* ✨ feat: Add MCP Reinitialization to MCPPanel
- Refactored tool caching to include user-specific tools in various service files.
- Refactored MCPManager class for clarity
- Added a new endpoint for reinitializing MCP servers, allowing for dynamic updates of server configurations.
- Enhanced the MCPPanel component to support server reinitialization with user feedback.
* 🔃 refactor: Simplify Plugin Deduplication and Clear Cache Post-MCP Initialization
- Replaced manual deduplication of tools with the dedicated `filterUniquePlugins` function for improved readability.
- Added back cache clearing for tools after MCP initialization to ensure fresh data is used.
- Removed unused exports from `PluginController.js` to clean up the codebase.
* 🏗️ refactor: Extract reasoning key logic into separate function
* refactor: Ensure `overrideProvider` is always defined in `getProviderConfig` result, and only used in `initializeAgent` if different from `agent.provider`
* feat: new title configuration options across services
- titlePrompt
- titleEndpoint
- titlePromptTemplate
- new "completion" titleMethod (new default)
* chore: update @librechat/agents and conform openai version to prevent SDK errors
* chore: add form-data package as a dependency and override to v4.0.4 to address CVE-2025-7783
* feat: add support for 'all' endpoint configuration in AppService and corresponding tests
* refactor: replace HttpsProxyAgent with ProxyAgent from undici for improved proxy handling in assistant initialization
* chore: update frontend review workflow to limit package paths to data-provider
* chore: update backend review workflow to include all package paths
- Added firecrawlOptions configuration field to librechat.yaml
- Refactored web.ts to live in packages/api rather than data-provider
- Updated imports from web.ts to reflect new location
- Added firecrawlOptions to FirecrawlConfig interface
- Added firecrawlOptions to authResult of loadWebSearchAuth so it gets properly passed to agents to be built into firecrawl payload
- Added tests for firecrawlOptions to web.spec.ts
* wip: first pass content strings
* 📦 chore: update @langchain/core to v0.3.62 for data-provider dev dep.
* 📦 chore: bump @langchain/core to v0.3.62 for api dep.
* 📦 chore: move @langchain/core to peerDependencies in package.json and package-lock.json
* fix: update formatContentStrings to create HumanMessage directly from formatted content
* chore: import order
* 🕒 refactor: Use Legacy Content for Custom Endpoints to Improve Compatibility
- Also applies to Azure serverless endpoints from AI Foundry
* chore: move useLegacyContent condition before early return
* fix: Ensure useLegacyContent is set only when options are available
* fix: Add `isRegenerate` flag to chat payload to avoid saving temporary response IDs
* fix: Remove unused `isResubmission` flag
* ci: Add tests for responseMessageId regeneration logic in BaseClient
* Fix scanning of the uploaded images folder on startup
* Re-write tests to pass linting
* Disable image output gzip scan by default
* Add `ENABLE_IMAGE_OUTPUT_GZIP_SCAN` to `.env.example`
Implements permission validation before allowing agent avatar uploads. Only admins, the agent's author, or users of collaborative agents can modify avatars. Also improves error handling by checking for agent existence upfront and simplifies avatar update logic.
Co-authored-by: Sai Nihas <sai.nihas@shopify.com>
- Added Constants import in PluginController for better organization.
- Renamed cachedTools to cachedToolsArray for clarity in PluginController.
- Ensured getCachedTools returns an empty object if no tools are found.
- Cleared tools array cache after MCP initialization in initializeMCP for consistency.
* fix: false flagging object as empty object when it has `additionalProperties` field
* 🔧 fix: Implement $ref resolution in JSON Schema handling
* 🔧 fix: Resolve JSON Schema references before conversion to Zod
* chore: move zod logic packages/api
* feat: Enhance loadServiceKey to support stringified JSON input
* chore: Update GOOGLE_SERVICE_KEY_FILE_PATH to GOOGLE_SERVICE_KEY_FILE for consistency
- Introduced new violation scores for TTS, STT, Fork, Import, and File Upload actions in the .env.example file.
- Updated logViolation function to accept a score parameter, allowing for dynamic severity levels based on the action type.
- Modified limiters for Fork, Import, Message, STT, TTS, Tool Call, and File Upload to utilize the new violation scores when logging violations.
* Updated the PATCH /memories/:key endpoint to allow key changes while ensuring no duplicate keys exist.
* Improved error handling in MemoryCreateDialog and MemoryEditDialog for key validation and duplication scenarios.
* Added a new translation for memory key validation error in translation.json.
* 🔧 refactor: Improve Google Key Handling in `loadAsyncEndpoints`
- Enhanced logic to check if GOOGLE_KEY is provided, including user-provided checks.
- Updated service key loading mechanism to only attempt loading if GOOGLE_KEY is not provided.
- Added error logging for service key loading failures.
* 🔧 refactor: Enhance service key loading logic in `initializeClient`
* chore: bump @librechat/agents to ^2.4.54 for anthropic web search support
* WIP: hardcoded web search tool usage
* feat: Implement web search functionality in Anthropic integration
- Updated parameters panel to include web search for anthropic models.
- Updated necessary schemas to accomodate toggle functionality
* chore: Set default web search option to false in anthropicSettings
* refactor: Rename webSearch to web_search for consistency across settings and schemas
* chore: bump @librechat/agents to v2.4.55
---------
Co-authored-by: Dustin Healy <dustinhealy1@gmail.com>
* 🔍 feat: Add SearXNG Web Search support and enhance ApiKeyDialog
- Updated WebSearch component to include authentication data for web search functionality so it won't show badge after being revoked
- Refactored ApiKeyDialog to streamline provider, scraper, and reranker selection with new InputSection component
- Added support for SearXNG as a search provider and updated translation files accordingly
- Improved form handling in useAuthSearchTool to accommodate new API keys and URLs
* 📜 chore: remove unused i18next key
* 📦 chore: address comments (swap API key and URL fields in SearXNG config, change input fields to 'text' from 'password'
* 📦 chore: make URL fields go first in ApiKeyDialog
* chore: bump @librechat/agents to v2.4.52
* ci: update webSearch configuration to include searxng fields in AppService.spec.js
---------
Co-authored-by: Danny Avila <danny@librechat.ai>
* chore: Improve error logging for fetching conversations, and use new TS packages for utils
* feat: Implement fork limiters for conversation forking requests
* chore: error message for conversation index deletion to clarify syncing behavior
* feat: Enhance error handling for forking with rate limit message
* refactor: Add validation schemas for agent creation and updates
* fix: Ensure author validation is applied in correct order for agent update handler
* ci: Add comprehensive unit tests for agent creation and update handlers with mass assignment protection
* fix: add missing web_search tool in system tools configuration
* feat: add useAgentCapabilities hook to manage agent capabilities
* refactor: move agents and endpoints configuration to AgentPanel context provider
* refactor: implement useGetAgentsConfig hook for consolidated agents and endpoints management
* refactor: enhance ToolsDropdown to utilize agent capabilities and streamline dropdown item rendering
* chore: reorder return values in useAgentCapabilities for improved clarity
* refactor: enhance agent capabilities handling in AttachFileMenu and update file handling logic to allow capabilities to be used for non-agents endpoints
* 🔍 feat: Introduce Web Search Functionality for OpenAI API
- Added a new web_search parameter to enable web search capabilities in the OpenAI configuration.
- Updated the DynamicSlider component for improved styling.
- Enhanced the useSetIndexOptions hook to auto-enable the Responses API when web search is activated.
- Modified relevant schemas, types, and translation files to support the new web search feature.
* chore: remove comments
* refactor: tool handling in initializeAgent for better clarity and functionality and reflection of openai features
---------
Co-authored-by: Danny Avila <danny@librechat.ai>
* ✨ feat: Add Google Search Grounding Feature and Update Agent Tool Initialization
- Introduced a new grounding option in the Google configuration to enable real-time web search results.
- Updated the agent initialization to concatenate additional tools from options.
- Enhanced translation files to include descriptions for the new grounding feature.
- Modified relevant schemas and parameter settings to support the grounding functionality.
* 🔑 chore: Update @librechat/agents dependency to version 2.4.50
* ✅ fix: Ensure tools array is initialized before concatenation in initializeAgent function
* chore: Update version of librechat-data-provider to 0.7.899 and add GOOGLE_TOOL_CONFLICT error type
* fix: Adjust label class for better text wrapping in DynamicSwitch component
* fix: Handle Google tool conflict error and update error messages in translation
* fix: Restore grounding setting in googleCol2 configuration
---------
Co-authored-by: Danny Avila <danny@librechat.ai>
* 🐛 fix: Enhance provider validation and error handling in getProviderConfig function
* WIP: edit text part
* refactor: Allow updating of both TEXT and THINK content types in message updates
* WIP: first pass, save & submit
* chore: remove legacy generation user message field
* feat: merge edited content
* fix: update placeholder and description for bedrock setting
* fix: remove unsupported warning message for AI resubmission
* Implemented new uploadGoogleVertexMistralOCR function for processing OCR using Google Vertex AI.
* Added vertexMistralOCRStrategy to handle file uploads.
* Updated FileSources and OCRStrategy enums to include vertexai_mistral_ocr.
* Introduced helper functions for JWT creation and Google service account configuration loading.
* refactor: access control logic to TypeScript
* chore: Change EndpointURLs to a constant object for improved type safety
* 🐛 fix: Enhance agent access control by adding skipAgentCheck functionality
* 🐛 fix: Add endpointFileConfig prop to AttachFileMenu and update file handling logic
* 🐛 fix: Update tool handling logic to support optional groupedTools and improve null checks, add dedicated tool dialog for Assistants
* chore: Export Accordion component from UI index for improved modularity
* feat: Add ActivePanelContext for managing active panel state across components
* chore: Replace string IDs with EModelEndpoint constants for assistants and agents in useSideNavLinks
* fix: Integrate access checks for agent creation and deletion routes in actions.js
* feat: Add configurable retention period for temporary chats
* Addressing eslint errors
* Fix: failing test due to missing registration
* Update: variable name and use hours instead of days for chat retention
* Addressing comments
* chore: fix import order in Conversation.js
* chore: import order in Message.js
* chore: fix import order in config.ts
* chore: move common methods to packages/api to reduce potential for circular dependencies
* refactor: update temp chat retention config type to Partial<TCustomConfig>
* refactor: remove unused config variable from AppService and update loadCustomConfig tests with logger mock
* refactor: handle model undefined edge case by moving Session model initialization inside methods
---------
Co-authored-by: Rakshit Tiwari <rak1729e@gmail.com>
* feat: support thinking budget parameter for Gemini 2.5 series (#6949, #7542)
https://ai.google.dev/gemini-api/docs/thinking#set-budget
* refactor: update thinking budget minimum value to -1 for dynamic thinking
- see: https://ai.google.dev/gemini-api/docs/thinking#set-budget
* chore: bump @librechat/agents to v2.4.43
* refactor: rename LLMConfigOptions to OpenAIConfigOptions for clarity and consistency
- Updated type definitions and references in initialize.ts, llm.ts, and openai.ts to reflect the new naming convention.
- Ensured that the OpenAI configuration options are consistently used across the relevant files.
* refactor: port Google LLM methods to TypeScript Package
* chore: update @librechat/agents version to 2.4.43 in package-lock.json and package.json
* refactor: update thinking budget description for clarity and adjust placeholder in parameter settings
* refactor: enhance googleSettings default value for thinking budget to support dynamic adjustment
* chore: update @librechat/agents to v2.4.44 for Vertex Dynamic Thinking workaround
* refactor: rename google config function, update `createRun` types, use `reasoning` as `reasoningKey` for Google
* refactor: simplify placeholder handling in DynamicInput component
* refactor: enhance thinking budget description for clarity and allow automatic decision by setting to "-1"
* refactor: update text styling in OptionHover component for improved readability
* chore: update @librechat/agents dependency to v2.4.46 in package.json and package-lock.json
* chore: update @librechat/api version to 1.2.5 in package.json and package-lock.json
* refactor: enhance `clientOptions` handling by filtering `omitTitleOptions`, add `json` field for Google models
---------
Co-authored-by: ciffelia <15273128+ciffelia@users.noreply.github.com>
* Enhanced existing tests for the `resolveHeaders` function to cover all user field placeholders and messy scenarios.
* Added basic integration tests for custom endpoints initialization file
