- Introduced dynamic provider registration functionality, allowing for flexible management of providers through a new registry system.
- Refactored HubProvider to streamline model resolution and improve error handling for unsupported models.
- Added utility functions for managing dynamic providers, including registration, cleanup, and alias resolution.
- Updated index exports to include new dynamic provider APIs, enhancing overall usability and integration.
- Removed outdated provider files and simplified the provider management structure for better maintainability.
- Added a new example file demonstrating the usage of the Hub Provider for routing to multiple underlying providers.
- Implemented the Hub Provider to support model ID parsing and routing based on a specified format.
- Refactored provider management by introducing a Registry Management class for better organization and retrieval of provider instances.
- Updated the Provider Initializer to streamline the initialization and registration of providers, enhancing overall flexibility and usability.
- Removed outdated files related to provider creation and dynamic registration to simplify the codebase.
- Removed outdated ConfigManager and factory files to streamline model management.
- Added ModelResolver for improved model ID resolution, supporting both traditional and namespaced formats.
- Introduced DynamicProviderRegistry for dynamic provider management, enhancing flexibility in model handling.
- Updated index exports to reflect the new structure and maintain compatibility with existing functionality.
- Upgraded `@ai-sdk/openai` to version 2.0.19 in `yarn.lock` and `package.json` for improved functionality and compatibility.
- Updated `@ai-sdk/provider-utils` to version 3.0.5, enhancing dependency management.
- Added `TypedToolError` type export in `index.ts` for better error handling.
- Removed unnecessary console logs in `webSearchPlugin` for cleaner code.
- Refactored type handling in `createProvider` to ensure proper type assertions.
- Enforced `topicId` as a required field in the `ModernAiProvider` configuration for stricter validation.
- Upgraded `@openrouter/ai-sdk-provider` to version ^1.1.2 in package.json and yarn.lock for improved functionality.
- Enhanced `convertMessageToSdkParam` and related functions to support additional model parameters, improving message conversion for various AI models.
- Integrated logging for error handling in file processing functions to aid in debugging and user feedback.
- Added support for native PDF input handling based on model capabilities, enhancing file processing features.
- Updated MemorySearchTool to utilize aiSdk for better integration and removed unused imports.
- Refactored WebSearchTool to streamline search results handling, changing from an array to a structured object for clarity.
- Adjusted MessageTool and MessageWebSearchTool components to reflect changes in tool response structure.
- Enhanced error handling and logging in tool callbacks for improved debugging and user feedback.
- Updated the ToolCallChunkHandler to support both MCPTool and NormalToolResponse types, improving flexibility in tool response management.
- Refactored type definitions for MCPToolResponse and introduced NormalToolResponse to better differentiate between tool response types.
- Enhanced logging in MCP utility functions for improved error tracking and debugging.
- Cleaned up type imports and ensured consistent handling of tool responses across various chunks.
- Upgraded multiple `@ai-sdk` packages in `yarn.lock` and `package.json` to their latest versions for enhanced functionality and compatibility.
- Improved type safety in `searchOrchestrationPlugin` by adding optional chaining to handle potential undefined values in knowledge bases.
- Cleaned up dependency declarations to use caret (^) for versioning, ensuring compatibility with future updates.
- Upgraded `@ai-sdk/gateway` to version 1.0.8 and `@ai-sdk/provider-utils` to version 3.0.4 in yarn.lock for enhanced functionality.
- Updated `ai` dependency in package.json to version ^5.0.16 for better compatibility.
- Added logging functionality in `AiSdkToChunkAdapter` to track chunk types and improve debugging.
- Refactored plugin imports to streamline code and enhance readability.
- Removed unnecessary console logs in `searchOrchestrationPlugin` to clean up the codebase.
- Extracted `StreamEventManager` and `ToolExecutor` classes from `promptToolUsePlugin.ts` to improve code organization and reduce complexity.
- Enhanced tool execution logic with better error handling and event management.
- Updated the `createPromptToolUsePlugin` function to utilize the new classes for cleaner implementation.
- Improved recursive call handling and result formatting for tool executions.
- Streamlined the overall flow of tool calls and event emissions within the plugin.
- Upgraded `vitest` dependency to version 3.2.4 in package.json and yarn.lock for improved testing capabilities.
- Removed console error logging in provider validation functions to streamline error handling.
- Added comprehensive tests for the AiProviderRegistry functionality, ensuring robust provider management and dynamic registration.
- Introduced new test cases for provider schemas to validate configurations and IDs.
- Deleted outdated registry test file to maintain a clean test suite.
- Updated the `defaultParseToolUse` function to return both parsed results and remaining content, improving usability.
- Introduced a new `TagExtractor` class for flexible tag extraction, supporting various tag formats.
- Modified type definitions to reflect changes in parsing function signatures.
- Enhanced handling of tool call events in the `ToolCallChunkHandler` for better integration with the new parsing logic.
- Added `isBuiltIn` property to the `MCPTool` interface for clearer tool categorization.
- Introduced a new Zod-based schema for provider validation, improving type safety and consistency.
- Added support for dynamic provider IDs and enhanced the provider registration process.
- Updated the AiProviderRegistry to utilize the new validation functions, ensuring robust provider management.
- Added tests for the provider registry to validate dynamic imports and functionality.
- Updated yarn.lock to reflect the latest dependency versions.
* fix: prevent incorrect @ symbol deletion in QuickPanel
- Track trigger source (input vs button) and @ position
- Only delete @ when triggered by input with model selection
- Button-triggered panels never delete text content
- Validate @ still exists at recorded position before deletion
* feat: delete search text along with @ symbol
- Pass searchText from QuickPanel to onClose callback
- Delete both @ and search text (e.g., @cla) when model selected
- Validate text matches before deletion for safety
- Fallback to deleting only @ if text doesn't match
* refactor: clarify ESC vs Backspace behavior in QuickPanel
- ESC: Cancel operation, delete @ + searchText when models selected
- Backspace: Natural editing, @ already deleted by browser, no extra action
- Clear separation of intent improves predictability and UX
* fix(export): robustly export reasoning and handle errors
* fix(export): normalize <br> to newline before notion parsing
* feat(i18n): add notion truncation and unify export warn keys
* refactor(export): add typing, state guards, and error logging
* fix(export): preserve existing <br> in reasoning when convert to html
* feat(export): add DOMPurify sanitization for reasoning content
* chore(deps): remove unused @types/dompurify dev dep
* chore(deps): remove dompurify dependency
Remove dompurify from package.json and yarn. The changes
delete dompurify entries and simplify the lockfile resolution so the
project no longer declares dompurify as a direct dependency.
This cleans up unused dependency declarations and prevents installing
dompurify when it is not required.
* refactor(ManageModelsPopup): remove margin of Empty
* chore: use destroyOnHidden rather than deprecated destroyTooltipOnHide
* refactor: center Empty
* fix: Update KnowledgeBase form and service to handle preprocess provider correctly
- Enhanced useKnowledgeBaseForm hook to set preprocessProvider with the correct providerId type.
- Modified getKnowledgeBaseParams function to retrieve preprocess provider from the store instead of the base, ensuring accurate provider assignment.
* fix: Remove unused providerId from preprocessProvider in useKnowledgeBaseForm hook
- Cleaned up the useKnowledgeBaseForm hook by removing the unused providerId property from the preprocessProvider object, ensuring a more accurate representation of the data structure.
* format code
* feat: Sync preprocess provider updates across knowledge bases
- Added a new action to synchronize updates to the preprocess provider across all knowledge bases that reference it.
- Updated the usePreprocessProvider hook to dispatch the new sync action after updating the provider.
- Modified getKnowledgeBaseParams to ensure the correct preprocess provider is assigned when retrieving knowledge base parameters.
* fix: Update sync logic for preprocess provider updates
- Modified the syncPreprocessProvider action to merge updates directly into the existing provider object instead of replacing it.
- Adjusted the usePreprocessProvider hook to only dispatch the sync action when specific fields (apiHost, apiKey, model) are updated, improving efficiency.
* refactor: replace axios and node fetch with electron's net module for network requests in preprocess providers
- Updated Doc2xPreprocessProvider and MineruPreprocessProvider to use net.fetch instead of axios for making HTTP requests.
- Improved error handling for network responses across various methods.
- Removed unnecessary AxiosRequestConfig and related code to streamline the implementation.
* lint
* refactor(Doc2xPreprocessProvider): enhance file validation and upload process
- Added file size validation to prevent loading files larger than 300MB into memory.
- Implemented file size check before reading the PDF to ensure efficient memory usage.
- Updated the file upload method to use a stream, setting the 'Content-Length' header for better handling of large files.
* refactor(brave-search): update net.fetch calls to use url.toString()
- Modified all instances of net.fetch to use url.toString() for better URL handling.
- Ensured consistency in how URLs are passed to the fetch method across various functions.
* refactor(MCPService): improve URL handling in net.fetch calls
- Updated net.fetch to use url.toString() for better type handling of URLs.
- Ensured consistent URL processing across the MCPService class.
* feat(ProxyManager): integrate axios with fetch proxy support
- Added axios as a dependency to enable fetch proxy usage.
- Implemented logic to set axios's adapter to 'fetch' for proxy handling.
- Preserved original axios adapter for restoration when disabling the proxy.
- Updated ProviderId type from an intersection of keyof ExtensibleProviderSettingsMap and string to a union, allowing for greater compatibility with dynamic provider settings.
- Changed ProviderId type from a union of keyof ExtensibleProviderSettingsMap and string to an intersection, enhancing type safety.
- Renamed appendTrace method to appendMessageTrace in SpanManagerService for clarity and consistency.
- Updated references to appendTrace in useMessageOperations and ApiService to use the new method name.
- Added a new appendTrace method in SpanManagerService to bind existing traces, improving trace management.
- Adjusted topicId handling in fetchMessagesSummary to default to an empty string for better consistency.
- Updated various package versions in yarn.lock for improved compatibility and performance.
- Removed obsolete patches for antd and openai, streamlining the dependency management.
- Adjusted icon imports in Dropdown and useIcons to utilize Lucide icons for better visual consistency.
- Renamed method from _getPreReleaseVersionFromGithub to _getReleaseVersionFromGithub for clarity.
- Enhanced logic to check for the latest release version using semver.
- Removed unnecessary checks related to test plans when updates are not available.
- Improved logging for better traceability of release version fetching.
- Enhanced the buildStreamTextParams function to include enableUrlContext in the capabilities object, improving the parameter set for AI interactions.
- Introduced a new isSupportedFlexServiceTier function to streamline model support checks, enhancing code clarity and maintainability.
- Introduced a patch to replace the RightOutlined icon with ChevronRight in the Dropdown component for improved visual consistency.
- Updated the useIcons hook to utilize ChevronDown instead of DownOutlined, enhancing the icon set with Lucide icons.
- Adjusted icon properties for better customization and styling options.
- Modified the constructor of ModernAiProvider to accept an optional provider parameter, enhancing flexibility in provider selection.
- Removed deprecated and unused functions related to search keyword extraction and search summary fetching, streamlining the codebase.
- Updated import statements and adjusted related logic to reflect the removal of obsolete functions, improving maintainability.
* fix(QuickPanel): auto-close panel when no commands match (#7824)
Fixes the issue where QuickPanel remains visible when user types
invalid slash commands. Now the panel intelligently closes after
300ms when no matching commands are found.
- Add smart delayed closing mechanism for unmatched searches
- Optimize memory management with proper timer cleanup
- Preserve existing trigger behavior for / and @ symbols
* feat(inputbar): intelligent @ symbol handling on model selection panel close
- Add smart @ character deletion when user selects models and closes panel with ESC/Backspace
- Preserve @ character when user closes panel without selecting any models
- Implement action tracking using useRef to detect user model interactions
- Support both ESC key and Backspace key for consistent behavior
- Use React setState instead of DOM manipulation for proper state management
Resolves user experience issue where @ symbol always remained after closing model selection panel
* perf(QuickPanel): optimize timer management and fix React anti-patterns
- Move side effects from useMemo to useEffect for proper React lifecycle
- Add automatic timer cleanup on component unmount and dependency changes
- Remove unnecessary timer creation/destruction on each search input
- Improve memory management and prevent potential memory leaks
- Maintain existing smart auto-close functionality with better performance
Fixes React anti-pattern where side effects were executed in useMemo,
which should be a pure function. This improves performance especially
when users type quickly in the search input.
* refactor(QuickPanel): remove redundant timer cleanup useEffect
Remove duplicate timer cleanup logic as the existing useEffect at line 141-164
already handles component unmount cleanup properly.
* refactor(QuickPanel): optimize useEffect dependencies and timer cleanup logic
- Replace overly broad `ctx` dependency with precise `[ctx.isVisible, searchText, list.length, ctx.close]`
- Move timer cleanup before visibility check to ensure proper cleanup on panel hide
- Add early return when panel is invisible to prevent unnecessary timer creation
- Improve performance by avoiding redundant effect executions
- Fix edge case where timers might not be cleared when panel becomes invisible
Addresses review feedback about dependency array optimization while maintaining
existing auto-close functionality and improving memory management.
* feat(QuickPanel): implement smart re-opening with dependency optimization
Features:
- Implement smart re-opening during deletion with real-time matching
- Only reopen panel when actual matches exist to avoid unnecessary interactions
- Add intelligent @ symbol handling on model selection panel close
- Optimize search text length limits (≤10 chars) for performance
Fixes:
- Fix useMemo dependency from overly broad [ctx, searchText] to precise [ctx.isVisible, ctx.symbol, ctx.list, searchText]
- Resolve trailing whitespace formatting issues
- Eliminate ESLint exhaustive-deps warnings while maintaining stability
- Prevent unnecessary re-renders when unrelated ctx properties change
Performance improvements ensure optimal QuickPanel responsiveness while maintaining
existing auto-close functionality and improving user experience.
* fix(ci): add eslint-disable comment for exhaustive-deps warning
The useEffect dependency array [ctx.isVisible, searchText, list.length, ctx.close]
is intentionally precise to avoid unnecessary re-renders when unrelated ctx
properties change. Adding ctx object would cause performance degradation.
* refactor(QuickPanel): remove smart re-opening logic during deletion
- Remove 62 lines of complex deletion detection logic from Inputbar component
- Eliminates performance overhead from frequent string matching during typing
- Reduces code complexity and potential edge cases
- Maintains simple and predictable QuickPanel behavior
- Improves maintainability by removing unnecessary "smart" features
The deletion-triggered smart reopening feature added unnecessary complexity
without significant user benefit. Users can simply type / or @ again to
reopen panels when needed.
- Modified MessageKnowledgeSearch component to display additional context from toolInput.
- Updated the fetch complete message to reflect the count of knowledgeReferences instead of toolOutput.
- Adjusted the mapping of results to iterate over knowledgeReferences for rendering.
- Integrated tracing capabilities into the ModernAiProvider, allowing for better tracking of AI completions and image generation processes.
- Added a new TelemetryPlugin to inject telemetry data into AI SDK requests, ensuring compatibility with existing tracing systems.
- Updated middleware and plugin configurations to support topic-based tracing, improving the overall observability of AI interactions.
- Introduced comprehensive logging throughout the AI SDK processes to facilitate debugging and performance monitoring.
- Added unit tests for new functionalities to ensure reliability and maintainability.
