ):
        raise HTTPException(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.UNAUTHORIZED,
@@ -309,6 +313,17 @@ async def update_tools_valves_by_id(
            status_code=status.HTTP_401_UNAUTHORIZED,
            detail=ERROR_MESSAGES.NOT_FOUND,
        )
+
+    if (
+        tools.user_id != user.id
+        and not has_access(user.id, "write", tools.access_control)
+        and user.role != "admin"
+    ):
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+        )
+
    if id in request.app.state.TOOLS:
        tools_module = request.app.state.TOOLS[id]
    else:
@@ -79,6 +79,7 @@ class ChatPermissions(BaseModel):
 class FeaturesPermissions(BaseModel):
    web_search: bool = True
    image_generation: bool = True
+    code_interpreter: bool = True


 class UserPermissions(BaseModel):
@@ -152,7 +153,7 @@ async def get_user_settings_by_session_user(user=Depends(get_verified_user)):
 async def update_user_settings_by_session_user(
    form_data: UserSettings, user=Depends(get_verified_user)
 ):
-    user = Users.update_user_by_id(user.id, {"settings": form_data.model_dump()})
+    user = Users.update_user_settings_by_id(user.id, form_data.model_dump())
    if user:
        return user.settings
    else:
@@ -4,45 +4,76 @@ import markdown
 from open_webui.models.chats import ChatTitleMessagesForm
 from open_webui.config import DATA_DIR, ENABLE_ADMIN_EXPORT
 from open_webui.constants import ERROR_MESSAGES
-from fastapi import APIRouter, Depends, HTTPException, Response, status
+from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
 from pydantic import BaseModel
 from starlette.responses import FileResponse
+
+
 from open_webui.utils.misc import get_gravatar_url
 from open_webui.utils.pdf_generator import PDFGenerator
-from open_webui.utils.auth import get_admin_user
+from open_webui.utils.auth import get_admin_user, get_verified_user
+from open_webui.utils.code_interpreter import execute_code_jupyter
+

 router = APIRouter()


@router.get("/gravatar")
-async def get_gravatar(
-    email: str,
-):
+async def get_gravatar(email: str, user=Depends(get_verified_user)):
    return get_gravatar_url(email)


-class CodeFormatRequest(BaseModel):
+class CodeForm(BaseModel):
    code: str


@router.post("/code/format")
-async def format_code(request: CodeFormatRequest):
+async def format_code(form_data: CodeForm, user=Depends(get_verified_user)):
    try:
-        formatted_code = black.format_str(request.code, mode=black.Mode())
+        formatted_code = black.format_str(form_data.code, mode=black.Mode())
        return {"code": formatted_code}
    except black.NothingChanged:
-        return {"code": request.code}
+        return {"code": form_data.code}
    except Exception as e:
        raise HTTPException(status_code=400, detail=str(e))


+@router.post("/code/execute")
+async def execute_code(
+    request: Request, form_data: CodeForm, user=Depends(get_verified_user)
+):
+    if request.app.state.config.CODE_EXECUTION_ENGINE == "jupyter":
+        output = await execute_code_jupyter(
+            request.app.state.config.CODE_EXECUTION_JUPYTER_URL,
+            form_data.code,
+            (
+                request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_TOKEN
+                if request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH == "token"
+                else None
+            ),
+            (
+                request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD
+                if request.app.state.config.CODE_EXECUTION_JUPYTER_AUTH == "password"
+                else None
+            ),
+            request.app.state.config.CODE_EXECUTION_JUPYTER_TIMEOUT,
+        )
+
+        return output
+    else:
+        raise HTTPException(
+            status_code=400,
+            detail="Code execution engine not supported",
+        )
+
+
 class MarkdownForm(BaseModel):
    md: str


@router.post("/markdown")
 async def get_html_from_markdown(
-    form_data: MarkdownForm,
+    form_data: MarkdownForm, user=Depends(get_verified_user)
 ):
    return {"html": markdown.markdown(form_data.md)}

@@ -54,7 +85,7 @@ class ChatForm(BaseModel):

@router.post("/pdf")
 async def download_chat_as_pdf(
-    form_data: ChatTitleMessagesForm,
+    form_data: ChatTitleMessagesForm, user=Depends(get_verified_user)
 ):
    try:
        pdf_bytes = PDFGenerator(form_data).generate_chat_pdf()
@@ -279,8 +279,8 @@ def get_event_emitter(request_info):
            await sio.emit(
                "chat-events",
                {
-                    "chat_id": request_info["chat_id"],
-                    "message_id": request_info["message_id"],
+                    "chat_id": request_info.get("chat_id", None),
+                    "message_id": request_info.get("message_id", None),
                    "data": event_data,
                },
                to=session_id,
@@ -325,19 +325,22 @@ def get_event_emitter(request_info):


 def get_event_call(request_info):
-    async def __event_call__(event_data):
+    async def __event_caller__(event_data):
        response = await sio.call(
            "chat-events",
            {
-                "chat_id": request_info["chat_id"],
-                "message_id": request_info["message_id"],
+                "chat_id": request_info.get("chat_id", None),
+                "message_id": request_info.get("message_id", None),
                "data": event_data,
            },
            to=request_info["session_id"],
        )
        return response

-    return __event_call__
+    return __event_caller__
+
+
+get_event_caller = get_event_call


 def get_user_id_from_session_pool(sid):
@@ -9308,5 +9308,3 @@
 	.json-schema-2020-12__title:first-of-type {
 	font-size: 16px;
 }
-
-/*# sourceMappingURL=swagger-ui.css.map*/
@@ -10,16 +10,23 @@ from open_webui.config import (
    S3_ACCESS_KEY_ID,
    S3_BUCKET_NAME,
    S3_ENDPOINT_URL,
+    S3_KEY_PREFIX,
    S3_REGION_NAME,
    S3_SECRET_ACCESS_KEY,
    GCS_BUCKET_NAME,
    GOOGLE_APPLICATION_CREDENTIALS_JSON,
+    AZURE_STORAGE_ENDPOINT,
+    AZURE_STORAGE_CONTAINER_NAME,
+    AZURE_STORAGE_KEY,
    STORAGE_PROVIDER,
    UPLOAD_DIR,
 )
 from google.cloud import storage
 from google.cloud.exceptions import GoogleCloudError, NotFound
 from open_webui.constants import ERROR_MESSAGES
+from azure.identity import DefaultAzureCredential
+from azure.storage.blob import BlobServiceClient
+from azure.core.exceptions import ResourceNotFoundError


 class StorageProvider(ABC):
@@ -93,15 +100,17 @@ class S3StorageProvider(StorageProvider):
            aws_secret_access_key=S3_SECRET_ACCESS_KEY,
        )
        self.bucket_name = S3_BUCKET_NAME
+        self.key_prefix = S3_KEY_PREFIX if S3_KEY_PREFIX else ""

    def upload_file(self, file: BinaryIO, filename: str) -> Tuple[bytes, str]:
        """Handles uploading of the file to S3 storage."""
        _, file_path = LocalStorageProvider.upload_file(file, filename)
        try:
-            self.s3_client.upload_file(file_path, self.bucket_name, filename)
+            s3_key = os.path.join(self.key_prefix, filename)
+            self.s3_client.upload_file(file_path, self.bucket_name, s3_key)
            return (
                open(file_path, "rb").read(),
-                "s3://" + self.bucket_name + "/" + filename,
+                "s3://" + self.bucket_name + "/" + s3_key,
            )
        except ClientError as e:
            raise RuntimeError(f"Error uploading file to S3: {e}")
@@ -109,18 +118,18 @@ class S3StorageProvider(StorageProvider):
    def get_file(self, file_path: str) -> str:
        """Handles downloading of the file from S3 storage."""
        try:
-            bucket_name, key = file_path.split("//")[1].split("/")
-            local_file_path = f"{UPLOAD_DIR}/{key}"
-            self.s3_client.download_file(bucket_name, key, local_file_path)
+            s3_key = self._extract_s3_key(file_path)
+            local_file_path = self._get_local_file_path(s3_key)
+            self.s3_client.download_file(self.bucket_name, s3_key, local_file_path)
            return local_file_path
        except ClientError as e:
            raise RuntimeError(f"Error downloading file from S3: {e}")

    def delete_file(self, file_path: str) -> None:
        """Handles deletion of the file from S3 storage."""
-        filename = file_path.split("/")[-1]
        try:
-            self.s3_client.delete_object(Bucket=self.bucket_name, Key=filename)
+            s3_key = self._extract_s3_key(file_path)
+            self.s3_client.delete_object(Bucket=self.bucket_name, Key=s3_key)
        except ClientError as e:
            raise RuntimeError(f"Error deleting file from S3: {e}")

@@ -133,6 +142,10 @@ class S3StorageProvider(StorageProvider):
            response = self.s3_client.list_objects_v2(Bucket=self.bucket_name)
            if "Contents" in response:
                for content in response["Contents"]:
+                    # Skip objects that were not uploaded from open-webui in the first place
+                    if not content["Key"].startswith(self.key_prefix):
+                        continue
+
                    self.s3_client.delete_object(
                        Bucket=self.bucket_name, Key=content["Key"]
                    )
@@ -142,6 +155,13 @@ class S3StorageProvider(StorageProvider):
        # Always delete from local storage
        LocalStorageProvider.delete_all_files()

+    # The s3 key is the name assigned to an object. It excludes the bucket name, but includes the internal path and the file name.
+    def _extract_s3_key(self, full_file_path: str) -> str:
+        return "/".join(full_file_path.split("//")[1].split("/")[1:])
+
+    def _get_local_file_path(self, s3_key: str) -> str:
+        return f"{UPLOAD_DIR}/{s3_key.split('/')[-1]}"
+

 class GCSStorageProvider(StorageProvider):
    def __init__(self):
@@ -207,6 +227,74 @@ class GCSStorageProvider(StorageProvider):
        LocalStorageProvider.delete_all_files()


+class AzureStorageProvider(StorageProvider):
+    def __init__(self):
+        self.endpoint = AZURE_STORAGE_ENDPOINT
+        self.container_name = AZURE_STORAGE_CONTAINER_NAME
+        storage_key = AZURE_STORAGE_KEY
+
+        if storage_key:
+            # Configure using the Azure Storage Account Endpoint and Key
+            self.blob_service_client = BlobServiceClient(
+                account_url=self.endpoint, credential=storage_key
+            )
+        else:
+            # Configure using the Azure Storage Account Endpoint and DefaultAzureCredential
+            # If the key is not configured, then the DefaultAzureCredential will be used to support Managed Identity authentication
+            self.blob_service_client = BlobServiceClient(
+                account_url=self.endpoint, credential=DefaultAzureCredential()
+            )
+        self.container_client = self.blob_service_client.get_container_client(
+            self.container_name
+        )
+
+    def upload_file(self, file: BinaryIO, filename: str) -> Tuple[bytes, str]:
+        """Handles uploading of the file to Azure Blob Storage."""
+        contents, file_path = LocalStorageProvider.upload_file(file, filename)
+        try:
+            blob_client = self.container_client.get_blob_client(filename)
+            blob_client.upload_blob(contents, overwrite=True)
+            return contents, f"{self.endpoint}/{self.container_name}/{filename}"
+        except Exception as e:
+            raise RuntimeError(f"Error uploading file to Azure Blob Storage: {e}")
+
+    def get_file(self, file_path: str) -> str:
+        """Handles downloading of the file from Azure Blob Storage."""
+        try:
+            filename = file_path.split("/")[-1]
+            local_file_path = f"{UPLOAD_DIR}/{filename}"
+            blob_client = self.container_client.get_blob_client(filename)
+            with open(local_file_path, "wb") as download_file:
+                download_file.write(blob_client.download_blob().readall())
+            return local_file_path
+        except ResourceNotFoundError as e:
+            raise RuntimeError(f"Error downloading file from Azure Blob Storage: {e}")
+
+    def delete_file(self, file_path: str) -> None:
+        """Handles deletion of the file from Azure Blob Storage."""
+        try:
+            filename = file_path.split("/")[-1]
+            blob_client = self.container_client.get_blob_client(filename)
+            blob_client.delete_blob()
+        except ResourceNotFoundError as e:
+            raise RuntimeError(f"Error deleting file from Azure Blob Storage: {e}")
+
+        # Always delete from local storage
+        LocalStorageProvider.delete_file(file_path)
+
+    def delete_all_files(self) -> None:
+        """Handles deletion of all files from Azure Blob Storage."""
+        try:
+            blobs = self.container_client.list_blobs()
+            for blob in blobs:
+                self.container_client.delete_blob(blob.name)
+        except Exception as e:
+            raise RuntimeError(f"Error deleting all files from Azure Blob Storage: {e}")
+
+        # Always delete from local storage
+        LocalStorageProvider.delete_all_files()
+
+
 def get_storage_provider(storage_provider: str):
    if storage_provider == "local":
        Storage = LocalStorageProvider()
@@ -214,6 +302,8 @@ def get_storage_provider(storage_provider: str):
        Storage = S3StorageProvider()
    elif storage_provider == "gcs":
        Storage = GCSStorageProvider()
+    elif storage_provider == "azure":
+        Storage = AzureStorageProvider()
    else:
        raise RuntimeError(f"Unsupported storage provider: {storage_provider}")
    return Storage
@@ -7,6 +7,8 @@ from moto import mock_aws
 from open_webui.storage import provider
 from gcp_storage_emulator.server import create_server
 from google.cloud import storage
+from azure.storage.blob import BlobServiceClient, ContainerClient, BlobClient
+from unittest.mock import MagicMock


 def mock_upload_dir(monkeypatch, tmp_path):
@@ -22,6 +24,7 @@ def test_imports():
    provider.LocalStorageProvider
    provider.S3StorageProvider
    provider.GCSStorageProvider
+    provider.AzureStorageProvider
    provider.Storage


@@ -32,6 +35,8 @@ def test_get_storage_provider():
    assert isinstance(Storage, provider.S3StorageProvider)
    Storage = provider.get_storage_provider("gcs")
    assert isinstance(Storage, provider.GCSStorageProvider)
+    Storage = provider.get_storage_provider("azure")
+    assert isinstance(Storage, provider.AzureStorageProvider)
    with pytest.raises(RuntimeError):
        provider.get_storage_provider("invalid")

@@ -48,6 +53,7 @@ def test_class_instantiation():
    provider.LocalStorageProvider()
    provider.S3StorageProvider()
    provider.GCSStorageProvider()
+    provider.AzureStorageProvider()


 class TestLocalStorageProvider:
@@ -272,3 +278,147 @@ class TestGCSStorageProvider:
        assert not (upload_dir / self.filename_extra).exists()
        assert self.Storage.bucket.get_blob(self.filename) == None
        assert self.Storage.bucket.get_blob(self.filename_extra) == None
+
+
+class TestAzureStorageProvider:
+    def __init__(self):
+        super().__init__()
+
+    @pytest.fixture(scope="class")
+    def setup_storage(self, monkeypatch):
+        # Create mock Blob Service Client and related clients
+        mock_blob_service_client = MagicMock()
+        mock_container_client = MagicMock()
+        mock_blob_client = MagicMock()
+
+        # Set up return values for the mock
+        mock_blob_service_client.get_container_client.return_value = (
+            mock_container_client
+        )
+        mock_container_client.get_blob_client.return_value = mock_blob_client
+
+        # Monkeypatch the Azure classes to return our mocks
+        monkeypatch.setattr(
+            azure.storage.blob,
+            "BlobServiceClient",
+            lambda *args, **kwargs: mock_blob_service_client,
+        )
+        monkeypatch.setattr(
+            azure.storage.blob,
+            "ContainerClient",
+            lambda *args, **kwargs: mock_container_client,
+        )
+        monkeypatch.setattr(
+            azure.storage.blob, "BlobClient", lambda *args, **kwargs: mock_blob_client
+        )
+
+        self.Storage = provider.AzureStorageProvider()
+        self.Storage.endpoint = "https://myaccount.blob.core.windows.net"
+        self.Storage.container_name = "my-container"
+        self.file_content = b"test content"
+        self.filename = "test.txt"
+        self.filename_extra = "test_extra.txt"
+        self.file_bytesio_empty = io.BytesIO()
+
+        # Apply mocks to the Storage instance
+        self.Storage.blob_service_client = mock_blob_service_client
+        self.Storage.container_client = mock_container_client
+
+    def test_upload_file(self, monkeypatch, tmp_path):
+        upload_dir = mock_upload_dir(monkeypatch, tmp_path)
+
+        # Simulate an error when container does not exist
+        self.Storage.container_client.get_blob_client.side_effect = Exception(
+            "Container does not exist"
+        )
+        with pytest.raises(Exception):
+            self.Storage.upload_file(io.BytesIO(self.file_content), self.filename)
+
+        # Reset side effect and create container
+        self.Storage.container_client.get_blob_client.side_effect = None
+        self.Storage.create_container()
+        contents, azure_file_path = self.Storage.upload_file(
+            io.BytesIO(self.file_content), self.filename
+        )
+
+        # Assertions
+        self.Storage.container_client.get_blob_client.assert_called_with(self.filename)
+        self.Storage.container_client.get_blob_client().upload_blob.assert_called_once_with(
+            self.file_content, overwrite=True
+        )
+        assert contents == self.file_content
+        assert (
+            azure_file_path
+            == f"https://myaccount.blob.core.windows.net/{self.Storage.container_name}/{self.filename}"
+        )
+        assert (upload_dir / self.filename).exists()
+        assert (upload_dir / self.filename).read_bytes() == self.file_content
+
+        with pytest.raises(ValueError):
+            self.Storage.upload_file(self.file_bytesio_empty, self.filename)
+
+    def test_get_file(self, monkeypatch, tmp_path):
+        upload_dir = mock_upload_dir(monkeypatch, tmp_path)
+        self.Storage.create_container()
+
+        # Mock upload behavior
+        self.Storage.upload_file(io.BytesIO(self.file_content), self.filename)
+        # Mock blob download behavior
+        self.Storage.container_client.get_blob_client().download_blob().readall.return_value = (
+            self.file_content
+        )
+
+        file_url = f"https://myaccount.blob.core.windows.net/{self.Storage.container_name}/{self.filename}"
+        file_path = self.Storage.get_file(file_url)
+
+        assert file_path == str(upload_dir / self.filename)
+        assert (upload_dir / self.filename).exists()
+        assert (upload_dir / self.filename).read_bytes() == self.file_content
+
+    def test_delete_file(self, monkeypatch, tmp_path):
+        upload_dir = mock_upload_dir(monkeypatch, tmp_path)
+        self.Storage.create_container()
+
+        # Mock file upload
+        self.Storage.upload_file(io.BytesIO(self.file_content), self.filename)
+        # Mock deletion
+        self.Storage.container_client.get_blob_client().delete_blob.return_value = None
+
+        file_url = f"https://myaccount.blob.core.windows.net/{self.Storage.container_name}/{self.filename}"
+        self.Storage.delete_file(file_url)
+
+        self.Storage.container_client.get_blob_client().delete_blob.assert_called_once()
+        assert not (upload_dir / self.filename).exists()
+
+    def test_delete_all_files(self, monkeypatch, tmp_path):
+        upload_dir = mock_upload_dir(monkeypatch, tmp_path)
+        self.Storage.create_container()
+
+        # Mock file uploads
+        self.Storage.upload_file(io.BytesIO(self.file_content), self.filename)
+        self.Storage.upload_file(io.BytesIO(self.file_content), self.filename_extra)
+
+        # Mock listing and deletion behavior
+        self.Storage.container_client.list_blobs.return_value = [
+            {"name": self.filename},
+            {"name": self.filename_extra},
+        ]
+        self.Storage.container_client.get_blob_client().delete_blob.return_value = None
+
+        self.Storage.delete_all_files()
+
+        self.Storage.container_client.list_blobs.assert_called_once()
+        self.Storage.container_client.get_blob_client().delete_blob.assert_any_call()
+        assert not (upload_dir / self.filename).exists()
+        assert not (upload_dir / self.filename_extra).exists()
+
+    def test_get_file_not_found(self, monkeypatch):
+        self.Storage.create_container()
+
+        file_url = f"https://myaccount.blob.core.windows.net/{self.Storage.container_name}/{self.filename}"
+        # Mock behavior to raise an error for missing blobs
+        self.Storage.container_client.get_blob_client().download_blob.side_effect = (
+            Exception("Blob not found")
+        )
+        with pytest.raises(Exception, match="Blob not found"):
+            self.Storage.get_file(file_url)
@@ -1,6 +1,12 @@
 import logging
 import uuid
 import jwt
+import base64
+import hmac
+import hashlib
+import requests
+import os
+

 from datetime import UTC, datetime, timedelta
 from typing import Optional, Union, List, Dict
@@ -8,7 +14,7 @@ from typing import Optional, Union, List, Dict
 from open_webui.models.users import Users

 from open_webui.constants import ERROR_MESSAGES
-from open_webui.env import WEBUI_SECRET_KEY
+from open_webui.env import WEBUI_SECRET_KEY, TRUSTED_SIGNATURE_KEY, STATIC_DIR

 from fastapi import Depends, HTTPException, Request, Response, status
 from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
@@ -24,6 +30,66 @@ ALGORITHM = "HS256"
 # Auth Utils
 ##############

+
+def verify_signature(payload: str, signature: str) -> bool:
+    """
+    Verifies the HMAC signature of the received payload.
+    """
+    try:
+        expected_signature = base64.b64encode(
+            hmac.new(TRUSTED_SIGNATURE_KEY, payload.encode(), hashlib.sha256).digest()
+        ).decode()
+
+        # Compare securely to prevent timing attacks
+        return hmac.compare_digest(expected_signature, signature)
+
+    except Exception:
+        return False
+
+
+def override_static(path: str, content: str):
+    # Ensure path is safe
+    if "/" in path or ".." in path:
+        print(f"Invalid path: {path}")
+        return
+
+    file_path = os.path.join(STATIC_DIR, path)
+    os.makedirs(os.path.dirname(file_path), exist_ok=True)
+
+    with open(file_path, "wb") as f:
+        f.write(base64.b64decode(content))  # Convert Base64 back to raw binary
+
+
+def get_license_data(app, key):
+    if key:
+        try:
+            res = requests.post(
+                "https://api.openwebui.com/api/v1/license",
+                json={"key": key, "version": "1"},
+                timeout=5,
+            )
+
+            if getattr(res, "ok", False):
+                payload = getattr(res, "json", lambda: {})()
+                for k, v in payload.items():
+                    if k == "resources":
+                        for p, c in v.items():
+                            globals().get("override_static", lambda a, b: None)(p, c)
+                    elif k == "user_count":
+                        setattr(app.state, "USER_COUNT", v)
+                    elif k == "webui_name":
+                        setattr(app.state, "WEBUI_NAME", v)
+
+                return True
+            else:
+                print(
+                    f"License: retrieval issue: {getattr(res, 'text', 'unknown error')}"
+                )
+        except Exception as ex:
+            print(f"License: Uncaught Exception: {ex}")
+    return False
+
+
 bearer_security = HTTPBearer(auto_error=False)
 pwd_context = CryptContext(schemes=["bcrypt"], deprecated="auto")

@@ -7,14 +7,17 @@ from typing import Any, Optional
 import random
 import json
 import inspect
+import uuid
+import asyncio

-from fastapi import Request
-from starlette.responses import Response, StreamingResponse
+from fastapi import Request, status
+from starlette.responses import Response, StreamingResponse, JSONResponse


 from open_webui.models.users import UserModel

 from open_webui.socket.main import (
+    sio,
    get_event_call,
    get_event_emitter,
 )
@@ -44,6 +47,10 @@ from open_webui.utils.response import (
    convert_response_ollama_to_openai,
    convert_streaming_response_ollama_to_openai,
 )
+from open_webui.utils.filter import (
+    get_sorted_filter_ids,
+    process_filter_functions,
+)

 from open_webui.env import SRC_LOG_LEVELS, GLOBAL_LOG_LEVEL, BYPASS_MODEL_ACCESS_CONTROL

@@ -53,108 +60,224 @@ log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["MAIN"])


+async def generate_direct_chat_completion(
+    request: Request,
+    form_data: dict,
+    user: Any,
+    models: dict,
+):
+    print("generate_direct_chat_completion")
+
+    metadata = form_data.pop("metadata", {})
+
+    user_id = metadata.get("user_id")
+    session_id = metadata.get("session_id")
+    request_id = str(uuid.uuid4())  # Generate a unique request ID
+
+    event_caller = get_event_call(metadata)
+
+    channel = f"{user_id}:{session_id}:{request_id}"
+
+    if form_data.get("stream"):
+        q = asyncio.Queue()
+
+        async def message_listener(sid, data):
+            """
+            Handle received socket messages and push them into the queue.
+            """
+            await q.put(data)
+
+        # Register the listener
+        sio.on(channel, message_listener)
+
+        # Start processing chat completion in background
+        res = await event_caller(
+            {
+                "type": "request:chat:completion",
+                "data": {
+                    "form_data": form_data,
+                    "model": models[form_data["model"]],
+                    "channel": channel,
+                    "session_id": session_id,
+                },
+            }
+        )
+
+        print("res", res)
+
+        if res.get("status", False):
+            # Define a generator to stream responses
+            async def event_generator():
+                nonlocal q
+                try:
+                    while True:
+                        data = await q.get()  # Wait for new messages
+                        if isinstance(data, dict):
+                            if "done" in data and data["done"]:
+                                break  # Stop streaming when 'done' is received
+
+                            yield f"data: {json.dumps(data)}\n\n"
+                        elif isinstance(data, str):
+                            yield data
+                except Exception as e:
+                    log.debug(f"Error in event generator: {e}")
+                    pass
+
+            # Define a background task to run the event generator
+            async def background():
+                try:
+                    del sio.handlers["/"][channel]
+                except Exception as e:
+                    pass
+
+            # Return the streaming response
+            return StreamingResponse(
+                event_generator(), media_type="text/event-stream", background=background
+            )
+        else:
+            raise Exception(str(res))
+    else:
+        res = await event_caller(
+            {
+                "type": "request:chat:completion",
+                "data": {
+                    "form_data": form_data,
+                    "model": models[form_data["model"]],
+                    "channel": channel,
+                    "session_id": session_id,
+                },
+            }
+        )
+
+        if "error" in res:
+            raise Exception(res["error"])
+
+        return res
+
+
 async def generate_chat_completion(
    request: Request,
    form_data: dict,
    user: Any,
    bypass_filter: bool = False,
 ):
+    log.debug(f"generate_chat_completion: {form_data}")
    if BYPASS_MODEL_ACCESS_CONTROL:
        bypass_filter = True

-    models = request.app.state.MODELS
+    if hasattr(request.state, "metadata"):
+        if "metadata" not in form_data:
+            form_data["metadata"] = request.state.metadata
+        else:
+            form_data["metadata"] = {
+                **form_data["metadata"],
+                **request.state.metadata,
+            }
+
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+        log.debug(f"direct connection to model: {models}")
+    else:
+        models = request.app.state.MODELS

    model_id = form_data["model"]
    if model_id not in models:
        raise Exception("Model not found")

-    # Process the form_data through the pipeline
-    try:
-        form_data = process_pipeline_inlet_filter(request, form_data, user, models)
-    except Exception as e:
-        raise e
-
    model = models[model_id]

-    # Check if user has access to the model
-    if not bypass_filter and user.role == "user":
-        try:
-            check_model_access(user, model)
-        except Exception as e:
-            raise e
-
-    if model["owned_by"] == "arena":
-        model_ids = model.get("info", {}).get("meta", {}).get("model_ids")
-        filter_mode = model.get("info", {}).get("meta", {}).get("filter_mode")
-        if model_ids and filter_mode == "exclude":
-            model_ids = [
-                model["id"]
-                for model in list(request.app.state.MODELS.values())
-                if model.get("owned_by") != "arena" and model["id"] not in model_ids
-            ]
-
-        selected_model_id = None
-        if isinstance(model_ids, list) and model_ids:
-            selected_model_id = random.choice(model_ids)
-        else:
-            model_ids = [
-                model["id"]
-                for model in list(request.app.state.MODELS.values())
-                if model.get("owned_by") != "arena"
-            ]
-            selected_model_id = random.choice(model_ids)
-
-        form_data["model"] = selected_model_id
-
-        if form_data.get("stream") == True:
-
-            async def stream_wrapper(stream):
-                yield f"data: {json.dumps({'selected_model_id': selected_model_id})}\n\n"
-                async for chunk in stream:
-                    yield chunk
-
-            response = await generate_chat_completion(
-                request, form_data, user, bypass_filter=True
-            )
-            return StreamingResponse(
-                stream_wrapper(response.body_iterator),
-                media_type="text/event-stream",
-                background=response.background,
-            )
-        else:
-            return {
-                **(
-                    await generate_chat_completion(
-                        request, form_data, user, bypass_filter=True
-                    )
-                ),
-                "selected_model_id": selected_model_id,
-            }
-
-    if model.get("pipe"):
-        # Below does not require bypass_filter because this is the only route the uses this function and it is already bypassing the filter
-        return await generate_function_chat_completion(
+    if getattr(request.state, "direct", False):
+        return await generate_direct_chat_completion(
            request, form_data, user=user, models=models
        )
-    if model["owned_by"] == "ollama":
-        # Using /ollama/api/chat endpoint
-        form_data = convert_payload_openai_to_ollama(form_data)
-        response = await generate_ollama_chat_completion(
-            request=request, form_data=form_data, user=user, bypass_filter=bypass_filter
-        )
-        if form_data.get("stream"):
-            response.headers["content-type"] = "text/event-stream"
-            return StreamingResponse(
-                convert_streaming_response_ollama_to_openai(response),
-                headers=dict(response.headers),
-                background=response.background,
-            )
-        else:
-            return convert_response_ollama_to_openai(response)
    else:
-        return await generate_openai_chat_completion(
-            request=request, form_data=form_data, user=user, bypass_filter=bypass_filter
-        )
+        # Check if user has access to the model
+        if not bypass_filter and user.role == "user":
+            try:
+                check_model_access(user, model)
+            except Exception as e:
+                raise e
+
+        if model.get("owned_by") == "arena":
+            model_ids = model.get("info", {}).get("meta", {}).get("model_ids")
+            filter_mode = model.get("info", {}).get("meta", {}).get("filter_mode")
+            if model_ids and filter_mode == "exclude":
+                model_ids = [
+                    model["id"]
+                    for model in list(request.app.state.MODELS.values())
+                    if model.get("owned_by") != "arena" and model["id"] not in model_ids
+                ]
+
+            selected_model_id = None
+            if isinstance(model_ids, list) and model_ids:
+                selected_model_id = random.choice(model_ids)
+            else:
+                model_ids = [
+                    model["id"]
+                    for model in list(request.app.state.MODELS.values())
+                    if model.get("owned_by") != "arena"
+                ]
+                selected_model_id = random.choice(model_ids)
+
+            form_data["model"] = selected_model_id
+
+            if form_data.get("stream") == True:
+
+                async def stream_wrapper(stream):
+                    yield f"data: {json.dumps({'selected_model_id': selected_model_id})}\n\n"
+                    async for chunk in stream:
+                        yield chunk
+
+                response = await generate_chat_completion(
+                    request, form_data, user, bypass_filter=True
+                )
+                return StreamingResponse(
+                    stream_wrapper(response.body_iterator),
+                    media_type="text/event-stream",
+                    background=response.background,
+                )
+            else:
+                return {
+                    **(
+                        await generate_chat_completion(
+                            request, form_data, user, bypass_filter=True
+                        )
+                    ),
+                    "selected_model_id": selected_model_id,
+                }
+
+        if model.get("pipe"):
+            # Below does not require bypass_filter because this is the only route the uses this function and it is already bypassing the filter
+            return await generate_function_chat_completion(
+                request, form_data, user=user, models=models
+            )
+        if model.get("owned_by") == "ollama":
+            # Using /ollama/api/chat endpoint
+            form_data = convert_payload_openai_to_ollama(form_data)
+            response = await generate_ollama_chat_completion(
+                request=request,
+                form_data=form_data,
+                user=user,
+                bypass_filter=bypass_filter,
+            )
+            if form_data.get("stream"):
+                response.headers["content-type"] = "text/event-stream"
+                return StreamingResponse(
+                    convert_streaming_response_ollama_to_openai(response),
+                    headers=dict(response.headers),
+                    background=response.background,
+                )
+            else:
+                return convert_response_ollama_to_openai(response)
+        else:
+            return await generate_openai_chat_completion(
+                request=request,
+                form_data=form_data,
+                user=user,
+                bypass_filter=bypass_filter,
+            )


 chat_completion = generate_chat_completion
@@ -163,7 +286,13 @@ chat_completion = generate_chat_completion
 async def chat_completed(request: Request, form_data: dict, user: Any):
    if not request.app.state.MODELS:
        await get_all_models(request)
-    models = request.app.state.MODELS
+
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    data = form_data
    model_id = data["model"]
@@ -173,120 +302,42 @@ async def chat_completed(request: Request, form_data: dict, user: Any):
    model = models[model_id]

    try:
-        data = process_pipeline_outlet_filter(request, data, user, models)
+        data = await process_pipeline_outlet_filter(request, data, user, models)
    except Exception as e:
        return Exception(f"Error: {e}")

-    __event_emitter__ = get_event_emitter(
-        {
-            "chat_id": data["chat_id"],
-            "message_id": data["id"],
-            "session_id": data["session_id"],
-            "user_id": user.id,
-        }
-    )
+    metadata = {
+        "chat_id": data["chat_id"],
+        "message_id": data["id"],
+        "session_id": data["session_id"],
+        "user_id": user.id,
+    }

-    __event_call__ = get_event_call(
-        {
-            "chat_id": data["chat_id"],
-            "message_id": data["id"],
-            "session_id": data["session_id"],
-            "user_id": user.id,
-        }
-    )
+    extra_params = {
+        "__event_emitter__": get_event_emitter(metadata),
+        "__event_call__": get_event_call(metadata),
+        "__user__": {
+            "id": user.id,
+            "email": user.email,
+            "name": user.name,
+            "role": user.role,
+        },
+        "__metadata__": metadata,
+        "__request__": request,
+        "__model__": model,
+    }

-    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)
-        return 0
-
-    filter_ids = [function.id for function in Functions.get_global_filter_functions()]
-    if "info" in model and "meta" in model["info"]:
-        filter_ids.extend(model["info"]["meta"].get("filterIds", []))
-        filter_ids = list(set(filter_ids))
-
-    enabled_filter_ids = [
-        function.id
-        for function in Functions.get_functions_by_type("filter", active_only=True)
-    ]
-    filter_ids = [
-        filter_id for filter_id in filter_ids if filter_id in enabled_filter_ids
-    ]
-
-    # Sort filter_ids by priority, using the get_priority function
-    filter_ids.sort(key=get_priority)
-
-    for filter_id in filter_ids:
-        filter = Functions.get_function_by_id(filter_id)
-        if not filter:
-            continue
-
-        if filter_id in request.app.state.FUNCTIONS:
-            function_module = request.app.state.FUNCTIONS[filter_id]
-        else:
-            function_module, _, _ = load_function_module_by_id(filter_id)
-            request.app.state.FUNCTIONS[filter_id] = function_module
-
-        if hasattr(function_module, "valves") and hasattr(function_module, "Valves"):
-            valves = Functions.get_function_valves_by_id(filter_id)
-            function_module.valves = function_module.Valves(
-                **(valves if valves else {})
-            )
-
-        if not hasattr(function_module, "outlet"):
-            continue
-        try:
-            outlet = function_module.outlet
-
-            # Get the signature of the function
-            sig = inspect.signature(outlet)
-            params = {"body": data}
-
-            # Extra parameters to be passed to the function
-            extra_params = {
-                "__model__": model,
-                "__id__": filter_id,
-                "__event_emitter__": __event_emitter__,
-                "__event_call__": __event_call__,
-                "__request__": request,
-            }
-
-            # Add extra params in contained in function signature
-            for key, value in extra_params.items():
-                if key in sig.parameters:
-                    params[key] = value
-
-            if "__user__" in sig.parameters:
-                __user__ = {
-                    "id": user.id,
-                    "email": user.email,
-                    "name": user.name,
-                    "role": user.role,
-                }
-
-                try:
-                    if hasattr(function_module, "UserValves"):
-                        __user__["valves"] = function_module.UserValves(
-                            **Functions.get_user_valves_by_id_and_user_id(
-                                filter_id, user.id
-                            )
-                        )
-                except Exception as e:
-                    print(e)
-
-                params = {**params, "__user__": __user__}
-
-            if inspect.iscoroutinefunction(outlet):
-                data = await outlet(**params)
-            else:
-                data = outlet(**params)
-
-        except Exception as e:
-            return Exception(f"Error: {e}")
-
-    return data
+    try:
+        result, _ = await process_filter_functions(
+            request=request,
+            filter_ids=get_sorted_filter_ids(model),
+            filter_type="outlet",
+            form_data=data,
+            extra_params=extra_params,
+        )
+        return result
+    except Exception as e:
+        return Exception(f"Error: {e}")


 async def chat_action(request: Request, action_id: str, form_data: dict, user: Any):
@@ -301,7 +352,13 @@ async def chat_action(request: Request, action_id: str, form_data: dict, user: A

    if not request.app.state.MODELS:
        await get_all_models(request)
-    models = request.app.state.MODELS
+
+    if getattr(request.state, "direct", False) and hasattr(request.state, "model"):
+        models = {
+            request.state.model["id"]: request.state.model,
+        }
+    else:
+        models = request.app.state.MODELS

    data = form_data
    model_id = data["model"]
@@ -0,0 +1,148 @@
+import asyncio
+import json
+import uuid
+import websockets
+import requests
+from urllib.parse import urljoin
+
+
+async def execute_code_jupyter(
+    jupyter_url, code, token=None, password=None, timeout=10
+):
+    """
+    Executes Python code in a Jupyter kernel.
+    Supports authentication with a token or password.
+    :param jupyter_url: Jupyter server URL (e.g., "http://localhost:8888")
+    :param code: Code to execute
+    :param token: Jupyter authentication token (optional)
+    :param password: Jupyter password (optional)
+    :param timeout: WebSocket timeout in seconds (default: 10s)
+    :return: Dictionary with stdout, stderr, and result
+             - Images are prefixed with "base64:image/png," and separated by newlines if multiple.
+    """
+    session = requests.Session()  # Maintain cookies
+    headers = {}  # Headers for requests
+
+    # Authenticate using password
+    if password and not token:
+        try:
+            login_url = urljoin(jupyter_url, "/login")
+            response = session.get(login_url)
+            response.raise_for_status()
+            xsrf_token = session.cookies.get("_xsrf")
+            if not xsrf_token:
+                raise ValueError("Failed to fetch _xsrf token")
+
+            login_data = {"_xsrf": xsrf_token, "password": password}
+            login_response = session.post(
+                login_url, data=login_data, cookies=session.cookies
+            )
+            login_response.raise_for_status()
+            headers["X-XSRFToken"] = xsrf_token
+        except Exception as e:
+            return {
+                "stdout": "",
+                "stderr": f"Authentication Error: {str(e)}",
+                "result": "",
+            }
+
+    # Construct API URLs with authentication token if provided
+    params = f"?token={token}" if token else ""
+    kernel_url = urljoin(jupyter_url, f"/api/kernels{params}")
+
+    try:
+        response = session.post(kernel_url, headers=headers, cookies=session.cookies)
+        response.raise_for_status()
+        kernel_id = response.json()["id"]
+
+        websocket_url = urljoin(
+            jupyter_url.replace("http", "ws"),
+            f"/api/kernels/{kernel_id}/channels{params}",
+        )
+
+        ws_headers = {}
+        if password and not token:
+            ws_headers["X-XSRFToken"] = session.cookies.get("_xsrf")
+            cookies = {name: value for name, value in session.cookies.items()}
+            ws_headers["Cookie"] = "; ".join(
+                [f"{name}={value}" for name, value in cookies.items()]
+            )
+
+        async with websockets.connect(
+            websocket_url, additional_headers=ws_headers
+        ) as ws:
+            msg_id = str(uuid.uuid4())
+            execute_request = {
+                "header": {
+                    "msg_id": msg_id,
+                    "msg_type": "execute_request",
+                    "username": "user",
+                    "session": str(uuid.uuid4()),
+                    "date": "",
+                    "version": "5.3",
+                },
+                "parent_header": {},
+                "metadata": {},
+                "content": {
+                    "code": code,
+                    "silent": False,
+                    "store_history": True,
+                    "user_expressions": {},
+                    "allow_stdin": False,
+                    "stop_on_error": True,
+                },
+                "channel": "shell",
+            }
+            await ws.send(json.dumps(execute_request))
+
+            stdout, stderr, result = "", "", []
+
+            while True:
+                try:
+                    message = await asyncio.wait_for(ws.recv(), timeout)
+                    message_data = json.loads(message)
+                    if message_data.get("parent_header", {}).get("msg_id") == msg_id:
+                        msg_type = message_data.get("msg_type")
+
+                        if msg_type == "stream":
+                            if message_data["content"]["name"] == "stdout":
+                                stdout += message_data["content"]["text"]
+                            elif message_data["content"]["name"] == "stderr":
+                                stderr += message_data["content"]["text"]
+
+                        elif msg_type in ("execute_result", "display_data"):
+                            data = message_data["content"]["data"]
+                            if "image/png" in data:
+                                result.append(
+                                    f"data:image/png;base64,{data['image/png']}"
+                                )
+                            elif "text/plain" in data:
+                                result.append(data["text/plain"])
+
+                        elif msg_type == "error":
+                            stderr += "\n".join(message_data["content"]["traceback"])
+
+                        elif (
+                            msg_type == "status"
+                            and message_data["content"]["execution_state"] == "idle"
+                        ):
+                            break
+
+                except asyncio.TimeoutError:
+                    stderr += "\nExecution timed out."
+                    break
+
+    except Exception as e:
+        return {"stdout": "", "stderr": f"Error: {str(e)}", "result": ""}
+
+    finally:
+        if kernel_id:
+            requests.delete(
+                f"{kernel_url}/{kernel_id}", headers=headers, cookies=session.cookies
+            )
+
+    return {
+        "stdout": stdout.strip(),
+        "stderr": stderr.strip(),
+        "result": "\n".join(result).strip() if result else "",
+    }
@@ -0,0 +1,99 @@
+import inspect
+from open_webui.utils.plugin import load_function_module_by_id
+from open_webui.models.functions import Functions
+
+
+def get_sorted_filter_ids(model):
+    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)
+        return 0
+
+    filter_ids = [function.id for function in Functions.get_global_filter_functions()]
+    if "info" in model and "meta" in model["info"]:
+        filter_ids.extend(model["info"]["meta"].get("filterIds", []))
+        filter_ids = list(set(filter_ids))
+
+    enabled_filter_ids = [
+        function.id
+        for function in Functions.get_functions_by_type("filter", active_only=True)
+    ]
+
+    filter_ids = [fid for fid in filter_ids if fid in enabled_filter_ids]
+    filter_ids.sort(key=get_priority)
+    return filter_ids
+
+
+async def process_filter_functions(
+    request, filter_ids, filter_type, form_data, extra_params
+):
+    skip_files = None
+
+    for filter_id in filter_ids:
+        filter = Functions.get_function_by_id(filter_id)
+        if not filter:
+            continue
+
+        if filter_id in request.app.state.FUNCTIONS:
+            function_module = request.app.state.FUNCTIONS[filter_id]
+        else:
+            function_module, _, _ = load_function_module_by_id(filter_id)
+            request.app.state.FUNCTIONS[filter_id] = function_module
+
+        # Check if the function has a file_handler variable
+        if filter_type == "inlet" and hasattr(function_module, "file_handler"):
+            skip_files = function_module.file_handler
+
+        # Apply valves to the function
+        if hasattr(function_module, "valves") and hasattr(function_module, "Valves"):
+            valves = Functions.get_function_valves_by_id(filter_id)
+            function_module.valves = function_module.Valves(
+                **(valves if valves else {})
+            )
+
+        # Prepare handler function
+        handler = getattr(function_module, filter_type, None)
+        if not handler:
+            continue
+
+        try:
+            # Prepare parameters
+            sig = inspect.signature(handler)
+            params = {"body": form_data} | {
+                k: v
+                for k, v in {
+                    **extra_params,
+                    "__id__": filter_id,
+                }.items()
+                if k in sig.parameters
+            }
+
+            # Handle user parameters
+            if "__user__" in sig.parameters:
+                if hasattr(function_module, "UserValves"):
+                    try:
+                        params["__user__"]["valves"] = function_module.UserValves(
+                            **Functions.get_user_valves_by_id_and_user_id(
+                                filter_id, params["__user__"]["id"]
+                            )
+                        )
+                    except Exception as e:
+                        print(e)
+
+            # Execute handler
+            if inspect.iscoroutinefunction(handler):
+                form_data = await handler(**params)
+            else:
+                form_data = handler(**params)
+
+        except Exception as e:
+            print(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["metadata"]["files"]
+
+    return form_data, {}
@@ -161,7 +161,7 @@ async def comfyui_generate_image(
                seed = (
                    payload.seed
                    if payload.seed
-                    else random.randint(0, 18446744073709551614)
+                    else random.randint(0, 1125899906842624)
                )
                for node_id in node.node_ids:
                    workflow[node_id]["inputs"][node.key] = seed
@@ -7,6 +7,18 @@ from pathlib import Path
 from typing import Callable, Optional


+import collections.abc
+
+
+def deep_update(d, u):
+    for k, v in u.items():
+        if isinstance(v, collections.abc.Mapping):
+            d[k] = deep_update(d.get(k, {}), v)
+        else:
+            d[k] = v
+    return d
+
+
 def get_message_list(messages, message_id):
    """
    Reconstructs a list of messages in order up to the specified message_id.
@@ -20,7 +32,7 @@ def get_message_list(messages, message_id):
    current_message = messages.get(message_id)

    if not current_message:
-        return f"Message ID {message_id} not found in the history."
+        return None

    # Reconstruct the chain by following the parentId links
    message_list = []
@@ -131,6 +143,44 @@ def add_or_update_system_message(content: str, messages: list[dict]):
    return messages


+def add_or_update_user_message(content: str, messages: list[dict]):
+    """
+    Adds a new user message at the end of the messages list
+    or updates the existing user message at the end.
+
+    :param msg: The message to be added or appended.
+    :param messages: The list of message dictionaries.
+    :return: The updated list of message dictionaries.
+    """
+
+    if messages and messages[-1].get("role") == "user":
+        messages[-1]["content"] = f"{messages[-1]['content']}\n{content}"
+    else:
+        # Insert at the end
+        messages.append({"role": "user", "content": content})
+
+    return messages
+
+
+def append_or_update_assistant_message(content: str, messages: list[dict]):
+    """
+    Adds a new assistant message at the end of the messages list
+    or updates the existing assistant message at the end.
+
+    :param msg: The message to be added or appended.
+    :param messages: The list of message dictionaries.
+    :return: The updated list of message dictionaries.
+    """
+
+    if messages and messages[-1].get("role") == "assistant":
+        messages[-1]["content"] = f"{messages[-1]['content']}\n{content}"
+    else:
+        # Insert at the end
+        messages.append({"role": "assistant", "content": content})
+
+    return messages
+
+
 def openai_chat_message_template(model: str):
    return {
        "id": f"{model}-{str(uuid.uuid4())}",
@@ -141,13 +191,24 @@ def openai_chat_message_template(model: str):


 def openai_chat_chunk_message_template(
-    model: str, message: Optional[str] = None, usage: Optional[dict] = None
+    model: str,
+    content: Optional[str] = None,
+    tool_calls: Optional[list[dict]] = None,
+    usage: Optional[dict] = None,
 ) -> dict:
    template = openai_chat_message_template(model)
    template["object"] = "chat.completion.chunk"
-    if message:
-        template["choices"][0]["delta"] = {"content": message}
-    else:
+
+    template["choices"][0]["index"] = 0
+    template["choices"][0]["delta"] = {}
+
+    if content:
+        template["choices"][0]["delta"]["content"] = content
+
+    if tool_calls:
+        template["choices"][0]["delta"]["tool_calls"] = tool_calls
+
+    if not content and not tool_calls:
        template["choices"][0]["finish_reason"] = "stop"

    if usage:
@@ -156,12 +217,20 @@ def openai_chat_chunk_message_template(


 def openai_chat_completion_message_template(
-    model: str, message: Optional[str] = None, usage: Optional[dict] = None
+    model: str,
+    message: Optional[str] = None,
+    tool_calls: Optional[list[dict]] = None,
+    usage: Optional[dict] = None,
 ) -> dict:
    template = openai_chat_message_template(model)
    template["object"] = "chat.completion"
    if message is not None:
-        template["choices"][0]["message"] = {"content": message, "role": "assistant"}
+        template["choices"][0]["message"] = {
+            "content": message,
+            "role": "assistant",
+            **({"tool_calls": tool_calls} if tool_calls else {}),
+        }
+
    template["choices"][0]["finish_reason"] = "stop"

    if usage:
@@ -183,11 +252,12 @@ def get_gravatar_url(email):
    return f"https://www.gravatar.com/avatar/{hash_hex}?d=mp"


-def calculate_sha256(file):
+def calculate_sha256(file_path, chunk_size):
+    # Compute SHA-256 hash of a file efficiently in chunks
    sha256 = hashlib.sha256()
-    # Read the file in chunks to efficiently handle large files
-    for chunk in iter(lambda: file.read(8192), b""):
-        sha256.update(chunk)
+    with open(file_path, "rb") as f:
+        while chunk := f.read(chunk_size):
+            sha256.update(chunk)
    return sha256.hexdigest()


@@ -142,7 +142,7 @@ async def get_all_models(request):
                    custom_model.base_model_id == model["id"]
                    or custom_model.base_model_id == model["id"].split(":")[0]
                ):
-                    owned_by = model["owned_by"]
+                    owned_by = model.get("owned_by", "unknown owner")
                    if "pipe" in model:
                        pipe = model["pipe"]
                    break
@@ -1,6 +1,7 @@
 import base64
 import logging
 import mimetypes
+import sys
 import uuid

 import aiohttp
@@ -35,12 +36,20 @@ from open_webui.config import (
    AppConfig,
 )
 from open_webui.constants import ERROR_MESSAGES, WEBHOOK_MESSAGES
-from open_webui.env import WEBUI_SESSION_COOKIE_SAME_SITE, WEBUI_SESSION_COOKIE_SECURE
+from open_webui.env import (
+    WEBUI_NAME,
+    WEBUI_AUTH_COOKIE_SAME_SITE,
+    WEBUI_AUTH_COOKIE_SECURE,
+)
 from open_webui.utils.misc import parse_duration
 from open_webui.utils.auth import get_password_hash, create_token
 from open_webui.utils.webhook import post_webhook

+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["OAUTH"])

 auth_manager_config = AppConfig()
 auth_manager_config.DEFAULT_USER_ROLE = DEFAULT_USER_ROLE
@@ -61,8 +70,9 @@ auth_manager_config.JWT_EXPIRES_IN = JWT_EXPIRES_IN


 class OAuthManager:
-    def __init__(self):
+    def __init__(self, app):
        self.oauth = OAuth()
+        self.app = app
        for _, provider_config in OAUTH_PROVIDERS.items():
            provider_config["register"](self.oauth)

@@ -72,17 +82,21 @@ class OAuthManager:
    def get_user_role(self, user, user_data):
        if user and Users.get_num_users() == 1:
            # If the user is the only user, assign the role "admin" - actually repairs role for single user on login
+            log.debug("Assigning the only user the admin role")
            return "admin"
        if not user and Users.get_num_users() == 0:
            # If there are no users, assign the role "admin", as the first user will be an admin
+            log.debug("Assigning the first user the admin role")
            return "admin"

        if auth_manager_config.ENABLE_OAUTH_ROLE_MANAGEMENT:
+            log.debug("Running OAUTH Role management")
            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
-            role = "pending"  # Default/fallback role if no matching roles are found
+            # Default/fallback role if no matching roles are found
+            role = auth_manager_config.DEFAULT_USER_ROLE

            # Next block extracts the roles from the user data, accepting nested claims of any depth
            if oauth_claim and oauth_allowed_roles and oauth_admin_roles:
@@ -92,17 +106,24 @@ class OAuthManager:
                    claim_data = claim_data.get(nested_claim, {})
                oauth_roles = claim_data if isinstance(claim_data, list) else None

+            log.debug(f"Oauth Roles claim: {oauth_claim}")
+            log.debug(f"User roles from oauth: {oauth_roles}")
+            log.debug(f"Accepted user roles: {oauth_allowed_roles}")
+            log.debug(f"Accepted admin roles: {oauth_admin_roles}")
+
            # If any roles are found, check if they match the allowed or admin roles
            if oauth_roles:
                # If role management is enabled, and matching roles are provided, use the roles
                for allowed_role in oauth_allowed_roles:
                    # If the user has any of the allowed roles, assign the role "user"
                    if allowed_role in oauth_roles:
+                        log.debug("Assigned user the user role")
                        role = "user"
                        break
                for admin_role in oauth_admin_roles:
                    # If the user has any of the admin roles, assign the role "admin"
                    if admin_role in oauth_roles:
+                        log.debug("Assigned user the admin role")
                        role = "admin"
                        break
        else:
@@ -116,16 +137,34 @@ class OAuthManager:
        return role

    def update_user_groups(self, user, user_data, default_permissions):
+        log.debug("Running OAUTH Group management")
        oauth_claim = auth_manager_config.OAUTH_GROUPS_CLAIM

-        user_oauth_groups: list[str] = user_data.get(oauth_claim, list())
+        # Nested claim search for groups claim
+        if oauth_claim:
+            claim_data = user_data
+            nested_claims = oauth_claim.split(".")
+            for nested_claim in nested_claims:
+                claim_data = claim_data.get(nested_claim, {})
+            user_oauth_groups = claim_data if isinstance(claim_data, list) else None
+
        user_current_groups: list[GroupModel] = Groups.get_groups_by_member_id(user.id)
        all_available_groups: list[GroupModel] = Groups.get_groups()

+        log.debug(f"Oauth Groups claim: {oauth_claim}")
+        log.debug(f"User oauth groups: {user_oauth_groups}")
+        log.debug(f"User's current groups: {[g.name for g in user_current_groups]}")
+        log.debug(
+            f"All groups available in OpenWebUI: {[g.name for g in all_available_groups]}"
+        )
+
        # 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:
                # Remove group from user
+                log.debug(
+                    f"Removing user from group {group_model.name} as it is no longer in their oauth groups"
+                )

                user_ids = group_model.user_ids
                user_ids = [i for i in user_ids if i != user.id]
@@ -151,6 +190,9 @@ class OAuthManager:
                gm.name == group_model.name for gm in user_current_groups
            ):
                # Add user to group
+                log.debug(
+                    f"Adding user to group {group_model.name} as it was found in their oauth groups"
+                )

                user_ids = group_model.user_ids
                user_ids.append(user.id)
@@ -170,7 +212,7 @@ class OAuthManager:
                    id=group_model.id, form_data=update_form, overwrite=False
                )

-    async def handle_login(self, provider, request):
+    async def handle_login(self, request, provider):
        if provider not in OAUTH_PROVIDERS:
            raise HTTPException(404)
        # If the provider has a custom redirect URL, use that, otherwise automatically generate one
@@ -182,7 +224,7 @@ class OAuthManager:
            raise HTTPException(404)
        return await client.authorize_redirect(request, redirect_uri)

-    async def handle_callback(self, provider, request, response):
+    async def handle_callback(self, request, provider, response):
        if provider not in OAUTH_PROVIDERS:
            raise HTTPException(404)
        client = self.get_client(provider)
@@ -192,7 +234,7 @@ class OAuthManager:
            log.warning(f"OAuth callback error: {e}")
            raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
        user_data: UserInfo = token.get("userinfo")
-        if not user_data:
+        if not user_data or "email" not in user_data:
            user_data: UserInfo = await client.userinfo(token=token)
        if not user_data:
            log.warning(f"OAuth callback failed, user data is missing: {token}")
@@ -204,11 +246,46 @@ class OAuthManager:
            raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
        provider_sub = f"{provider}@{sub}"
        email_claim = auth_manager_config.OAUTH_EMAIL_CLAIM
-        email = user_data.get(email_claim, "").lower()
+        email = user_data.get(email_claim, "")
        # We currently mandate that email addresses are provided
        if not email:
-            log.warning(f"OAuth callback failed, email is missing: {user_data}")
-            raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
+            # If the provider is GitHub,and public email is not provided, we can use the access token to fetch the user's email
+            if provider == "github":
+                try:
+                    access_token = token.get("access_token")
+                    headers = {"Authorization": f"Bearer {access_token}"}
+                    async with aiohttp.ClientSession() as session:
+                        async with session.get(
+                            "https://api.github.com/user/emails", headers=headers
+                        ) as resp:
+                            if resp.ok:
+                                emails = await resp.json()
+                                # use the primary email as the user's email
+                                primary_email = next(
+                                    (e["email"] for e in emails if e.get("primary")),
+                                    None,
+                                )
+                                if primary_email:
+                                    email = primary_email
+                                else:
+                                    log.warning(
+                                        "No primary email found in GitHub response"
+                                    )
+                                    raise HTTPException(
+                                        400, detail=ERROR_MESSAGES.INVALID_CRED
+                                    )
+                            else:
+                                log.warning("Failed to fetch GitHub email")
+                                raise HTTPException(
+                                    400, detail=ERROR_MESSAGES.INVALID_CRED
+                                )
+                except Exception as e:
+                    log.warning(f"Error fetching GitHub email: {e}")
+                    raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
+            else:
+                log.warning(f"OAuth callback failed, email is missing: {user_data}")
+                raise HTTPException(400, detail=ERROR_MESSAGES.INVALID_CRED)
+        email = email.lower()
        if (
            "*" not in auth_manager_config.OAUTH_ALLOWED_DOMAINS
            and email.split("@")[-1] not in auth_manager_config.OAUTH_ALLOWED_DOMAINS
@@ -236,12 +313,21 @@ class OAuthManager:
                Users.update_user_role_by_id(user.id, determined_role)

        if not user:
+            user_count = Users.get_num_users()
+
+            if (
+                request.app.state.USER_COUNT
+                and user_count >= request.app.state.USER_COUNT
+            ):
+                raise HTTPException(
+                    403,
+                    detail=ERROR_MESSAGES.ACCESS_PROHIBITED,
+                )
+
            # If the user does not exist, check if signups are enabled
            if auth_manager_config.ENABLE_OAUTH_SIGNUP:
                # Check if an existing user with the same email already exists
-                existing_user = Users.get_user_by_email(
-                    user_data.get("email", "").lower()
-                )
+                existing_user = Users.get_user_by_email(email)
                if existing_user:
                    raise HTTPException(400, detail=ERROR_MESSAGES.EMAIL_TAKEN)

@@ -260,24 +346,35 @@ class OAuthManager:
                            }
                        async with aiohttp.ClientSession() as session:
                            async with session.get(picture_url, **get_kwargs) as resp:
-                                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}"
+                                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 = ""
+                        picture_url = "/user.png"
                if not picture_url:
                    picture_url = "/user.png"
+
                username_claim = auth_manager_config.OAUTH_USERNAME_CLAIM

+                name = user_data.get(username_claim)
+                if not name:
+                    log.warning("Username claim is missing, using email as name")
+                    name = email
+
                role = self.get_user_role(None, user_data)

                user = Auths.insert_new_auth(
@@ -285,7 +382,7 @@ class OAuthManager:
                    password=get_password_hash(
                        str(uuid.uuid4())
                    ),  # Random password, not used
-                    name=user_data.get(username_claim, "User"),
+                    name=name,
                    profile_image_url=picture_url,
                    role=role,
                    oauth_sub=provider_sub,
@@ -293,6 +390,7 @@ class OAuthManager:

                if auth_manager_config.WEBHOOK_URL:
                    post_webhook(
+                        WEBUI_NAME,
                        auth_manager_config.WEBHOOK_URL,
                        WEBHOOK_MESSAGES.USER_SIGNUP(user.name),
                        {
@@ -323,8 +421,8 @@ class OAuthManager:
            key="token",
            value=jwt_token,
            httponly=True,  # Ensures the cookie is not accessible via JavaScript
-            samesite=WEBUI_SESSION_COOKIE_SAME_SITE,
-            secure=WEBUI_SESSION_COOKIE_SECURE,
+            samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+            secure=WEBUI_AUTH_COOKIE_SECURE,
        )

        if ENABLE_OAUTH_SIGNUP.value:
@@ -333,12 +431,9 @@ class OAuthManager:
                key="oauth_id_token",
                value=oauth_id_token,
                httponly=True,
-                samesite=WEBUI_SESSION_COOKIE_SAME_SITE,
-                secure=WEBUI_SESSION_COOKIE_SECURE,
+                samesite=WEBUI_AUTH_COOKIE_SAME_SITE,
+                secure=WEBUI_AUTH_COOKIE_SECURE,
            )
        # Redirect back to the frontend with the JWT token
        redirect_url = f"{request.base_url}auth#token={jwt_token}"
        return RedirectResponse(url=redirect_url, headers=response.headers)
-
-
-oauth_manager = OAuthManager()
@@ -1,17 +1,27 @@
-from open_webui.utils.task import prompt_template
+from open_webui.utils.task import prompt_template, prompt_variables_template
 from open_webui.utils.misc import (
    add_or_update_system_message,
 )

 from typing import Callable, Optional
+import json


 # inplace function: form_data is modified
-def apply_model_system_prompt_to_body(params: dict, form_data: dict, user) -> dict:
+def apply_model_system_prompt_to_body(
+    params: dict, form_data: dict, metadata: Optional[dict] = None, user=None
+) -> dict:
    system = params.get("system", None)
    if not system:
        return form_data

+    # Metadata (WebUI Usage)
+    if metadata:
+        variables = metadata.get("variables", {})
+        if variables:
+            system = prompt_variables_template(system, variables)
+
+    # Legacy (API Usage)
    if user:
        template_params = {
            "user_name": user.name,
@@ -19,7 +29,9 @@ def apply_model_system_prompt_to_body(params: dict, form_data: dict, user) -> di
        }
    else:
        template_params = {}
+
    system = prompt_template(system, **template_params)
+
    form_data["messages"] = add_or_update_system_message(
        system, form_data.get("messages", [])
    )
@@ -55,38 +67,49 @@ def apply_model_params_to_body_openai(params: dict, form_data: dict) -> dict:


 def apply_model_params_to_body_ollama(params: dict, form_data: dict) -> dict:
-    opts = [
-        "temperature",
-        "top_p",
-        "seed",
-        "mirostat",
-        "mirostat_eta",
-        "mirostat_tau",
-        "num_ctx",
-        "num_batch",
-        "num_keep",
-        "repeat_last_n",
-        "tfs_z",
-        "top_k",
-        "min_p",
-        "use_mmap",
-        "use_mlock",
-        "num_thread",
-        "num_gpu",
-    ]
-    mappings = {i: lambda x: x for i in opts}
-    form_data = apply_model_params_to_body(params, form_data, mappings)
-
+    # Convert OpenAI parameter names to Ollama parameter names if needed.
    name_differences = {
        "max_tokens": "num_predict",
-        "frequency_penalty": "repeat_penalty",
    }

    for key, value in name_differences.items():
        if (param := params.get(key, None)) is not None:
-            form_data[value] = param
+            # Copy the parameter to new name then delete it, to prevent Ollama warning of invalid option provided
+            params[value] = params[key]
+            del params[key]

-    return form_data
+    # See https://github.com/ollama/ollama/blob/main/docs/api.md#request-8
+    mappings = {
+        "temperature": float,
+        "top_p": float,
+        "seed": lambda x: x,
+        "mirostat": int,
+        "mirostat_eta": float,
+        "mirostat_tau": float,
+        "num_ctx": int,
+        "num_batch": int,
+        "num_keep": int,
+        "num_predict": int,
+        "repeat_last_n": int,
+        "top_k": int,
+        "min_p": float,
+        "typical_p": float,
+        "repeat_penalty": float,
+        "presence_penalty": float,
+        "frequency_penalty": float,
+        "penalize_newline": bool,
+        "stop": lambda x: [bytes(s, "utf-8").decode("unicode_escape") for s in x],
+        "numa": bool,
+        "num_gpu": int,
+        "main_gpu": int,
+        "low_vram": bool,
+        "vocab_only": bool,
+        "use_mmap": bool,
+        "use_mlock": bool,
+        "num_thread": int,
+    }
+
+    return apply_model_params_to_body(params, form_data, mappings)


 def convert_messages_openai_to_ollama(messages: list[dict]) -> list[dict]:
@@ -97,11 +120,38 @@ def convert_messages_openai_to_ollama(messages: list[dict]) -> list[dict]:
        new_message = {"role": message["role"]}

        content = message.get("content", [])
+        tool_calls = message.get("tool_calls", None)
+        tool_call_id = message.get("tool_call_id", None)

        # Check if the content is a string (just a simple message)
        if isinstance(content, str):
            # If the content is a string, it's pure text
            new_message["content"] = content
+
+            # If message is a tool call, add the tool call id to the message
+            if tool_call_id:
+                new_message["tool_call_id"] = tool_call_id
+
+        elif tool_calls:
+            # If tool calls are present, add them to the message
+            ollama_tool_calls = []
+            for tool_call in tool_calls:
+                ollama_tool_call = {
+                    "index": tool_call.get("index", 0),
+                    "id": tool_call.get("id", None),
+                    "function": {
+                        "name": tool_call.get("function", {}).get("name", ""),
+                        "arguments": json.loads(
+                            tool_call.get("function", {}).get("arguments", {})
+                        ),
+                    },
+                }
+                ollama_tool_calls.append(ollama_tool_call)
+            new_message["tool_calls"] = ollama_tool_calls
+
+            # Put the content to empty string (Ollama requires an empty string for tool calls)
+            new_message["content"] = ""
+
        else:
            # Otherwise, assume the content is a list of dicts, e.g., text followed by an image URL
            content_text = ""
@@ -155,37 +205,32 @@ def convert_payload_openai_to_ollama(openai_payload: dict) -> dict:
    )
    ollama_payload["stream"] = openai_payload.get("stream", False)

+    if "tools" in openai_payload:
+        ollama_payload["tools"] = openai_payload["tools"]
+
    if "format" in openai_payload:
        ollama_payload["format"] = openai_payload["format"]

    # If there are advanced parameters in the payload, format them in Ollama's options field
-    ollama_options = {}
-
    if openai_payload.get("options"):
        ollama_payload["options"] = openai_payload["options"]
        ollama_options = openai_payload["options"]

-    # Handle parameters which map directly
-    for param in ["temperature", "top_p", "seed"]:
-        if param in openai_payload:
-            ollama_options[param] = openai_payload[param]
+        # Re-Mapping OpenAI's `max_tokens` -> Ollama's `num_predict`
+        if "max_tokens" in ollama_options:
+            ollama_options["num_predict"] = ollama_options["max_tokens"]
+            del ollama_options[
+                "max_tokens"
+            ]  # To prevent Ollama warning of invalid option provided

-    # Mapping OpenAI's `max_tokens` -> Ollama's `num_predict`
-    if "max_completion_tokens" in openai_payload:
-        ollama_options["num_predict"] = openai_payload["max_completion_tokens"]
-    elif "max_tokens" in openai_payload:
-        ollama_options["num_predict"] = openai_payload["max_tokens"]
+        # Ollama lacks a "system" prompt option. It has to be provided as a direct parameter, so we copy it down.
+        if "system" in ollama_options:
+            ollama_payload["system"] = ollama_options["system"]
+            del ollama_options[
+                "system"
+            ]  # To prevent Ollama warning of invalid option provided

-    # Handle frequency / presence_penalty, which needs renaming and checking
-    if "frequency_penalty" in openai_payload:
-        ollama_options["repeat_penalty"] = openai_payload["frequency_penalty"]
-
-    if "presence_penalty" in openai_payload and "penalty" not in ollama_options:
-        # We are assuming presence penalty uses a similar concept in Ollama, which needs custom handling if exists.
-        ollama_options["new_topic_penalty"] = openai_payload["presence_penalty"]
-
-    # Add options to payload if any have been set
-    if ollama_options:
-        ollama_payload["options"] = ollama_options
+    if "metadata" in openai_payload:
+        ollama_payload["metadata"] = openai_payload["metadata"]

    return ollama_payload
@@ -2,6 +2,7 @@ from datetime import datetime
 from io import BytesIO
 from pathlib import Path
 from typing import Dict, Any, List
+from html import escape

 from markdown import markdown

@@ -41,13 +42,13 @@ class PDFGenerator:

    def _build_html_message(self, message: Dict[str, Any]) -> str:
        """Build HTML for a single message."""
-        role = message.get("role", "user")
-        content = message.get("content", "")
+        role = escape(message.get("role", "user"))
+        content = escape(message.get("content", ""))
        timestamp = message.get("timestamp")

-        model = message.get("model") if role == "assistant" else ""
+        model = escape(message.get("model") if role == "assistant" else "")

-        date_str = self.format_timestamp(timestamp) if timestamp else ""
+        date_str = escape(self.format_timestamp(timestamp) if timestamp else "")

        # extends pymdownx extension to convert markdown to html.
        # - https://facelessuser.github.io/pymdown-extensions/usage_notes/
@@ -76,6 +77,7 @@ class PDFGenerator:

    def _generate_html_body(self) -> str:
        """Generate the full HTML body for the PDF."""
+        escaped_title = escape(self.form_data.title)
        return f"""
        <html>
            <head>
@@ -84,7 +86,7 @@ class PDFGenerator:
            <body>
            <div>
                <div>
-                    <h2>{self.form_data.title}</h2>
+                    <h2>{escaped_title}</h2>
                    {self.messages_html}
                </div>
            </div>
@@ -167,9 +167,14 @@ def load_function_module_by_id(function_id, content=None):

 def install_frontmatter_requirements(requirements):
    if requirements:
-        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])
+        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])
+        except Exception as e:
+            log.error(f"Error installing package: {req}")
+            raise e
+
    else:
        log.info("No requirements found in frontmatter.")
@@ -1,15 +1,101 @@
 import json
+from uuid import uuid4
 from open_webui.utils.misc import (
    openai_chat_chunk_message_template,
    openai_chat_completion_message_template,
 )


+def convert_ollama_tool_call_to_openai(tool_calls: dict) -> dict:
+    openai_tool_calls = []
+    for tool_call in tool_calls:
+        openai_tool_call = {
+            "index": tool_call.get("index", 0),
+            "id": tool_call.get("id", f"call_{str(uuid4())}"),
+            "type": "function",
+            "function": {
+                "name": tool_call.get("function", {}).get("name", ""),
+                "arguments": json.dumps(
+                    tool_call.get("function", {}).get("arguments", {})
+                ),
+            },
+        }
+        openai_tool_calls.append(openai_tool_call)
+    return openai_tool_calls
+
+
+def convert_ollama_usage_to_openai(data: dict) -> dict:
+    return {
+        "response_token/s": (
+            round(
+                (
+                    (
+                        data.get("eval_count", 0)
+                        / ((data.get("eval_duration", 0) / 10_000_000))
+                    )
+                    * 100
+                ),
+                2,
+            )
+            if data.get("eval_duration", 0) > 0
+            else "N/A"
+        ),
+        "prompt_token/s": (
+            round(
+                (
+                    (
+                        data.get("prompt_eval_count", 0)
+                        / ((data.get("prompt_eval_duration", 0) / 10_000_000))
+                    )
+                    * 100
+                ),
+                2,
+            )
+            if data.get("prompt_eval_duration", 0) > 0
+            else "N/A"
+        ),
+        "total_duration": data.get("total_duration", 0),
+        "load_duration": data.get("load_duration", 0),
+        "prompt_eval_count": data.get("prompt_eval_count", 0),
+        "prompt_tokens": int(
+            data.get("prompt_eval_count", 0)
+        ),  # This is the OpenAI compatible key
+        "prompt_eval_duration": data.get("prompt_eval_duration", 0),
+        "eval_count": data.get("eval_count", 0),
+        "completion_tokens": int(
+            data.get("eval_count", 0)
+        ),  # This is the OpenAI compatible key
+        "eval_duration": data.get("eval_duration", 0),
+        "approximate_total": (lambda s: f"{s // 3600}h{(s % 3600) // 60}m{s % 60}s")(
+            (data.get("total_duration", 0) or 0) // 1_000_000_000
+        ),
+        "total_tokens": int(  # This is the OpenAI compatible key
+            data.get("prompt_eval_count", 0) + data.get("eval_count", 0)
+        ),
+        "completion_tokens_details": {  # This is the OpenAI compatible key
+            "reasoning_tokens": 0,
+            "accepted_prediction_tokens": 0,
+            "rejected_prediction_tokens": 0,
+        },
+    }
+
+
 def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
    model = ollama_response.get("model", "ollama")
    message_content = ollama_response.get("message", {}).get("content", "")
+    tool_calls = ollama_response.get("message", {}).get("tool_calls", None)
+    openai_tool_calls = None

-    response = openai_chat_completion_message_template(model, message_content)
+    if tool_calls:
+        openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
+
+    data = ollama_response
+
+    usage = convert_ollama_usage_to_openai(data)
+
+    response = openai_chat_completion_message_template(
+        model, message_content, openai_tool_calls, usage
+    )
    return response


@@ -19,52 +105,20 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)

        model = data.get("model", "ollama")
        message_content = data.get("message", {}).get("content", "")
+        tool_calls = data.get("message", {}).get("tool_calls", None)
+        openai_tool_calls = None
+
+        if tool_calls:
+            openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
+
        done = data.get("done", False)

        usage = None
        if done:
-            usage = {
-                "response_token/s": (
-                    round(
-                        (
-                            (
-                                data.get("eval_count", 0)
-                                / ((data.get("eval_duration", 0) / 10_000_000))
-                            )
-                            * 100
-                        ),
-                        2,
-                    )
-                    if data.get("eval_duration", 0) > 0
-                    else "N/A"
-                ),
-                "prompt_token/s": (
-                    round(
-                        (
-                            (
-                                data.get("prompt_eval_count", 0)
-                                / ((data.get("prompt_eval_duration", 0) / 10_000_000))
-                            )
-                            * 100
-                        ),
-                        2,
-                    )
-                    if data.get("prompt_eval_duration", 0) > 0
-                    else "N/A"
-                ),
-                "total_duration": data.get("total_duration", 0),
-                "load_duration": data.get("load_duration", 0),
-                "prompt_eval_count": data.get("prompt_eval_count", 0),
-                "prompt_eval_duration": data.get("prompt_eval_duration", 0),
-                "eval_count": data.get("eval_count", 0),
-                "eval_duration": data.get("eval_duration", 0),
-                "approximate_total": (
-                    lambda s: f"{s // 3600}h{(s % 3600) // 60}m{s % 60}s"
-                )((data.get("total_duration", 0) or 0) // 1_000_000_000),
-            }
+            usage = convert_ollama_usage_to_openai(data)

        data = openai_chat_chunk_message_template(
-            model, message_content if not done else None, usage
+            model, message_content if not done else None, openai_tool_calls, usage
        )

        line = f"data: {json.dumps(data)}\n\n"
@@ -22,7 +22,7 @@ def get_task_model_id(
    # Set the task model
    task_model_id = default_model_id
    # Check if the user has a custom task model and use that model
-    if models[task_model_id]["owned_by"] == "ollama":
+    if models[task_model_id].get("owned_by") == "ollama":
        if task_model and task_model in models:
            task_model_id = task_model
    else:
@@ -32,6 +32,12 @@ def get_task_model_id(
    return task_model_id


+def prompt_variables_template(template: str, variables: dict[str, str]) -> str:
+    for variable, value in variables.items():
+        template = template.replace(variable, value)
+    return template
+
+
 def prompt_template(
    template: str, user_name: Optional[str] = None, user_location: Optional[str] = None
 ) -> str:
@@ -61,6 +61,12 @@ def get_tools(
            )

        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"
+
            # Remove internal parameters
            spec["parameters"]["properties"] = {
                key: val
@@ -73,6 +79,13 @@ def get_tools(
            # 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)
+
+            if callable.__doc__ and callable.__doc__.strip() != "":
+                s = re.split(":(param|return)", callable.__doc__, 1)
+                spec["description"] = s[0]
+            else:
+                spec["description"] = function_name
+
            # TODO: This needs to be a pydantic model
            tool_dict = {
                "toolkit_id": tool_id,
@@ -2,14 +2,14 @@ import json
 import logging

 import requests
-from open_webui.config import WEBUI_FAVICON_URL, WEBUI_NAME
+from open_webui.config import WEBUI_FAVICON_URL
 from open_webui.env import SRC_LOG_LEVELS, VERSION

 log = logging.getLogger(__name__)
 log.setLevel(SRC_LOG_LEVELS["WEBHOOK"])


-def post_webhook(url: str, message: str, event_data: dict) -> bool:
+def post_webhook(name: str, url: str, message: str, event_data: dict) -> bool:
    try:
        log.debug(f"post_webhook: {url}, {message}, {event_data}")
        payload = {}
@@ -39,7 +39,7 @@ def post_webhook(url: str, message: str, event_data: dict) -> bool:
                "sections": [
                    {
                        "activityTitle": message,
-                        "activitySubtitle": f"{WEBUI_NAME} ({VERSION}) - {action}",
+                        "activitySubtitle": f"{name} ({VERSION}) - {action}",
                        "activityImage": WEBUI_FAVICON_URL,
                        "facts": facts,
                        "markdown": True,
@@ -1,17 +1,14 @@
-fastapi==0.111.0
+fastapi==0.115.7
 uvicorn[standard]==0.30.6
-pydantic==2.9.2
+pydantic==2.10.6
 python-multipart==0.0.18

-Flask==3.1.0
-Flask-Cors==5.0.0
-
 python-socketio==5.11.3
-python-jose==3.3.0
+python-jose==3.4.0
 passlib[bcrypt]==1.7.4

 requests==2.32.3
-aiohttp==3.11.8
+aiohttp==3.11.11
 async-timeout
 aiocache
 aiofiles
@@ -32,6 +29,8 @@ boto3==1.35.53
 argon2-cffi==23.1.0
 APScheduler==3.10.4

+RestrictedPython==8.0
+
 # AI libraries
 openai
 anthropic
@@ -45,8 +44,8 @@ fake-useragent==1.5.1
 chromadb==0.6.2
 pymilvus==2.5.0
 qdrant-client~=1.12.0
-opensearch-py==2.7.1
-
+opensearch-py==2.8.0
+playwright==1.49.1 # Caution: version must match docker-compose.playwright.yaml

 transformers
 sentence-transformers==3.3.1
@@ -57,10 +56,10 @@ einops==0.8.0
 ftfy==6.2.3
 pypdf==4.3.1
 fpdf2==2.8.2
-pymdown-extensions==10.11.2
+pymdown-extensions==10.14.2
 docx2txt==0.8
 python-pptx==1.0.0
-unstructured==0.15.9
+unstructured==0.16.17
 nltk==3.9.1
 Markdown==3.7
 pypandoc==1.13
@@ -71,16 +70,16 @@ xlrd==2.0.1
 validators==0.34.0
 psutil
 sentencepiece
-soundfile==0.12.1
+soundfile==0.13.1

-opencv-python-headless==4.10.0.84
+opencv-python-headless==4.11.0.86
 rapidocr-onnxruntime==1.3.24
 rank-bm25==0.2.2

-faster-whisper==1.0.3
+faster-whisper==1.1.1

 PyJWT[crypto]==2.10.1
-authlib==1.3.2
+authlib==1.4.1

 black==24.8.0
 langfuse==2.44.0
@@ -89,7 +88,7 @@ pytube==15.0.0

 extract_msg
 pydub
-duckduckgo-search~=7.2.1
+duckduckgo-search~=7.3.2

 ## Google Drive
 google-api-python-client
@@ -104,5 +103,12 @@ pytest-docker~=3.1.1
 googleapis-common-protos==1.63.2
 google-cloud-storage==2.19.0

+azure-identity==1.20.0
+azure-storage-blob==12.24.1
+
+
 ## LDAP
 ldap3==2.9.1
+
+## Firecrawl
+firecrawl-py==1.12.0
@@ -3,6 +3,17 @@
 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
+        echo "Installing Playwright browsers..."
+        playwright install chromium
+        playwright install-deps chromium
+    fi
+
+    python -c "import nltk; nltk.download('punkt_tab')"
+fi
+
 KEY_FILE=.webui_secret_key

 PORT="${PORT:-8080}"
@@ -6,6 +6,17 @@ SETLOCAL ENABLEDELAYEDEXPANSION
 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%" == "" (
+        echo Installing Playwright browsers...
+        playwright install chromium
+        playwright install-deps chromium
+    )
+
+    python -c "import nltk; nltk.download('punkt_tab')"
+)
+
 SET "KEY_FILE=.webui_secret_key"
 IF "%PORT%"=="" SET PORT=8080
 IF "%HOST%"=="" SET HOST=0.0.0.0
@@ -0,0 +1,10 @@
+services:
+  playwright:
+    image: mcr.microsoft.com/playwright:v1.49.1-noble # Version must match requirements.txt
+    container_name: playwright
+    command: npx -y playwright@1.49.1 run-server --port 3000 --host 0.0.0.0
+
+  open-webui:
+    environment:
+      - 'RAG_WEB_LOADER_ENGINE=playwright'
+      - 'PLAYWRIGHT_WS_URI=ws://playwright:3000'
@@ -1,6 +1,6 @@
 {
 	"name": "open-webui",
-	"version": "0.5.6",
+	"version": "0.5.16",
 	"private": true,
 	"scripts": {
 		"dev": "npm run pyodide:fetch && vite dev --host",
@@ -25,10 +25,11 @@
 		"@sveltejs/adapter-static": "^3.0.2",
 		"@sveltejs/kit": "^2.5.20",
 		"@sveltejs/vite-plugin-svelte": "^3.1.1",
+		"@tailwindcss/container-queries": "^0.1.1",
+		"@tailwindcss/postcss": "^4.0.0",
 		"@tailwindcss/typography": "^0.5.13",
 		"@typescript-eslint/eslint-plugin": "^6.17.0",
 		"@typescript-eslint/parser": "^6.17.0",
-		"autoprefixer": "^10.4.16",
 		"cypress": "^13.15.0",
 		"eslint": "^8.56.0",
 		"eslint-config-prettier": "^9.1.0",
@@ -42,11 +43,11 @@
 		"svelte": "^4.2.18",
 		"svelte-check": "^3.8.5",
 		"svelte-confetti": "^1.3.2",
-		"tailwindcss": "^3.3.3",
+		"tailwindcss": "^4.0.0",
 		"tslib": "^2.4.1",
 		"typescript": "^5.5.4",
 		"vite": "^5.4.14",
-		"vitest": "^1.6.0"
+		"vitest": "^1.6.1"
 	},
 	"type": "module",
 	"dependencies": {
@@ -84,6 +85,7 @@
 		"idb": "^7.1.1",
 		"js-sha256": "^0.10.1",
 		"katex": "^0.16.21",
+		"kokoro-js": "^1.1.1",
 		"marked": "^9.1.0",
 		"mermaid": "^10.9.3",
 		"paneforge": "^0.0.6",
@@ -98,13 +100,15 @@
 		"prosemirror-schema-list": "^1.4.1",
 		"prosemirror-state": "^1.4.3",
 		"prosemirror-view": "^1.34.3",
-		"pyodide": "^0.26.1",
+		"pyodide": "^0.27.2",
 		"socket.io-client": "^4.2.0",
 		"sortablejs": "^1.15.2",
 		"svelte-sonner": "^0.3.19",
 		"tippy.js": "^6.3.7",
 		"turndown": "^7.2.0",
-		"uuid": "^9.0.1"
+		"undici": "^7.3.0",
+		"uuid": "^9.0.1",
+		"vite-plugin-static-copy": "^2.2.0"
 	},
 	"engines": {
 		"node": ">=18.13.0 <=22.x.x",
@@ -1,6 +1,5 @@
 export default {
 	plugins: {
-		tailwindcss: {},
-		autoprefixer: {}
+		'@tailwindcss/postcss': {}
 	}
 };
@@ -6,20 +6,17 @@ authors = [
 ]
 license = { file = "LICENSE" }
 dependencies = [
-    "fastapi==0.111.0",
+    "fastapi==0.115.7",
    "uvicorn[standard]==0.30.6",
-    "pydantic==2.9.2",
+    "pydantic==2.10.6",
    "python-multipart==0.0.18",

-    "Flask==3.1.0",
-    "Flask-Cors==5.0.0",
-
    "python-socketio==5.11.3",
-    "python-jose==3.3.0",
+    "python-jose==3.4.0",
    "passlib[bcrypt]==1.7.4",

    "requests==2.32.3",
-    "aiohttp==3.11.8",
+    "aiohttp==3.11.11",
    "async-timeout",
    "aiocache",
    "aiofiles",
@@ -40,6 +37,9 @@ dependencies = [
    "argon2-cffi==23.1.0",
    "APScheduler==3.10.4",

+
+    "RestrictedPython==8.0",
+
    "openai",
    "anthropic",
    "google-generativeai==0.7.2",
@@ -52,7 +52,8 @@ dependencies = [
    "chromadb==0.6.2",
    "pymilvus==2.5.0",
    "qdrant-client~=1.12.0",
-    "opensearch-py==2.7.1",
+    "opensearch-py==2.8.0",
+    "playwright==1.49.1",

    "transformers",
    "sentence-transformers==3.3.1",
@@ -62,10 +63,10 @@ dependencies = [
    "ftfy==6.2.3",
    "pypdf==4.3.1",
    "fpdf2==2.8.2",
-    "pymdown-extensions==10.11.2",
+    "pymdown-extensions==10.14.2",
    "docx2txt==0.8",
    "python-pptx==1.0.0",
-    "unstructured==0.15.9",
+    "unstructured==0.16.17",
    "nltk==3.9.1",
    "Markdown==3.7",
    "pypandoc==1.13",
@@ -76,16 +77,16 @@ dependencies = [
    "validators==0.34.0",
    "psutil",
    "sentencepiece",
-    "soundfile==0.12.1",
+    "soundfile==0.13.1",

-    "opencv-python-headless==4.10.0.84",
+    "opencv-python-headless==4.11.0.86",
    "rapidocr-onnxruntime==1.3.24",
    "rank-bm25==0.2.2",

-    "faster-whisper==1.0.3",
+    "faster-whisper==1.1.1",

    "PyJWT[crypto]==2.10.1",
-    "authlib==1.3.2",
+    "authlib==1.4.1",

    "black==24.8.0",
    "langfuse==2.44.0",
@@ -94,7 +95,7 @@ dependencies = [

    "extract_msg",
    "pydub",
-    "duckduckgo-search~=7.2.1",
+    "duckduckgo-search~=7.3.2",

    "google-api-python-client",
    "google-auth-httplib2",
@@ -108,7 +109,13 @@ dependencies = [
    "googleapis-common-protos==1.63.2",
    "google-cloud-storage==2.19.0",

+    "azure-identity==1.20.0",
+    "azure-storage-blob==12.24.1",
+
    "ldap3==2.9.1",
+
+    "firecrawl-py==1.12.0",
+
    "gcp-storage-emulator>=2024.8.3",
 ]
 readme = "README.md"
@@ -74,6 +74,7 @@ usage() {
    echo "  --enable-api[port=PORT]    Enable API and expose it on the specified port."
    echo "  --webui[port=PORT]         Set the port for the web user interface."
    echo "  --data[folder=PATH]        Bind mount for ollama data folder (by default will create the 'ollama' volume)."
+    echo "  --playwright               Enable Playwright support for web scraping."
    echo "  --build                    Build the docker image before running the compose project."
    echo "  --drop                     Drop the compose project."
    echo "  -q, --quiet                Run script in headless mode."
@@ -100,6 +101,7 @@ webui_port=3000
 headless=false
 build_image=false
 kill_compose=false
+enable_playwright=false

 # Function to extract value from the parameter
 extract_value() {
@@ -129,6 +131,9 @@ while [[ $# -gt 0 ]]; do
            value=$(extract_value "$key")
            data_dir=${value:-"./ollama-data"}
            ;;
+        --playwright)
+            enable_playwright=true
+            ;;
        --drop)
            kill_compose=true
            ;;
@@ -182,6 +187,9 @@ else
        DEFAULT_COMPOSE_COMMAND+=" -f docker-compose.data.yaml"
        export OLLAMA_DATA_DIR=$data_dir # Set OLLAMA_DATA_DIR environment variable
    fi
+    if [[ $enable_playwright == true ]]; then
+        DEFAULT_COMPOSE_COMMAND+=" -f docker-compose.playwright.yaml"
+    fi
    if [[ -n $webui_port ]]; then
        export OPEN_WEBUI_PORT=$webui_port # Set OPEN_WEBUI_PORT environment variable
    fi
@@ -201,6 +209,7 @@ echo -e "   ${GREEN}${BOLD}GPU Count:${NC} ${OLLAMA_GPU_COUNT:-Not Enabled}"
 echo -e "   ${GREEN}${BOLD}WebAPI Port:${NC} ${OLLAMA_WEBAPI_PORT:-Not Enabled}"
 echo -e "   ${GREEN}${BOLD}Data Folder:${NC} ${data_dir:-Using ollama volume}"
 echo -e "   ${GREEN}${BOLD}WebUI Port:${NC} $webui_port"
+echo -e "   ${GREEN}${BOLD}Playwright:${NC} ${enable_playwright:-false}"
 echo

 if [[ $headless == true ]]; then
@@ -9,12 +9,46 @@ const packages = [
 	'scikit-learn',
 	'scipy',
 	'regex',
-	'seaborn'
+	'sympy',
+	'tiktoken',
+	'seaborn',
+	'pytz'
 ];

 import { loadPyodide } from 'pyodide';
+import { setGlobalDispatcher, ProxyAgent } from 'undici';
 import { writeFile, readFile, copyFile, readdir, rmdir } from 'fs/promises';

+/**
+ * Loading network proxy configurations from the environment variables.
+ * And the proxy config with lowercase name has the highest priority to use.
+ */
+function initNetworkProxyFromEnv() {
+	// we assume all subsequent requests in this script are HTTPS:
+	// https://cdn.jsdelivr.net
+	// https://pypi.org
+	// https://files.pythonhosted.org
+	const allProxy = process.env.all_proxy || process.env.ALL_PROXY;
+	const httpsProxy = process.env.https_proxy || process.env.HTTPS_PROXY;
+	const httpProxy = process.env.http_proxy || process.env.HTTP_PROXY;
+	const preferedProxy = httpsProxy || allProxy || httpProxy;
+	/**
+	 * use only http(s) proxy because socks5 proxy is not supported currently:
+	 * @see https://github.com/nodejs/undici/issues/2224
+	 */
+	if (!preferedProxy || !preferedProxy.startsWith('http')) return;
+	let preferedProxyURL;
+	try {
+		preferedProxyURL = new URL(preferedProxy).toString();
+	} catch {
+		console.warn(`Invalid network proxy URL: "${preferedProxy}"`);
+		return;
+	}
+	const dispatcher = new ProxyAgent({ uri: preferedProxyURL });
+	setGlobalDispatcher(dispatcher);
+	console.log(`Initialized network proxy "${preferedProxy}" from env`);
+}
+
 async function downloadPackages() {
 	console.log('Setting up pyodide + micropip');

@@ -81,5 +115,6 @@ async function copyPyodide() {
 	}
 }

+initNetworkProxyFromEnv();
 await downloadPackages();
 await copyPyodide();
@@ -1,3 +1,5 @@
+@reference "./tailwind.css";
+
@font-face {
 	font-family: 'Inter';
 	src: url('/assets/fonts/Inter-Variable.ttf');
@@ -53,11 +55,11 @@ math {
 }

 .markdown-prose {
-	@apply prose dark:prose-invert prose-blockquote:border-gray-100 prose-blockquote:dark:border-gray-800 prose-blockquote:border-l-2 prose-blockquote:not-italic prose-blockquote:font-normal  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;
+	@apply prose dark:prose-invert prose-blockquote:border-s-gray-100 prose-blockquote:dark:border-gray-800 prose-blockquote:border-s-2 prose-blockquote:not-italic prose-blockquote:font-normal  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;
 }

 .markdown-prose-xs {
-	@apply text-xs prose dark:prose-invert prose-blockquote:border-gray-100 prose-blockquote:dark:border-gray-800 prose-blockquote:border-l-2 prose-blockquote:not-italic prose-blockquote:font-normal  prose-headings:font-semibold prose-hr:my-0  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;
+	@apply text-xs prose dark:prose-invert prose-blockquote:border-s-gray-100 prose-blockquote:dark:border-gray-800 prose-blockquote:border-s-2 prose-blockquote:not-italic prose-blockquote:font-normal  prose-headings:font-semibold prose-hr:my-0  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;
 }

 .markdown a {
@@ -99,7 +101,7 @@ li p {

 /* Dark theme scrollbar styles */
 .dark ::-webkit-scrollbar-thumb {
-	background-color: rgba(33, 33, 33, 0.8); /* Darker color for dark theme */
+	background-color: rgba(42, 42, 42, 0.8); /* Darker color for dark theme */
 	border-color: rgba(0, 0, 0, var(--tw-border-opacity));
 }

@@ -217,8 +219,18 @@ input[type='number'] {
 	width: 100%;
 }

-.cm-scroller {
-	@apply scrollbar-hidden;
+.cm-scroller:active::-webkit-scrollbar-thumb,
+.cm-scroller:focus::-webkit-scrollbar-thumb,
+.cm-scroller:hover::-webkit-scrollbar-thumb {
+	visibility: visible;
+}
+
+.cm-scroller::-webkit-scrollbar-thumb {
+	visibility: hidden;
+}
+
+.cm-scroller::-webkit-scrollbar-corner {
+	display: none;
 }

 .cm-editor.cm-focused {
@@ -21,6 +21,7 @@
 			title="Open WebUI"
 			href="/opensearch.xml"
 		/>
+		<script src="/static/loader.js" defer></script>

 		<script>
 			function resizeIframe(obj) {
@@ -459,7 +459,7 @@ export const getChatById = async (token: string, id: string) => {
 			return json;
 		})
 		.catch((err) => {
-			error = err;
+			error = err.detail;

 			console.log(err);
 			return null;
@@ -580,7 +580,7 @@ export const toggleChatPinnedStatusById = async (token: string, id: string) => {
 	return res;
 };

-export const cloneChatById = async (token: string, id: string) => {
+export const cloneChatById = async (token: string, id: string, title?: string) => {
 	let error = null;

 	const res = await fetch(`${WEBUI_API_BASE_URL}/chats/${id}/clone`, {
@@ -589,7 +589,10 @@ export const cloneChatById = async (token: string, id: string) => {
 			Accept: 'application/json',
 			'Content-Type': 'application/json',
 			...(token && { authorization: `Bearer ${token}` })
-		}
+		},
+		body: JSON.stringify({
+			...(title && { title: title })
+		})
 	})
 		.then(async (res) => {
 			if (!res.ok) throw await res.json();
@@ -58,6 +58,120 @@ export const exportConfig = async (token: string) => {
 	return res;
 };

+export const getDirectConnectionsConfig = async (token: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/direct_connections`, {
+		method: 'GET',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		}
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			error = err.detail;
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const setDirectConnectionsConfig = async (token: string, config: object) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/direct_connections`, {
+		method: 'POST',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			...config
+		})
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			error = err.detail;
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const getCodeExecutionConfig = async (token: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/code_execution`, {
+		method: 'GET',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		}
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+			error = err.detail;
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const setCodeExecutionConfig = async (token: string, config: object) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/configs/code_execution`, {
+		method: 'POST',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			...config
+		})
+	})
+		.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 getModelsConfig = async (token: string) => {
 	let error = null;

@@ -1,6 +1,11 @@
 import { WEBUI_API_BASE_URL, WEBUI_BASE_URL } from '$lib/constants';
+import { getOpenAIModelsDirect } from './openai';

-export const getModels = async (token: string = '', base: boolean = false) => {
+export const getModels = async (
+	token: string = '',
+	connections: object | null = null,
+	base: boolean = false
+) => {
 	let error = null;
 	const res = await fetch(`${WEBUI_BASE_URL}/api/models${base ? '/base' : ''}`, {
 		method: 'GET',
@@ -25,6 +30,111 @@ export const getModels = async (token: string = '', base: boolean = false) => {
 	}

 	let models = res?.data ?? [];
+
+	if (connections && !base) {
+		let localModels = [];
+
+		if (connections) {
+			const OPENAI_API_BASE_URLS = connections.OPENAI_API_BASE_URLS;
+			const OPENAI_API_KEYS = connections.OPENAI_API_KEYS;
+			const OPENAI_API_CONFIGS = connections.OPENAI_API_CONFIGS;
+
+			const requests = [];
+			for (const idx in OPENAI_API_BASE_URLS) {
+				const url = OPENAI_API_BASE_URLS[idx];
+
+				if (idx.toString() in OPENAI_API_CONFIGS) {
+					const apiConfig = OPENAI_API_CONFIGS[idx.toString()] ?? {};
+
+					const enable = apiConfig?.enable ?? true;
+					const modelIds = apiConfig?.model_ids ?? [];
+
+					if (enable) {
+						if (modelIds.length > 0) {
+							const modelList = {
+								object: 'list',
+								data: modelIds.map((modelId) => ({
+									id: modelId,
+									name: modelId,
+									owned_by: 'openai',
+									openai: { id: modelId },
+									urlIdx: idx
+								}))
+							};
+
+							requests.push(
+								(async () => {
+									return modelList;
+								})()
+							);
+						} else {
+							requests.push(
+								(async () => {
+									return await getOpenAIModelsDirect(url, OPENAI_API_KEYS[idx])
+										.then((res) => {
+											return res;
+										})
+										.catch((err) => {
+											return {
+												object: 'list',
+												data: [],
+												urlIdx: idx
+											};
+										});
+								})()
+							);
+						}
+					} else {
+						requests.push(
+							(async () => {
+								return {
+									object: 'list',
+									data: [],
+									urlIdx: idx
+								};
+							})()
+						);
+					}
+				}
+			}
+
+			const responses = await Promise.all(requests);
+
+			for (const idx in responses) {
+				const response = responses[idx];
+				const apiConfig = OPENAI_API_CONFIGS[idx.toString()] ?? {};
+
+				let models = Array.isArray(response) ? response : (response?.data ?? []);
+				models = models.map((model) => ({ ...model, openai: { id: model.id }, urlIdx: idx }));
+
+				const prefixId = apiConfig.prefix_id;
+				if (prefixId) {
+					for (const model of models) {
+						model.id = `${prefixId}.${model.id}`;
+					}
+				}
+
+				localModels = localModels.concat(models);
+			}
+		}
+
+		models = models.concat(
+			localModels.map((model) => ({
+				...model,
+				name: model?.name ?? model?.id,
+				direct: true
+			}))
+		);
+
+		// Remove duplicates
+		const modelsMap = {};
+		for (const model of models) {
+			modelsMap[model.id] = model;
+		}
+
+		models = Object.values(modelsMap);
+	}
+
 	return models;
 };

@@ -880,13 +990,14 @@ export const getChangelog = async () => {
 	return res;
 };

-export const getVersionUpdates = async () => {
+export const getVersionUpdates = async (token: string) => {
 	let error = null;

 	const res = await fetch(`${WEBUI_BASE_URL}/api/version/updates`, {
 		method: 'GET',
 		headers: {
-			'Content-Type': 'application/json'
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
 		}
 	})
 		.then(async (res) => {
@@ -219,7 +219,7 @@ export const deleteModelById = async (token: string, id: string) => {
 			return json;
 		})
 		.catch((err) => {
-			error = err;
+			error = err.detail;

 			console.log(err);
 			return null;
@@ -208,6 +208,33 @@ export const updateOpenAIKeys = async (token: string = '', keys: string[]) => {
 	return res.OPENAI_API_KEYS;
 };

+export const getOpenAIModelsDirect = async (url: string, key: string) => {
+	let error = null;
+
+	const res = await fetch(`${url}/models`, {
+		method: 'GET',
+		headers: {
+			Accept: 'application/json',
+			'Content-Type': 'application/json',
+			...(key && { authorization: `Bearer ${key}` })
+		}
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			error = `OpenAI: ${err?.error?.message ?? 'Network Problem'}`;
+			return [];
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
 export const getOpenAIModels = async (token: string, urlIdx?: number) => {
 	let error = null;

@@ -241,33 +268,62 @@ export const getOpenAIModels = async (token: string, urlIdx?: number) => {
 export const verifyOpenAIConnection = async (
 	token: string = '',
 	url: string = 'https://api.openai.com/v1',
-	key: string = ''
+	key: string = '',
+	direct: boolean = false
 ) => {
+	if (!url) {
+		throw 'OpenAI: URL is required';
+	}
+
 	let error = null;
+	let res = null;

-	const res = await fetch(`${OPENAI_API_BASE_URL}/verify`, {
-		method: 'POST',
-		headers: {
-			Accept: 'application/json',
-			Authorization: `Bearer ${token}`,
-			'Content-Type': 'application/json'
-		},
-		body: JSON.stringify({
-			url,
-			key
+	if (direct) {
+		res = await fetch(`${url}/models`, {
+			method: 'GET',
+			headers: {
+				Accept: 'application/json',
+				Authorization: `Bearer ${key}`,
+				'Content-Type': 'application/json'
+			}
 		})
-	})
-		.then(async (res) => {
-			if (!res.ok) throw await res.json();
-			return res.json();
-		})
-		.catch((err) => {
-			error = `OpenAI: ${err?.error?.message ?? 'Network Problem'}`;
-			return [];
-		});
+			.then(async (res) => {
+				if (!res.ok) throw await res.json();
+				return res.json();
+			})
+			.catch((err) => {
+				error = `OpenAI: ${err?.error?.message ?? 'Network Problem'}`;
+				return [];
+			});

-	if (error) {
-		throw error;
+		if (error) {
+			throw error;
+		}
+	} else {
+		res = await fetch(`${OPENAI_API_BASE_URL}/verify`, {
+			method: 'POST',
+			headers: {
+				Accept: 'application/json',
+				Authorization: `Bearer ${token}`,
+				'Content-Type': 'application/json'
+			},
+			body: JSON.stringify({
+				url,
+				key
+			})
+		})
+			.then(async (res) => {
+				if (!res.ok) throw await res.json();
+				return res.json();
+			})
+			.catch((err) => {
+				error = `OpenAI: ${err?.error?.message ?? 'Network Problem'}`;
+				return [];
+			});
+
+		if (error) {
+			throw error;
+		}
 	}

 	return res;
@@ -322,7 +378,7 @@ export const generateOpenAIChatCompletion = async (
 			return res.json();
 		})
 		.catch((err) => {
-			error = `${err?.detail ?? 'Network Problem'}`;
+			error = `${err?.detail ?? err}`;
 			return null;
 		});

@@ -284,14 +284,16 @@ export const updateUserInfo = async (token: string, info: object) => {

 export const getAndUpdateUserLocation = async (token: string) => {
 	const location = await getUserPosition().catch((err) => {
-		throw err;
+		console.log(err);
+		return null;
 	});

 	if (location) {
 		await updateUserInfo(token, { location: location });
 		return location;
 	} else {
-		throw new Error('Failed to get user location');
+		console.log('Failed to get user location');
+		return null;
 	}
 };

@@ -1,12 +1,13 @@
 import { WEBUI_API_BASE_URL } from '$lib/constants';

-export const getGravatarUrl = async (email: string) => {
+export const getGravatarUrl = async (token: string, email: string) => {
 	let error = null;

 	const res = await fetch(`${WEBUI_API_BASE_URL}/utils/gravatar?email=${email}`, {
 		method: 'GET',
 		headers: {
-			'Content-Type': 'application/json'
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
 		}
 	})
 		.then(async (res) => {
@@ -22,13 +23,14 @@ export const getGravatarUrl = async (email: string) => {
 	return res;
 };

-export const formatPythonCode = async (code: string) => {
+export const executeCode = async (token: string, code: string) => {
 	let error = null;

-	const res = await fetch(`${WEBUI_API_BASE_URL}/utils/code/format`, {
+	const res = await fetch(`${WEBUI_API_BASE_URL}/utils/code/execute`, {
 		method: 'POST',
 		headers: {
-			'Content-Type': 'application/json'
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
 		},
 		body: JSON.stringify({
 			code: code
@@ -55,13 +57,48 @@ export const formatPythonCode = async (code: string) => {
 	return res;
 };

-export const downloadChatAsPDF = async (title: string, messages: object[]) => {
+export const formatPythonCode = async (token: string, code: string) => {
+	let error = null;
+
+	const res = await fetch(`${WEBUI_API_BASE_URL}/utils/code/format`, {
+		method: 'POST',
+		headers: {
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
+		},
+		body: JSON.stringify({
+			code: code
+		})
+	})
+		.then(async (res) => {
+			if (!res.ok) throw await res.json();
+			return res.json();
+		})
+		.catch((err) => {
+			console.log(err);
+
+			error = err;
+			if (err.detail) {
+				error = err.detail;
+			}
+			return null;
+		});
+
+	if (error) {
+		throw error;
+	}
+
+	return res;
+};
+
+export const downloadChatAsPDF = async (token: string, title: string, messages: object[]) => {
 	let error = null;

 	const blob = await fetch(`${WEBUI_API_BASE_URL}/utils/pdf`, {
 		method: 'POST',
 		headers: {
-			'Content-Type': 'application/json'
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
 		},
 		body: JSON.stringify({
 			title: title,
@@ -81,13 +118,14 @@ export const downloadChatAsPDF = async (title: string, messages: object[]) => {
 	return blob;
 };

-export const getHTMLFromMarkdown = async (md: string) => {
+export const getHTMLFromMarkdown = async (token: string, md: string) => {
 	let error = null;

 	const res = await fetch(`${WEBUI_API_BASE_URL}/utils/markdown`, {
 		method: 'POST',
 		headers: {
-			'Content-Type': 'application/json'
+			'Content-Type': 'application/json',
+			Authorization: `Bearer ${token}`
 		},
 		body: JSON.stringify({
 			md: md
@@ -20,7 +20,9 @@

 	export let show = false;
 	export let edit = false;
+
 	export let ollama = false;
+	export let direct = false;

 	export let connection = null;

@@ -46,9 +48,11 @@
 	};

 	const verifyOpenAIHandler = async () => {
-		const res = await verifyOpenAIConnection(localStorage.token, url, key).catch((error) => {
-			toast.error(`${error}`);
-		});
+		const res = await verifyOpenAIConnection(localStorage.token, url, key, direct).catch(
+			(error) => {
+				toast.error(`${error}`);
+			}
+		);

 		if (res) {
 			toast.success($i18n.t('Server connection verified'));
@@ -165,7 +169,7 @@

 								<div class="flex-1">
 									<input
-										class="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-none"
+										class="w-full 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')}
@@ -198,7 +202,7 @@
 								</button>
 							</Tooltip>

-							<div class="flex flex-col flex-shrink-0 self-end">
+							<div class="flex flex-col shrink-0 self-end">
 								<Tooltip content={enable ? $i18n.t('Enabled') : $i18n.t('Disabled')}>
 									<Switch bind:state={enable} />
 								</Tooltip>
@@ -211,7 +215,7 @@

 								<div class="flex-1">
 									<SensitiveInput
-										className="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-none"
+										className="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
 										bind:value={key}
 										placeholder={$i18n.t('API Key')}
 										required={!ollama}
@@ -229,7 +233,7 @@
 										)}
 									>
 										<input
-											class="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-none"
+											class="w-full text-sm bg-transparent placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
 											type="text"
 											bind:value={prefixId}
 											placeholder={$i18n.t('Prefix ID')}
@@ -254,7 +258,7 @@
 											<div class=" text-sm flex-1 py-1 rounded-lg">
 												{modelId}
 											</div>
-											<div class="flex-shrink-0">
+											<div class="shrink-0">
 												<button
 													type="button"
 													on:click={() => {
@@ -288,7 +292,7 @@
 							<input
 								class="w-full py-1 text-sm rounded-lg bg-transparent {modelId
 									? ''
-									: 'text-gray-500'} placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-none"
+									: 'text-gray-500'} placeholder:text-gray-300 dark:placeholder:text-gray-700 outline-hidden"
 								bind:value={modelId}
 								placeholder={$i18n.t('Add a model ID')}
 							/>
@@ -68,7 +68,7 @@
 								v{version} - {changelog[version].date}
 							</div>

-							<hr class=" dark:border-gray-800 my-2" />
+							<hr class="border-gray-100 dark:border-gray-850 my-2" />

 							{#each Object.keys(changelog[version]).filter((section) => section !== 'date') as section}
 								<div class="">
@@ -31,13 +31,13 @@
 </script>

 <button
-	class="flex gap-2.5 text-left min-w-[var(--width)] w-full dark:bg-gray-850 dark:text-white bg-white text-black border border-gray-50 dark:border-gray-800 rounded-xl px-3.5 py-3.5"
+	class="flex gap-2.5 text-left min-w-[var(--width)] w-full dark:bg-gray-850 dark:text-white bg-white text-black border border-gray-100 dark:border-gray-850 rounded-xl px-3.5 py-3.5"
 	on:click={() => {
 		onClick();
 		dispatch('closeToast');
 	}}
 >
-	<div class="flex-shrink-0 self-top -translate-y-0.5">
+	<div class="shrink-0 self-top -translate-y-0.5">
 		<img src={'/static/favicon.png'} alt="favicon" class="size-7 rounded-full" />
 	</div>

@@ -30,10 +30,10 @@
 		<SlideShow duration={5000} />

 		<div
-			class="w-full h-full absolute top-0 left-0 bg-gradient-to-t from-20% from-black to-transparent"
+			class="w-full h-full absolute top-0 left-0 bg-linear-to-t from-20% from-black to-transparent"
 		></div>

-		<div class="w-full h-full absolute top-0 left-0 backdrop-blur-sm bg-black/50"></div>
+		<div class="w-full h-full absolute top-0 left-0 backdrop-blur-xs bg-black/50"></div>

 		<div class="relative bg-transparent w-full min-h-screen flex z-10">
 			<div class="flex flex-col justify-end w-full items-center pb-10 text-center">
@@ -65,7 +65,7 @@
 	};

 	const shareHandler = async () => {
-		toast.success($i18n.t('Redirecting you to OpenWebUI Community'));
+		toast.success($i18n.t('Redirecting you to Open WebUI Community'));

 		// remove snapshot from feedbacks
 		const feedbacksToShare = feedbacks.map((f) => {
@@ -131,14 +131,16 @@
 	</div>
 </div>

-<div class="scrollbar-hidden relative whitespace-nowrap overflow-x-auto max-w-full rounded pt-0.5">
+<div
+	class="scrollbar-hidden relative whitespace-nowrap overflow-x-auto max-w-full rounded-sm pt-0.5"
+>
 	{#if (feedbacks ?? []).length === 0}
 		<div class="text-center text-xs text-gray-500 dark:text-gray-400 py-1">
 			{$i18n.t('No feedbacks found')}
 		</div>
 	{:else}
 		<table
-			class="w-full text-sm text-left text-gray-500 dark:text-gray-400 table-auto max-w-full rounded"
+			class="w-full text-sm text-left text-gray-500 dark:text-gray-400 table-auto max-w-full rounded-sm"
 		>
 			<thead
 				class="text-xs text-gray-700 uppercase bg-gray-50 dark:bg-gray-850 dark:text-gray-400 -translate-y-0.5"
@@ -169,7 +171,7 @@
 						<td class=" py-0.5 text-right font-semibold">
 							<div class="flex justify-center">
 								<Tooltip content={feedback?.user?.name}>
-									<div class="flex-shrink-0">
+									<div class="shrink-0">
 										<img
 											src={feedback?.user?.profile_image_url ?? '/user.png'}
 											alt={feedback?.user?.name}
@@ -266,7 +268,7 @@
 					}}
 				>
 					<div class=" self-center mr-2 font-medium line-clamp-1">
-						{$i18n.t('Share to OpenWebUI Community')}
+						{$i18n.t('Share to Open WebUI Community')}
 					</div>

 					<div class=" self-center">
@@ -1,6 +1,8 @@
 <script lang="ts">
 	import * as ort from 'onnxruntime-web';
-	import { AutoModel, AutoTokenizer } from '@huggingface/transformers';
+	import { env, AutoModel, AutoTokenizer } from '@huggingface/transformers';
+
+	env.backends.onnx.wasm.wasmPaths = '/wasm/';

 	import { onMount, getContext } from 'svelte';
 	import { models } from '$lib/stores';
@@ -286,7 +288,7 @@
 					<MagnifyingGlass className="size-3" />
 				</div>
 				<input
-					class=" w-full text-sm pr-4 py-1 rounded-r-xl outline-none bg-transparent"
+					class=" w-full text-sm pr-4 py-1 rounded-r-xl outline-hidden bg-transparent"
 					bind:value={query}
 					placeholder={$i18n.t('Search')}
 					on:focus={() => {
@@ -298,7 +300,9 @@
 	</div>
 </div>

-<div class="scrollbar-hidden relative whitespace-nowrap overflow-x-auto max-w-full rounded pt-0.5">
+<div
+	class="scrollbar-hidden relative whitespace-nowrap overflow-x-auto max-w-full rounded-sm pt-0.5"
+>
 	{#if loadingLeaderboard}
 		<div class=" absolute top-0 bottom-0 left-0 right-0 flex">
 			<div class="m-auto">
@@ -347,7 +351,7 @@
 						</td>
 						<td class="px-3 py-1.5 flex flex-col justify-center">
 							<div class="flex items-center gap-2">
-								<div class="flex-shrink-0">
+								<div class="shrink-0">
 									<img
 										src={model?.info?.meta?.profile_image_url ?? '/favicon.png'}
 										alt={model.name}
@@ -3,7 +3,7 @@
 	import fileSaver from 'file-saver';
 	const { saveAs } = fileSaver;

-	import { WEBUI_NAME, config, functions, models } from '$lib/stores';
+	import { WEBUI_NAME, config, functions, models, settings } from '$lib/stores';
 	import { onMount, getContext, tick } from 'svelte';

 	import { goto } from '$app/navigation';
@@ -65,7 +65,7 @@
 			return null;
 		});

-		toast.success($i18n.t('Redirecting you to OpenWebUI Community'));
+		toast.success($i18n.t('Redirecting you to Open WebUI Community'));

 		const url = 'https://openwebui.com';

@@ -126,7 +126,12 @@
 			toast.success($i18n.t('Function deleted successfully'));

 			functions.set(await getFunctions(localStorage.token));
-			models.set(await getModels(localStorage.token));
+			models.set(
+				await getModels(
+					localStorage.token,
+					$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+				)
+			);
 		}
 	};

@@ -147,7 +152,12 @@
 			}

 			functions.set(await getFunctions(localStorage.token));
-			models.set(await getModels(localStorage.token));
+			models.set(
+				await getModels(
+					localStorage.token,
+					$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+				)
+			);
 		}
 	};

@@ -170,12 +180,12 @@

 		window.addEventListener('keydown', onKeyDown);
 		window.addEventListener('keyup', onKeyUp);
-		window.addEventListener('blur', onBlur);
+		window.addEventListener('blur-sm', onBlur);

 		return () => {
 			window.removeEventListener('keydown', onKeyDown);
 			window.removeEventListener('keyup', onKeyUp);
-			window.removeEventListener('blur', onBlur);
+			window.removeEventListener('blur-sm', onBlur);
 		};
 	});
 </script>
@@ -201,7 +211,7 @@
 				<Search className="size-3.5" />
 			</div>
 			<input
-				class=" w-full text-sm pr-4 py-1 rounded-r-xl outline-none bg-transparent"
+				class=" w-full text-sm pr-4 py-1 rounded-r-xl outline-hidden bg-transparent"
 				bind:value={query}
 				placeholder={$i18n.t('Search Functions')}
 			/>
@@ -231,14 +241,14 @@
 					<div class=" flex-1 self-center pl-1">
 						<div class=" font-semibold flex items-center gap-1.5">
 							<div
-								class=" text-xs font-bold px-1 rounded uppercase line-clamp-1 bg-gray-500/20 text-gray-700 dark:text-gray-200"
+								class=" text-xs font-bold px-1 rounded-sm uppercase line-clamp-1 bg-gray-500/20 text-gray-700 dark:text-gray-200"
 							>
 								{func.type}
 							</div>

 							{#if func?.meta?.manifest?.version}
 								<div
-									class="text-xs font-bold px-1 rounded line-clamp-1 bg-gray-500/20 text-gray-700 dark:text-gray-200"
+									class="text-xs font-bold px-1 rounded-sm line-clamp-1 bg-gray-500/20 text-gray-700 dark:text-gray-200"
 								>
 									v{func?.meta?.manifest?.version ?? ''}
 								</div>
@@ -250,7 +260,7 @@
 						</div>

 						<div class="flex gap-1.5 px-1">
-							<div class=" text-gray-500 text-xs font-medium flex-shrink-0">{func.id}</div>
+							<div class=" text-gray-500 text-xs font-medium shrink-0">{func.id}</div>

 							<div class=" text-xs overflow-hidden text-ellipsis line-clamp-1">
 								{func.meta.description}
@@ -359,7 +369,13 @@
 							bind:state={func.is_active}
 							on:change={async (e) => {
 								toggleFunctionById(localStorage.token, func.id);
-								models.set(await getModels(localStorage.token));
+								models.set(
+									await getModels(
+										localStorage.token,
+										$config?.features?.enable_direct_connections &&
+											($settings?.directConnections ?? null)
+									)
+								);
 							}}
 						/>
 					</Tooltip>
@@ -453,7 +469,7 @@
 {#if $config?.features.enable_community_sharing}
 	<div class=" my-16">
 		<div class=" text-xl font-medium mb-1 line-clamp-1">
-			{$i18n.t('Made by OpenWebUI Community')}
+			{$i18n.t('Made by Open WebUI Community')}
 		</div>

 		<a
@@ -496,7 +512,12 @@
 	id={selectedFunction?.id ?? null}
 	on:save={async () => {
 		await tick();
-		models.set(await getModels(localStorage.token));
+		models.set(
+			await getModels(
+				localStorage.token,
+				$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+			)
+		);
 	}}
 />

@@ -517,7 +538,12 @@

 			toast.success($i18n.t('Functions imported successfully'));
 			functions.set(await getFunctions(localStorage.token));
-			models.set(await getModels(localStorage.token));
+			models.set(
+				await getModels(
+					localStorage.token,
+					$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+				)
+			);
 		};

 		reader.readAsText(importFiles[0]);
@@ -300,7 +300,7 @@ class Pipe:
 			<div class="flex flex-col flex-1 overflow-auto h-0 rounded-lg">
 				<div class="w-full mb-2 flex flex-col gap-0.5">
 					<div class="flex w-full items-center">
-						<div class=" flex-shrink-0 mr-2">
+						<div class=" shrink-0 mr-2">
 							<Tooltip content={$i18n.t('Back')}>
 								<button
 									class="w-full text-left text-sm py-1.5 px-1 rounded-lg dark:text-gray-300 dark:hover:text-white hover:bg-black/5 dark:hover:bg-gray-850"
@@ -317,7 +317,7 @@ class Pipe:
 						<div class="flex-1">
 							<Tooltip content={$i18n.t('e.g. My Filter')} placement="top-start">
 								<input
-									class="w-full text-2xl font-medium bg-transparent outline-none font-primary"
+									class="w-full text-2xl font-medium bg-transparent outline-hidden font-primary"
 									type="text"
 									placeholder={$i18n.t('Function Name')}
 									bind:value={name}
@@ -333,13 +333,13 @@ class Pipe:

 					<div class=" flex gap-2 px-1 items-center">
 						{#if edit}
-							<div class="text-sm text-gray-500 flex-shrink-0">
+							<div class="text-sm text-gray-500 shrink-0">
 								{id}
 							</div>
 						{:else}
 							<Tooltip className="w-full" content={$i18n.t('e.g. my_filter')} placement="top-start">
 								<input
-									class="w-full text-sm disabled:text-gray-500 bg-transparent outline-none"
+									class="w-full text-sm disabled:text-gray-500 bg-transparent outline-hidden"
 									type="text"
 									placeholder={$i18n.t('Function ID')}
 									bind:value={id}
@@ -355,7 +355,7 @@ class Pipe:
 							placement="top-start"
 						>
 							<input
-								class="w-full text-sm bg-transparent outline-none"
+								class="w-full text-sm bg-transparent outline-hidden"
 								type="text"
 								placeholder={$i18n.t('Function Description')}
 								bind:value={meta.description}
@@ -42,7 +42,7 @@

 	<div slot="content">
 		<DropdownMenu.Content
-			class="w-full max-w-[180px] 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"
+			class="w-full max-w-[180px] 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"
@@ -63,7 +63,7 @@
 					</div>
 				</div>

-				<hr class="border-gray-100 dark:border-gray-800 my-1" />
+				<hr class="border-gray-100 dark:border-gray-850 my-1" />
 			{/if}

 			<DropdownMenu.Item
@@ -122,7 +122,7 @@
 				<div class="flex items-center">{$i18n.t('Export')}</div>
 			</DropdownMenu.Item>

-			<hr class="border-gray-100 dark:border-gray-800 my-1" />
+			<hr class="border-gray-100 dark:border-gray-850 my-1" />

 			<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"
@@ -19,6 +19,7 @@
 	import ChartBar from '../icons/ChartBar.svelte';
 	import DocumentChartBar from '../icons/DocumentChartBar.svelte';
 	import Evaluations from './Settings/Evaluations.svelte';
+	import CodeExecution from './Settings/CodeExecution.svelte';

 	const i18n = getContext('i18n');

@@ -188,6 +189,32 @@
 			<div class=" self-center">{$i18n.t('Web Search')}</div>
 		</button>

+		<button
+			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
+			'code-execution'
+				? ''
+				: ' text-gray-300 dark:text-gray-600 hover:text-gray-700 dark:hover:text-white'}"
+			on:click={() => {
+				selectedTab = 'code-execution';
+			}}
+		>
+			<div class=" self-center mr-2">
+				<svg
+					xmlns="http://www.w3.org/2000/svg"
+					viewBox="0 0 16 16"
+					fill="currentColor"
+					class="size-4"
+				>
+					<path
+						fill-rule="evenodd"
+						d="M2 4a2 2 0 0 1 2-2h8a2 2 0 0 1 2 2v8a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2V4Zm2.22 1.97a.75.75 0 0 0 0 1.06l.97.97-.97.97a.75.75 0 1 0 1.06 1.06l1.5-1.5a.75.75 0 0 0 0-1.06l-1.5-1.5a.75.75 0 0 0-1.06 0ZM8.75 8.5a.75.75 0 0 0 0 1.5h2.5a.75.75 0 0 0 0-1.5h-2.5Z"
+						clip-rule="evenodd"
+					/>
+				</svg>
+			</div>
+			<div class=" self-center">{$i18n.t('Code Execution')}</div>
+		</button>
+
 		<button
 			class="px-0.5 py-1 min-w-fit rounded-lg flex-1 md:flex-none flex text-right transition {selectedTab ===
 			'interface'
@@ -364,6 +391,15 @@
 					await config.set(await getBackendConfig());
 				}}
 			/>
+		{:else if selectedTab === 'code-execution'}
+			<CodeExecution
+				saveHandler={async () => {
+					toast.success($i18n.t('Settings saved successfully!'));
+
+					await tick();
+					await config.set(await getBackendConfig());
+				}}
+			/>
 		{:else if selectedTab === 'interface'}
 			<Interface
 				on:save={() => {
@@ -10,7 +10,7 @@
 		getModels as _getModels,
 		getVoices as _getVoices
 	} from '$lib/apis/audio';
-	import { config } from '$lib/stores';
+	import { config, settings } from '$lib/stores';

 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';

@@ -39,6 +39,7 @@
 	let STT_ENGINE = '';
 	let STT_MODEL = '';
 	let STT_WHISPER_MODEL = '';
+	let STT_DEEPGRAM_API_KEY = '';

 	let STT_WHISPER_MODEL_LOADING = false;

@@ -50,8 +51,11 @@
 		if (TTS_ENGINE === '') {
 			models = [];
 		} else {
-			const res = await _getModels(localStorage.token).catch((e) => {
-				toast.error(e);
+			const res = await _getModels(
+				localStorage.token,
+				$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+			).catch((e) => {
+				toast.error(`${e}`);
 			});

 			if (res) {
@@ -74,7 +78,7 @@
 			}, 100);
 		} else {
 			const res = await _getVoices(localStorage.token).catch((e) => {
-				toast.error(e);
+				toast.error(`${e}`);
 			});

 			if (res) {
@@ -103,7 +107,8 @@
 				OPENAI_API_KEY: STT_OPENAI_API_KEY,
 				ENGINE: STT_ENGINE,
 				MODEL: STT_MODEL,
-				WHISPER_MODEL: STT_WHISPER_MODEL
+				WHISPER_MODEL: STT_WHISPER_MODEL,
+				DEEPGRAM_API_KEY: STT_DEEPGRAM_API_KEY
 			}
 		});

@@ -143,6 +148,7 @@
 			STT_ENGINE = res.stt.ENGINE;
 			STT_MODEL = res.stt.MODEL;
 			STT_WHISPER_MODEL = res.stt.WHISPER_MODEL;
+			STT_DEEPGRAM_API_KEY = res.stt.DEEPGRAM_API_KEY;
 		}

 		await getVoices();
@@ -166,13 +172,14 @@
 					<div class=" self-center text-xs font-medium">{$i18n.t('Speech-to-Text Engine')}</div>
 					<div class="flex items-center relative">
 						<select
-							class="dark:bg-gray-900 cursor-pointer w-fit pr-8 rounded px-2 p-1 text-xs bg-transparent outline-none text-right"
+							class="dark:bg-gray-900 cursor-pointer w-fit pr-8 rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
 							bind:value={STT_ENGINE}
 							placeholder="Select an engine"
 						>
 							<option value="">{$i18n.t('Whisper (Local)')}</option>
 							<option value="openai">OpenAI</option>
 							<option value="web">{$i18n.t('Web API')}</option>
+							<option value="deepgram">Deepgram</option>
 						</select>
 					</div>
 				</div>
@@ -181,7 +188,7 @@
 					<div>
 						<div class="mt-1 flex gap-2 mb-1">
 							<input
-								class="flex-1 w-full bg-transparent outline-none"
+								class="flex-1 w-full bg-transparent outline-hidden"
 								placeholder={$i18n.t('API Base URL')}
 								bind:value={STT_OPENAI_API_BASE_URL}
 								required
@@ -191,7 +198,7 @@
 						</div>
 					</div>

-					<hr class=" dark:border-gray-850 my-2" />
+					<hr class="border-gray-100 dark:border-gray-850 my-2" />

 					<div>
 						<div class=" mb-1.5 text-sm font-medium">{$i18n.t('STT Model')}</div>
@@ -199,7 +206,7 @@
 							<div class="flex-1">
 								<input
 									list="model-list"
-									class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+									class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 									bind:value={STT_MODEL}
 									placeholder="Select a model"
 								/>
@@ -210,6 +217,37 @@
 							</div>
 						</div>
 					</div>
+				{:else if STT_ENGINE === 'deepgram'}
+					<div>
+						<div class="mt-1 flex gap-2 mb-1">
+							<SensitiveInput placeholder={$i18n.t('API Key')} bind:value={STT_DEEPGRAM_API_KEY} />
+						</div>
+					</div>
+
+					<hr class="border-gray-100 dark:border-gray-850 my-2" />
+
+					<div>
+						<div class=" mb-1.5 text-sm font-medium">{$i18n.t('STT Model')}</div>
+						<div class="flex w-full">
+							<div class="flex-1">
+								<input
+									class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
+									bind:value={STT_MODEL}
+									placeholder="Select a model (optional)"
+								/>
+							</div>
+						</div>
+						<div class="mt-2 mb-1 text-xs text-gray-400 dark:text-gray-500">
+							{$i18n.t('Leave model field empty to use the default model.')}
+							<a
+								class=" hover:underline dark:text-gray-200 text-gray-800"
+								href="https://developers.deepgram.com/docs/models"
+								target="_blank"
+							>
+								{$i18n.t('Click here to see available models.')}
+							</a>
+						</div>
+					</div>
 				{:else if STT_ENGINE === ''}
 					<div>
 						<div class=" mb-1.5 text-sm font-medium">{$i18n.t('STT Model')}</div>
@@ -217,7 +255,7 @@
 						<div class="flex w-full">
 							<div class="flex-1 mr-2">
 								<input
-									class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+									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('Set whisper model')}
 									bind:value={STT_WHISPER_MODEL}
 								/>
@@ -295,7 +333,7 @@
 				{/if}
 			</div>

-			<hr class=" dark:border-gray-800" />
+			<hr class="border-gray-100 dark:border-gray-850" />

 			<div>
 				<div class=" mb-1 text-sm font-medium">{$i18n.t('TTS Settings')}</div>
@@ -304,7 +342,7 @@
 					<div class=" self-center text-xs font-medium">{$i18n.t('Text-to-Speech Engine')}</div>
 					<div class="flex items-center relative">
 						<select
-							class=" dark:bg-gray-900 w-fit pr-8 cursor-pointer rounded px-2 p-1 text-xs bg-transparent outline-none text-right"
+							class=" dark:bg-gray-900 w-fit pr-8 cursor-pointer rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
 							bind:value={TTS_ENGINE}
 							placeholder="Select a mode"
 							on:change={async (e) => {
@@ -334,7 +372,7 @@
 					<div>
 						<div class="mt-1 flex gap-2 mb-1">
 							<input
-								class="flex-1 w-full bg-transparent outline-none"
+								class="flex-1 w-full bg-transparent outline-hidden"
 								placeholder={$i18n.t('API Base URL')}
 								bind:value={TTS_OPENAI_API_BASE_URL}
 								required
@@ -347,7 +385,7 @@
 					<div>
 						<div class="mt-1 flex gap-2 mb-1">
 							<input
-								class="flex-1 w-full rounded-lg py-2 pl-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+								class="flex-1 w-full rounded-lg py-2 pl-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 								placeholder={$i18n.t('API Key')}
 								bind:value={TTS_API_KEY}
 								required
@@ -358,13 +396,13 @@
 					<div>
 						<div class="mt-1 flex gap-2 mb-1">
 							<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-none"
+								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={$i18n.t('API Key')}
 								bind:value={TTS_API_KEY}
 								required
 							/>
 							<input
-								class="flex-1 w-full rounded-lg py-2 pl-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+								class="flex-1 w-full rounded-lg py-2 pl-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 								placeholder={$i18n.t('Azure Region')}
 								bind:value={TTS_AZURE_SPEECH_REGION}
 								required
@@ -373,7 +411,7 @@
 					</div>
 				{/if}

-				<hr class=" dark:border-gray-850 my-2" />
+				<hr class="border-gray-100 dark:border-gray-850 my-2" />

 				{#if TTS_ENGINE === ''}
 					<div>
@@ -381,7 +419,7 @@
 						<div class="flex w-full">
 							<div class="flex-1">
 								<select
-									class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+									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={TTS_VOICE}
 								>
 									<option value="" selected={TTS_VOICE !== ''}>{$i18n.t('Default')}</option>
@@ -404,7 +442,7 @@
 							<div class="flex-1">
 								<input
 									list="model-list"
-									class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+									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={TTS_MODEL}
 									placeholder="CMU ARCTIC speaker embedding name"
 								/>
@@ -446,7 +484,7 @@
 								<div class="flex-1">
 									<input
 										list="voice-list"
-										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+										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={TTS_VOICE}
 										placeholder="Select a voice"
 									/>
@@ -465,7 +503,7 @@
 								<div class="flex-1">
 									<input
 										list="tts-model-list"
-										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+										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={TTS_MODEL}
 										placeholder="Select a model"
 									/>
@@ -487,7 +525,7 @@
 								<div class="flex-1">
 									<input
 										list="voice-list"
-										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+										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={TTS_VOICE}
 										placeholder="Select a voice"
 									/>
@@ -506,7 +544,7 @@
 								<div class="flex-1">
 									<input
 										list="tts-model-list"
-										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+										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={TTS_MODEL}
 										placeholder="Select a model"
 									/>
@@ -528,7 +566,7 @@
 								<div class="flex-1">
 									<input
 										list="voice-list"
-										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+										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={TTS_VOICE}
 										placeholder="Select a voice"
 									/>
@@ -555,7 +593,7 @@
 								<div class="flex-1">
 									<input
 										list="tts-model-list"
-										class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+										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={TTS_AZURE_SPEECH_OUTPUT_FORMAT}
 										placeholder="Select a output format"
 									/>
@@ -565,13 +603,13 @@
 					</div>
 				{/if}

-				<hr class="dark:border-gray-850 my-2" />
+				<hr class="border-gray-100 dark:border-gray-850 my-2" />

 				<div class="pt-0.5 flex w-full justify-between">
 					<div class="self-center text-xs font-medium">{$i18n.t('Response splitting')}</div>
 					<div class="flex items-center relative">
 						<select
-							class="dark:bg-gray-900 w-fit pr-8 cursor-pointer rounded px-2 p-1 text-xs bg-transparent outline-none text-right"
+							class="dark:bg-gray-900 w-fit pr-8 cursor-pointer rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
 							aria-label="Select how to split message text for TTS requests"
 							bind:value={TTS_SPLIT_ON}
 						>
@@ -0,0 +1,317 @@
+<script lang="ts">
+	import { toast } from 'svelte-sonner';
+	import { onMount, getContext } from 'svelte';
+	import { getCodeExecutionConfig, setCodeExecutionConfig } from '$lib/apis/configs';
+
+	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
+
+	import Tooltip from '$lib/components/common/Tooltip.svelte';
+	import Textarea from '$lib/components/common/Textarea.svelte';
+	import Switch from '$lib/components/common/Switch.svelte';
+
+	const i18n = getContext('i18n');
+
+	export let saveHandler: Function;
+
+	let config = null;
+
+	let engines = ['pyodide', 'jupyter'];
+
+	const submitHandler = async () => {
+		const res = await setCodeExecutionConfig(localStorage.token, config);
+	};
+
+	onMount(async () => {
+		const res = await getCodeExecutionConfig(localStorage.token);
+
+		if (res) {
+			config = res;
+		}
+	});
+</script>
+
+<form
+	class="flex flex-col h-full justify-between space-y-3 text-sm"
+	on:submit|preventDefault={async () => {
+		await submitHandler();
+		saveHandler();
+	}}
+>
+	<div class=" space-y-3 overflow-y-scroll scrollbar-hidden h-full">
+		{#if config}
+			<div>
+				<div class="mb-3.5">
+					<div class=" mb-2.5 text-base font-medium">{$i18n.t('General')}</div>
+
+					<hr class=" border-gray-100 dark:border-gray-850 my-2" />
+
+					<div class="mb-2.5">
+						<div class="flex w-full justify-between">
+							<div class=" self-center text-xs font-medium">{$i18n.t('Code Execution Engine')}</div>
+							<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={config.CODE_EXECUTION_ENGINE}
+									placeholder={$i18n.t('Select a engine')}
+									required
+								>
+									<option disabled selected value="">{$i18n.t('Select a engine')}</option>
+									{#each engines as engine}
+										<option value={engine}>{engine}</option>
+									{/each}
+								</select>
+							</div>
+						</div>
+
+						{#if config.CODE_EXECUTION_ENGINE === 'jupyter'}
+							<div class="text-gray-500 text-xs">
+								{$i18n.t(
+									'Warning: Jupyter execution enables arbitrary code execution, posing severe security risks—proceed with extreme caution.'
+								)}
+							</div>
+						{/if}
+					</div>
+
+					{#if config.CODE_EXECUTION_ENGINE === 'jupyter'}
+						<div class="mb-2.5 flex flex-col gap-1.5 w-full">
+							<div class="text-xs font-medium">
+								{$i18n.t('Jupyter URL')}
+							</div>
+
+							<div class="flex w-full">
+								<div class="flex-1">
+									<input
+										class="w-full text-sm py-0.5 placeholder:text-gray-300 dark:placeholder:text-gray-700 bg-transparent outline-hidden"
+										type="text"
+										placeholder={$i18n.t('Enter Jupyter URL')}
+										bind:value={config.CODE_EXECUTION_JUPYTER_URL}
+										autocomplete="off"
+									/>
+								</div>
+							</div>
+						</div>
+
+						<div class="mb-2.5 flex flex-col gap-1.5 w-full">
+							<div class=" flex gap-2 w-full items-center justify-between">
+								<div class="text-xs font-medium">
+									{$i18n.t('Jupyter Auth')}
+								</div>
+
+								<div>
+									<select
+										class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-left"
+										bind:value={config.CODE_EXECUTION_JUPYTER_AUTH}
+										placeholder={$i18n.t('Select an auth method')}
+									>
+										<option selected value="">{$i18n.t('None')}</option>
+										<option value="token">{$i18n.t('Token')}</option>
+										<option value="password">{$i18n.t('Password')}</option>
+									</select>
+								</div>
+							</div>
+
+							{#if config.CODE_EXECUTION_JUPYTER_AUTH}
+								<div class="flex w-full gap-2">
+									<div class="flex-1">
+										{#if config.CODE_EXECUTION_JUPYTER_AUTH === 'password'}
+											<SensitiveInput
+												type="text"
+												placeholder={$i18n.t('Enter Jupyter Password')}
+												bind:value={config.CODE_EXECUTION_JUPYTER_AUTH_PASSWORD}
+												autocomplete="off"
+											/>
+										{:else}
+											<SensitiveInput
+												type="text"
+												placeholder={$i18n.t('Enter Jupyter Token')}
+												bind:value={config.CODE_EXECUTION_JUPYTER_AUTH_TOKEN}
+												autocomplete="off"
+											/>
+										{/if}
+									</div>
+								</div>
+							{/if}
+						</div>
+
+						<div class="flex gap-2 w-full items-center justify-between">
+							<div class="text-xs font-medium">
+								{$i18n.t('Code Execution Timeout')}
+							</div>
+
+							<div class="">
+								<Tooltip content={$i18n.t('Enter timeout in seconds')}>
+									<input
+										class="dark:bg-gray-900 w-fit rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
+										type="number"
+										bind:value={config.CODE_EXECUTION_JUPYTER_TIMEOUT}
+										placeholder={$i18n.t('e.g. 60')}
+										autocomplete="off"
+									/>
+								</Tooltip>
+							</div>
+						</div>
+					{/if}
+				</div>
+
+				<div class="mb-3.5">
+					<div class=" mb-2.5 text-base font-medium">{$i18n.t('Code Interpreter')}</div>
+
+					<hr class=" border-gray-100 dark:border-gray-850 my-2" />
+
+					<div class="mb-2.5">
+						<div class=" flex w-full justify-between">
+							<div class=" self-center text-xs font-medium">
+								{$i18n.t('Enable Code Interpreter')}
+							</div>
+
+							<Switch bind:state={config.ENABLE_CODE_INTERPRETER} />
+						</div>
+					</div>
+
+					{#if config.ENABLE_CODE_INTERPRETER}
+						<div class="mb-2.5">
+							<div class="  flex w-full justify-between">
+								<div class=" self-center text-xs font-medium">
+									{$i18n.t('Code Interpreter Engine')}
+								</div>
+								<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={config.CODE_INTERPRETER_ENGINE}
+										placeholder={$i18n.t('Select a engine')}
+										required
+									>
+										<option disabled selected value="">{$i18n.t('Select a engine')}</option>
+										{#each engines as engine}
+											<option value={engine}>{engine}</option>
+										{/each}
+									</select>
+								</div>
+							</div>
+
+							{#if config.CODE_INTERPRETER_ENGINE === 'jupyter'}
+								<div class="text-gray-500 text-xs">
+									{$i18n.t(
+										'Warning: Jupyter execution enables arbitrary code execution, posing severe security risks—proceed with extreme caution.'
+									)}
+								</div>
+							{/if}
+						</div>
+
+						{#if config.CODE_INTERPRETER_ENGINE === 'jupyter'}
+							<div class="mb-2.5 flex flex-col gap-1.5 w-full">
+								<div class="text-xs font-medium">
+									{$i18n.t('Jupyter URL')}
+								</div>
+
+								<div class="flex w-full">
+									<div class="flex-1">
+										<input
+											class="w-full text-sm py-0.5 placeholder:text-gray-300 dark:placeholder:text-gray-700 bg-transparent outline-hidden"
+											type="text"
+											placeholder={$i18n.t('Enter Jupyter URL')}
+											bind:value={config.CODE_INTERPRETER_JUPYTER_URL}
+											autocomplete="off"
+										/>
+									</div>
+								</div>
+							</div>
+
+							<div class="mb-2.5 flex flex-col gap-1.5 w-full">
+								<div class="flex gap-2 w-full items-center justify-between">
+									<div class="text-xs font-medium">
+										{$i18n.t('Jupyter Auth')}
+									</div>
+
+									<div>
+										<select
+											class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-left"
+											bind:value={config.CODE_INTERPRETER_JUPYTER_AUTH}
+											placeholder={$i18n.t('Select an auth method')}
+										>
+											<option selected value="">{$i18n.t('None')}</option>
+											<option value="token">{$i18n.t('Token')}</option>
+											<option value="password">{$i18n.t('Password')}</option>
+										</select>
+									</div>
+								</div>
+
+								{#if config.CODE_INTERPRETER_JUPYTER_AUTH}
+									<div class="flex w-full gap-2">
+										<div class="flex-1">
+											{#if config.CODE_INTERPRETER_JUPYTER_AUTH === 'password'}
+												<SensitiveInput
+													type="text"
+													placeholder={$i18n.t('Enter Jupyter Password')}
+													bind:value={config.CODE_INTERPRETER_JUPYTER_AUTH_PASSWORD}
+													autocomplete="off"
+												/>
+											{:else}
+												<SensitiveInput
+													type="text"
+													placeholder={$i18n.t('Enter Jupyter Token')}
+													bind:value={config.CODE_INTERPRETER_JUPYTER_AUTH_TOKEN}
+													autocomplete="off"
+												/>
+											{/if}
+										</div>
+									</div>
+								{/if}
+							</div>
+
+							<div class="flex gap-2 w-full items-center justify-between">
+								<div class="text-xs font-medium">
+									{$i18n.t('Code Execution Timeout')}
+								</div>
+
+								<div class="">
+									<Tooltip content={$i18n.t('Enter timeout in seconds')}>
+										<input
+											class="dark:bg-gray-900 w-fit rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
+											type="number"
+											bind:value={config.CODE_INTERPRETER_JUPYTER_TIMEOUT}
+											placeholder={$i18n.t('e.g. 60')}
+											autocomplete="off"
+										/>
+									</Tooltip>
+								</div>
+							</div>
+						{/if}
+
+						<hr class="border-gray-100 dark:border-gray-850 my-2" />
+
+						<div>
+							<div class="py-0.5 w-full">
+								<div class=" mb-2.5 text-xs font-medium">
+									{$i18n.t('Code Interpreter Prompt Template')}
+								</div>
+
+								<Tooltip
+									content={$i18n.t(
+										'Leave empty to use the default prompt, or enter a custom prompt'
+									)}
+									placement="top-start"
+								>
+									<Textarea
+										bind:value={config.CODE_INTERPRETER_PROMPT_TEMPLATE}
+										placeholder={$i18n.t(
+											'Leave empty to use the default prompt, or enter a custom prompt'
+										)}
+									/>
+								</Tooltip>
+							</div>
+						</div>
+					{/if}
+				</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"
+		>
+			{$i18n.t('Save')}
+		</button>
+	</div>
+</form>
@@ -7,8 +7,9 @@
 	import { getOllamaConfig, updateOllamaConfig } from '$lib/apis/ollama';
 	import { getOpenAIConfig, updateOpenAIConfig, getOpenAIModels } from '$lib/apis/openai';
 	import { getModels as _getModels } from '$lib/apis';
+	import { getDirectConnectionsConfig, setDirectConnectionsConfig } from '$lib/apis/configs';

-	import { models, user } from '$lib/stores';
+	import { config, models, settings, user } from '$lib/stores';

 	import Switch from '$lib/components/common/Switch.svelte';
 	import Spinner from '$lib/components/common/Spinner.svelte';
@@ -16,13 +17,16 @@
 	import Plus from '$lib/components/icons/Plus.svelte';

 	import OpenAIConnection from './Connections/OpenAIConnection.svelte';
-	import AddConnectionModal from './Connections/AddConnectionModal.svelte';
+	import AddConnectionModal from '$lib/components/AddConnectionModal.svelte';
 	import OllamaConnection from './Connections/OllamaConnection.svelte';

 	const i18n = getContext('i18n');

 	const getModels = async () => {
-		const models = await _getModels(localStorage.token);
+		const models = await _getModels(
+			localStorage.token,
+			$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+		);
 		return models;
 	};

@@ -37,6 +41,8 @@
 	let ENABLE_OPENAI_API: null | boolean = null;
 	let ENABLE_OLLAMA_API: null | boolean = null;

+	let directConnectionsConfig = null;
+
 	let pipelineUrls = {};
 	let showAddOpenAIConnectionModal = false;
 	let showAddOllamaConnectionModal = false;
@@ -98,17 +104,33 @@
 		}
 	};

+	const updateDirectConnectionsHandler = async () => {
+		const res = await setDirectConnectionsConfig(localStorage.token, directConnectionsConfig).catch(
+			(error) => {
+				toast.error(`${error}`);
+			}
+		);
+
+		if (res) {
+			toast.success($i18n.t('Direct Connections settings updated'));
+			await models.set(await getModels());
+		}
+	};
+
 	const addOpenAIConnectionHandler = async (connection) => {
 		OPENAI_API_BASE_URLS = [...OPENAI_API_BASE_URLS, connection.url];
 		OPENAI_API_KEYS = [...OPENAI_API_KEYS, connection.key];
-		OPENAI_API_CONFIGS[OPENAI_API_BASE_URLS.length] = connection.config;
+		OPENAI_API_CONFIGS[OPENAI_API_BASE_URLS.length - 1] = connection.config;

 		await updateOpenAIHandler();
 	};

 	const addOllamaConnectionHandler = async (connection) => {
 		OLLAMA_BASE_URLS = [...OLLAMA_BASE_URLS, connection.url];
-		OLLAMA_API_CONFIGS[OLLAMA_BASE_URLS.length] = connection.config;
+		OLLAMA_API_CONFIGS[OLLAMA_BASE_URLS.length - 1] = {
+			...connection.config,
+			key: connection.key
+		};

 		await updateOllamaHandler();
 	};
@@ -124,6 +146,9 @@
 				})(),
 				(async () => {
 					openaiConfig = await getOpenAIConfig(localStorage.token);
+				})(),
+				(async () => {
+					directConnectionsConfig = await getDirectConnectionsConfig(localStorage.token);
 				})()
 			]);

@@ -167,6 +192,14 @@
 			}
 		}
 	});
+
+	const submitHandler = async () => {
+		updateOpenAIHandler();
+		updateOllamaHandler();
+		updateDirectConnectionsHandler();
+
+		dispatch('save');
+	};
 </script>

 <AddConnectionModal
@@ -180,17 +213,9 @@
 	onSubmit={addOllamaConnectionHandler}
 />

-<form
-	class="flex flex-col h-full justify-between text-sm"
-	on:submit|preventDefault={() => {
-		updateOpenAIHandler();
-		updateOllamaHandler();
-
-		dispatch('save');
-	}}
->
+<form class="flex flex-col h-full justify-between text-sm" on:submit|preventDefault={submitHandler}>
 	<div class=" overflow-y-scroll scrollbar-hidden h-full">
-		{#if ENABLE_OPENAI_API !== null && ENABLE_OLLAMA_API !== null}
+		{#if ENABLE_OPENAI_API !== null && ENABLE_OLLAMA_API !== null && directConnectionsConfig !== null}
 			<div class="my-2">
 				<div class="mt-2 space-y-2 pr-1.5">
 					<div class="flex justify-between items-center text-sm">
@@ -209,7 +234,7 @@
 					</div>

 					{#if ENABLE_OPENAI_API}
-						<hr class=" border-gray-50 dark:border-gray-850" />
+						<hr class=" border-gray-100 dark:border-gray-850" />

 						<div class="">
 							<div class="flex justify-between items-center">
@@ -244,7 +269,11 @@
 											);
 											OPENAI_API_KEYS = OPENAI_API_KEYS.filter((key, keyIdx) => idx !== keyIdx);

-											delete OPENAI_API_CONFIGS[idx];
+											let newConfig = {};
+											OPENAI_API_BASE_URLS.forEach((url, newIdx) => {
+												newConfig[newIdx] = OPENAI_API_CONFIGS[newIdx < idx ? newIdx : newIdx + 1];
+											});
+											OPENAI_API_CONFIGS = newConfig;
 										}}
 									/>
 								{/each}
@@ -254,7 +283,7 @@
 				</div>
 			</div>

-			<hr class=" border-gray-50 dark:border-gray-850" />
+			<hr class=" border-gray-100 dark:border-gray-850" />

 			<div class="pr-1.5 my-2">
 				<div class="flex justify-between items-center text-sm mb-2">
@@ -271,7 +300,7 @@
 				</div>

 				{#if ENABLE_OLLAMA_API}
-					<hr class=" border-gray-50 dark:border-gray-850 my-2" />
+					<hr class=" border-gray-100 dark:border-gray-850 my-2" />

 					<div class="">
 						<div class="flex justify-between items-center">
@@ -302,7 +331,12 @@
 										}}
 										onDelete={() => {
 											OLLAMA_BASE_URLS = OLLAMA_BASE_URLS.filter((url, urlIdx) => idx !== urlIdx);
-											delete OLLAMA_API_CONFIGS[idx];
+
+											let newConfig = {};
+											OLLAMA_BASE_URLS.forEach((url, newIdx) => {
+												newConfig[newIdx] = OLLAMA_API_CONFIGS[newIdx < idx ? newIdx : newIdx + 1];
+											});
+											OLLAMA_API_CONFIGS = newConfig;
 										}}
 									/>
 								{/each}
@@ -322,6 +356,33 @@
 					</div>
 				{/if}
 			</div>
+
+			<hr class=" border-gray-100 dark:border-gray-850" />
+
+			<div class="pr-1.5 my-2">
+				<div class="flex justify-between items-center text-sm">
+					<div class="  font-medium">{$i18n.t('Direct Connections')}</div>
+
+					<div class="flex items-center">
+						<div class="">
+							<Switch
+								bind:state={directConnectionsConfig.ENABLE_DIRECT_CONNECTIONS}
+								on:change={async () => {
+									updateDirectConnectionsHandler();
+								}}
+							/>
+						</div>
+					</div>
+				</div>
+
+				<div class="mt-1.5">
+					<div class="text-xs text-gray-500">
+						{$i18n.t(
+							'Direct Connections allow users to connect to their own OpenAI compatible API endpoints.'
+						)}
+					</div>
+				</div>
+			</div>
 		{:else}
 			<div class="flex h-full justify-center">
 				<div class="my-auto">
@@ -16,7 +16,7 @@
 			<div
 				class="flex w-full justify-between items-center text-lg font-medium self-center font-primary"
 			>
-				<div class=" flex-shrink-0">
+				<div class=" shrink-0">
 					{$i18n.t('Manage Ollama')}
 				</div>
 			</div>
@@ -4,7 +4,7 @@

 	import Tooltip from '$lib/components/common/Tooltip.svelte';
 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
-	import AddConnectionModal from './AddConnectionModal.svelte';
+	import AddConnectionModal from '$lib/components/AddConnectionModal.svelte';

 	import Cog6 from '$lib/components/icons/Cog6.svelte';
 	import Wrench from '$lib/components/icons/Wrench.svelte';
@@ -56,7 +56,7 @@
 		{/if}

 		<input
-			class="w-full text-sm bg-transparent outline-none"
+			class="w-full text-sm bg-transparent outline-hidden"
 			placeholder={$i18n.t('Enter URL (e.g. http://localhost:11434)')}
 			bind:value={url}
 		/>
@@ -5,7 +5,8 @@
 	import Tooltip from '$lib/components/common/Tooltip.svelte';
 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
 	import Cog6 from '$lib/components/icons/Cog6.svelte';
-	import AddConnectionModal from './AddConnectionModal.svelte';
+	import AddConnectionModal from '$lib/components/AddConnectionModal.svelte';
+
 	import { connect } from 'socket.io-client';

 	export let onDelete = () => {};
@@ -53,7 +54,7 @@
 		<div class="flex w-full">
 			<div class="flex-1 relative">
 				<input
-					class=" outline-none w-full bg-transparent {pipeline ? 'pr-8' : ''}"
+					class=" outline-hidden w-full bg-transparent {pipeline ? 'pr-8' : ''}"
 					placeholder={$i18n.t('API Base URL')}
 					bind:value={url}
 					autocomplete="off"
@@ -84,7 +85,7 @@
 			</div>

 			<SensitiveInput
-				inputClassName=" outline-none bg-transparent w-full"
+				inputClassName=" outline-hidden bg-transparent w-full"
 				placeholder={$i18n.t('API Key')}
 				bind:value={key}
 			/>
@@ -119,7 +119,7 @@
 				</div>
 			</button>

-			<hr class=" dark:border-gray-850 my-1" />
+			<hr class="border-gray-100 dark:border-gray-850 my-1" />

 			{#if $config?.features.enable_admin_export ?? true}
 				<div class="  flex w-full justify-between">
@@ -27,7 +27,6 @@
 	import SensitiveInput from '$lib/components/common/SensitiveInput.svelte';
 	import Tooltip from '$lib/components/common/Tooltip.svelte';
 	import Switch from '$lib/components/common/Switch.svelte';
-	import { text } from '@sveltejs/kit';
 	import Textarea from '$lib/components/common/Textarea.svelte';

 	const i18n = getContext('i18n');
@@ -56,6 +55,8 @@
 	let chunkOverlap = 0;
 	let pdfExtractImages = true;

+	let RAG_FULL_CONTEXT = false;
+
 	let enableGoogleDriveIntegration = false;

 	let OpenAIUrl = '';
@@ -182,6 +183,7 @@
 				max_size: fileMaxSize === '' ? null : fileMaxSize,
 				max_count: fileMaxCount === '' ? null : fileMaxCount
 			},
+			RAG_FULL_CONTEXT: RAG_FULL_CONTEXT,
 			chunk: {
 				text_splitter: textSplitter,
 				chunk_overlap: chunkOverlap,
@@ -242,6 +244,8 @@
 			chunkSize = res.chunk.chunk_size;
 			chunkOverlap = res.chunk.chunk_overlap;

+			RAG_FULL_CONTEXT = res.RAG_FULL_CONTEXT;
+
 			contentExtractionEngine = res.content_extraction.engine;
 			tikaServerUrl = res.content_extraction.tika_server_url;
 			showTikaServerUrl = contentExtractionEngine === 'tika';
@@ -296,7 +300,7 @@
 				<div class=" self-center text-xs font-medium">{$i18n.t('Embedding Model Engine')}</div>
 				<div class="flex items-center relative">
 					<select
-						class="dark:bg-gray-900 w-fit pr-8 rounded px-2 p-1 text-xs bg-transparent outline-none text-right"
+						class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 p-1 text-xs bg-transparent outline-hidden text-right"
 						bind:value={embeddingEngine}
 						placeholder="Select an embedding model engine"
 						on:change={(e) => {
@@ -319,7 +323,7 @@
 			{#if embeddingEngine === 'openai'}
 				<div class="my-0.5 flex gap-2 pr-2">
 					<input
-						class="flex-1 w-full rounded-lg text-sm bg-transparent outline-none"
+						class="flex-1 w-full rounded-lg text-sm bg-transparent outline-hidden"
 						placeholder={$i18n.t('API Base URL')}
 						bind:value={OpenAIUrl}
 						required
@@ -330,7 +334,7 @@
 			{:else if embeddingEngine === 'ollama'}
 				<div class="my-0.5 flex gap-2 pr-2">
 					<input
-						class="flex-1 w-full rounded-lg text-sm bg-transparent outline-none"
+						class="flex-1 w-full rounded-lg text-sm bg-transparent outline-hidden"
 						placeholder={$i18n.t('API Base URL')}
 						bind:value={OllamaUrl}
 						required
@@ -375,7 +379,7 @@
 				<div class=" self-center text-xs font-medium">{$i18n.t('Hybrid Search')}</div>

 				<button
-					class="p-1 px-3 text-xs flex rounded transition"
+					class="p-1 px-3 text-xs flex rounded-sm transition"
 					on:click={() => {
 						toggleHybridSearch();
 					}}
@@ -388,9 +392,22 @@
 					{/if}
 				</button>
 			</div>
+
+			<div class=" py-0.5 flex w-full justify-between">
+				<div class=" self-center text-xs font-medium">{$i18n.t('Full Context Mode')}</div>
+				<div class="flex items-center relative">
+					<Tooltip
+						content={RAG_FULL_CONTEXT
+							? 'Inject entire contents 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.'}
+					>
+						<Switch bind:state={RAG_FULL_CONTEXT} />
+					</Tooltip>
+				</div>
+			</div>
 		</div>

-		<hr class="dark:border-gray-850" />
+		<hr class="border-gray-100 dark:border-gray-850" />

 		<div class="space-y-2" />
 		<div>
@@ -400,7 +417,7 @@
 				<div class="flex w-full">
 					<div class="flex-1 mr-2">
 						<input
-							class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+							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={embeddingModel}
 							placeholder={$i18n.t('Set embedding model')}
 							required
@@ -411,7 +428,7 @@
 				<div class="flex w-full">
 					<div class="flex-1 mr-2">
 						<input
-							class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+							class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 							placeholder={$i18n.t('Set embedding model (e.g. {{model}})', {
 								model: embeddingModel.slice(-40)
 							})}
@@ -490,7 +507,7 @@
 					<div class="flex w-full">
 						<div class="flex-1 mr-2">
 							<input
-								class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+								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('Set reranking model (e.g. {{model}})', {
 									model: 'BAAI/bge-reranker-v2-m3'
 								})}
@@ -555,7 +572,7 @@
 			{/if}
 		</div>

-		<hr class=" dark:border-gray-850" />
+		<hr class=" border-gray-100 dark:border-gray-850" />

 		<div class="">
 			<div class="text-sm font-medium mb-1">{$i18n.t('Content Extraction')}</div>
@@ -564,7 +581,7 @@
 				<div class="self-center text-xs font-medium">{$i18n.t('Engine')}</div>
 				<div class="flex items-center relative">
 					<select
-						class="dark:bg-gray-900 w-fit pr-8 rounded px-2 text-xs bg-transparent outline-none text-right"
+						class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 text-xs bg-transparent outline-hidden text-right"
 						bind:value={contentExtractionEngine}
 						on:change={(e) => {
 							showTikaServerUrl = e.target.value === 'tika';
@@ -580,7 +597,7 @@
 				<div class="flex w-full mt-1">
 					<div class="flex-1 mr-2">
 						<input
-							class="w-full rounded-lg py-2 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+							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 Tika Server URL')}
 							bind:value={tikaServerUrl}
 						/>
@@ -589,7 +606,7 @@
 			{/if}
 		</div>

-		<hr class=" dark:border-gray-850" />
+		<hr class=" border-gray-100 dark:border-gray-850" />

 		<div class="text-sm font-medium mb-1">{$i18n.t('Google Drive')}</div>

@@ -602,7 +619,7 @@
 			</div>
 		</div>

-		<hr class=" dark:border-gray-850" />
+		<hr class=" border-gray-100 dark:border-gray-850" />

 		<div class=" ">
 			<div class=" text-sm font-medium mb-1">{$i18n.t('Query Params')}</div>
@@ -613,7 +630,7 @@

 					<div class="w-full">
 						<input
-							class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+							class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 							type="number"
 							placeholder={$i18n.t('Enter Top K')}
 							bind:value={querySettings.k}
@@ -631,7 +648,7 @@

 						<div class="w-full">
 							<input
-								class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+								class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 								type="number"
 								step="0.01"
 								placeholder={$i18n.t('Enter Score')}
@@ -667,7 +684,7 @@
 			</div>
 		</div>

-		<hr class=" dark:border-gray-850" />
+		<hr class=" border-gray-100 dark:border-gray-850" />

 		<div class=" ">
 			<div class="mb-1 text-sm font-medium">{$i18n.t('Chunk Params')}</div>
@@ -676,7 +693,7 @@
 				<div class="self-center text-xs font-medium">{$i18n.t('Text Splitter')}</div>
 				<div class="flex items-center relative">
 					<select
-						class="dark:bg-gray-900 w-fit pr-8 rounded px-2 text-xs bg-transparent outline-none text-right"
+						class="dark:bg-gray-900 w-fit pr-8 rounded-sm px-2 text-xs bg-transparent outline-hidden text-right"
 						bind:value={textSplitter}
 					>
 						<option value="">{$i18n.t('Default')} ({$i18n.t('Character')})</option>
@@ -692,7 +709,7 @@
 					</div>
 					<div class="self-center">
 						<input
-							class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+							class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 							type="number"
 							placeholder={$i18n.t('Enter Chunk Size')}
 							bind:value={chunkSize}
@@ -709,7 +726,7 @@

 					<div class="self-center">
 						<input
-							class="w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+							class="w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 							type="number"
 							placeholder={$i18n.t('Enter Chunk Overlap')}
 							bind:value={chunkOverlap}
@@ -731,7 +748,7 @@
 			</div>
 		</div>

-		<hr class=" dark:border-gray-850" />
+		<hr class=" border-gray-100 dark:border-gray-850" />

 		<div class="">
 			<div class="text-sm font-medium mb-1">{$i18n.t('Files')}</div>
@@ -750,7 +767,7 @@
 							placement="top-start"
 						>
 							<input
-								class="w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+								class="w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 								type="number"
 								placeholder={$i18n.t('Leave empty for unlimited')}
 								bind:value={fileMaxSize}
@@ -773,7 +790,7 @@
 							placement="top-start"
 						>
 							<input
-								class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-none"
+								class=" w-full rounded-lg py-1.5 px-4 text-sm bg-gray-50 dark:text-gray-300 dark:bg-gray-850 outline-hidden"
 								type="number"
 								placeholder={$i18n.t('Leave empty for unlimited')}
 								bind:value={fileMaxCount}
@@ -786,7 +803,7 @@
 			</div>
 		</div>

-		<hr class=" dark:border-gray-850" />
+		<hr class=" border-gray-100 dark:border-gray-850" />

 		<div>
 			<button
@@ -1,6 +1,6 @@
 <script lang="ts">
 	import { toast } from 'svelte-sonner';
-	import { models, user } from '$lib/stores';
+	import { models, settings, user, config } from '$lib/stores';
 	import { createEventDispatcher, onMount, getContext, tick } from 'svelte';

 	const dispatch = createEventDispatcher();
@@ -16,49 +16,69 @@

 	const i18n = getContext('i18n');

-	let config = null;
+	let evaluationConfig = null;
 	let showAddModel = false;

 	const submitHandler = async () => {
-		config = await updateConfig(localStorage.token, config).catch((err) => {
+		evaluationConfig = await updateConfig(localStorage.token, evaluationConfig).catch((err) => {
 			toast.error(err);
 			return null;
 		});

-		if (config) {
+		if (evaluationConfig) {
 			toast.success('Settings saved successfully');
-			models.set(await getModels(localStorage.token));
+			models.set(
+				await getModels(
+					localStorage.token,
+					$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+				)
+			);
 		}
 	};

 	const addModelHandler = async (model) => {
-		config.EVALUATION_ARENA_MODELS.push(model);
-		config.EVALUATION_ARENA_MODELS = [...config.EVALUATION_ARENA_MODELS];
+		evaluationConfig.EVALUATION_ARENA_MODELS.push(model);
+		evaluationConfig.EVALUATION_ARENA_MODELS = [...evaluationConfig.EVALUATION_ARENA_MODELS];

 		await submitHandler();
-		models.set(await getModels(localStorage.token));
+		models.set(
+			await getModels(
+				localStorage.token,
+				$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+			)
+		);
 	};

 	const editModelHandler = async (model, modelIdx) => {
-		config.EVALUATION_ARENA_MODELS[modelIdx] = model;
-		config.EVALUATION_ARENA_MODELS = [...config.EVALUATION_ARENA_MODELS];
+		evaluationConfig.EVALUATION_ARENA_MODELS[modelIdx] = model;
+		evaluationConfig.EVALUATION_ARENA_MODELS = [...evaluationConfig.EVALUATION_ARENA_MODELS];

 		await submitHandler();
-		models.set(await getModels(localStorage.token));
+		models.set(
+			await getModels(
+				localStorage.token,
+				$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+			)
+		);
 	};

 	const deleteModelHandler = async (modelIdx) => {
-		config.EVALUATION_ARENA_MODELS = config.EVALUATION_ARENA_MODELS.filter(
+		evaluationConfig.EVALUATION_ARENA_MODELS = evaluationConfig.EVALUATION_ARENA_MODELS.filter(
 			(m, mIdx) => mIdx !== modelIdx
 		);

 		await submitHandler();
-		models.set(await getModels(localStorage.token));
+		models.set(
+			await getModels(
+				localStorage.token,
+				$config?.features?.enable_direct_connections && ($settings?.directConnections ?? null)
+			)
+		);
 	};

 	onMount(async () => {
 		if ($user.role === 'admin') {
-			config = await getConfig(localStorage.token).catch((err) => {
+			evaluationConfig = await getConfig(localStorage.token).catch((err) => {
 				toast.error(err);
 				return null;
 			});
@@ -81,7 +101,7 @@
 	}}
 >
 	<div class="overflow-y-scroll scrollbar-hidden h-full">
-		{#if config !== null}
+		{#if evaluationConfig !== null}
 			<div class="">
 				<div class="text-sm font-medium mb-2">{$i18n.t('General Settings')}</div>

@@ -90,12 +110,12 @@
 						<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={config.ENABLE_EVALUATION_ARENA_MODELS} />
+							<Switch bind:state={evaluationConfig.ENABLE_EVALUATION_ARENA_MODELS} />
 						</Tooltip>
 					</div>
 				</div>

-				{#if config.ENABLE_EVALUATION_ARENA_MODELS}
+				{#if evaluationConfig.ENABLE_EVALUATION_ARENA_MODELS}
 					<hr class=" border-gray-50 dark:border-gray-700/10 my-2" />

 					<div class="flex justify-between items-center mb-2">
@@ -117,8 +137,8 @@
 					</div>

 					<div class="flex flex-col gap-2">
-						{#if (config?.EVALUATION_ARENA_MODELS ?? []).length > 0}
-							{#each config.EVALUATION_ARENA_MODELS as model, index}
+						{#if (evaluationConfig?.EVALUATION_ARENA_MODELS ?? []).length > 0}
+							{#each evaluationConfig.EVALUATION_ARENA_MODELS as model, index}
 								<Model
 									{model}
 									on:edit={(e) => {
--- a/Show More
+++ b/Show More