• Add new Textarea component
• Auto-resize between 40px-120px height
• Support Enter to submit, Shift+Enter for newline
• Add form autocomplete attributes
• Reset height after message submission
• Add refresh loading state with spinner
• Dynamic tooltips based on current label
• Improved refresh logic for labels
• Force dropdown re-render on refresh
• Update i18n for new tooltip messages
- Fix AsyncSelect component where cmdk's internal filtering was hiding search history items
- Change CommandItem value logic: use empty string when searchTerm is empty to match cmdk's filtering behavior
- Ensure search history displays correctly on initial load regardless of current queryLabel value
- Maintain proper selected state indication and all existing functionality
- Preserve normal search behavior when user types in search input
Fixes issue where dropdown only showed "*" and current queryLabel instead of full search history from localStorage.
- Replace hardcoded black text colors (#000) with dynamic theme-based colors
- Add createSigmaSettings function to generate settings based on theme
- Use labelColorDarkTheme (#B2EBF2) for dark mode and black (#000) for light mode
- Update GraphViewer component to react to theme changes
- Fixes issue where node and edge labels were invisible in dark mode
Fixes: Dark mode text visibility in knowledge graph visualization
- Add thinkingProcessed ref to track state
- Extract thinking content only once
- Preserve real-time thinking updates
- Reset processed flag on new queries
- Always update display content stream
- Add metadata formatting with timestamps
- Replace warning emoji with proper icons
- Show tooltips for errors and metadata
- Import AlertTriangle and Info icons
- Improve status cell layout and styling
- Fix documents being grouped by status after pagination and sorting
- Use backend-sorted data directly from currentPageDocs instead of re-grouping
- Preserve backend sort order to prevent status-based grouping
- Maintain backward compatibility with legacy docs structure
- Ensure all sorting fields (file name, dates, ID) work correctly without status grouping
The issue occurred because the frontend was re-grouping already-sorted data
from the backend by status, breaking the intended sort order. Now documents
are displayed in the exact order returned by the backend API.
Fixes: Document list sorting by file name was grouping by status instead of
maintaining proper sort order across all documents.
- Remove separate ENABLE_RERANK flag in favor of rerank_binding="null"
- Change default rerank binding from "cohere" to "null" (disabled)
- Update UI to display both rerank binding and model information
- Add intelligent refresh function to handle boundary cases
- Replace manual refresh with smart page preservation logic
- Auto-redirect to last page when current page becomes invalid
- Maintain user's browsing position during pipeline start/stop
- Fix issue where document list would reset to first page after pipeline operations
- Increase chunk_top_k from 10 to 20
- Reduce max_entity_tokens to 6000
- Reduce max_relation_tokens to 8000
- Update web UI default values
- Fix max_total_tokens to 30000
- Remove optional 'modes' parameter from aclear_cache() and clear_cache() methods
- Replace deprecated drop_cache_by_modes() with drop() method for complete cache clearing
- Update API endpoint to ignore mode-specific parameters and clear all cache
- Simplify frontend clearCache() function to send empty request body
This change ensures all LLM cache is cleared together.
- Remove History Turns UI component from QuerySettings
- Update settings store version to 17 with migration
- Force history_turns parameter to always be 0 in queries
- Prevent future modifications to history_turns setting
