- Added pre-calculation of estimated chunk count for HNSW capacity in `generate_dense_embeddings_centralized` to optimize indexing performance.
- Implemented binary vector generation with memory-mapped storage for efficient cascade search, including metadata saving.
- Introduced SPLADE sparse index generation with improved handling and metadata storage.
- Updated `ChainSearchEngine` to prefer centralized binary searcher for improved performance and added fallback to legacy binary index.
- Deprecated `BinaryANNIndex` in favor of `BinarySearcher` for better memory management and performance.
- Enhanced `SpladeEncoder` with warmup functionality to reduce latency spikes during first-time inference.
- Improved `SpladeIndex` with cache size adjustments for better query performance.
- Added methods for managing binary vectors in `VectorMetadataStore`, including batch insertion and retrieval.
- Created a new `BinarySearcher` class for efficient binary vector search using Hamming distance, supporting both memory-mapped and database loading modes.
- Added centralized SPLADE database and vector storage configuration in config.py.
- Updated embedding_manager.py to support centralized SPLADE database path.
- Enhanced generate_embeddings and generate_embeddings_recursive functions for centralized storage.
- Introduced centralized ANN index creation in ann_index.py.
- Modified hybrid_search.py to utilize centralized vector index for searches.
- Implemented methods to discover and manage centralized SPLADE and HNSW files.
- Added category support for programming and documentation languages in Config.
- Implemented category-based filtering in HybridSearchEngine to improve search relevance based on query intent.
- Introduced functions for filtering results by category and determining file categories based on extensions.
- Updated VectorStore to include a category column in the database schema and modified chunk addition methods to support category tagging.
- Enhanced the WatcherConfig to ignore additional common directories and files.
- Created a benchmark script to compare performance between Binary Cascade, SPLADE, and Vector semantic search methods, including detailed result analysis and overlap comparison.
- Added `splade_encoder.py` for ONNX-optimized SPLADE encoding, including methods for encoding text and batch processing.
- Created `SPLADE_IMPLEMENTATION.md` to document the SPLADE encoder's functionality, design patterns, and integration points.
- Introduced migration script `migration_009_add_splade.py` to add SPLADE metadata and posting list tables to the database.
- Developed `splade_index.py` for managing the SPLADE inverted index, supporting efficient sparse vector retrieval.
- Added verification script `verify_watcher.py` to test FileWatcher event filtering and debouncing functionality.
Architecture refactoring for multi-provider rotation:
Backend:
- Add EmbeddingPoolConfig type with autoDiscover support
- Implement discoverProvidersForModel() for auto-aggregation
- Add GET/PUT /api/litellm-api/embedding-pool endpoints
- Add GET /api/litellm-api/embedding-pool/discover/:model preview
- Convert ccw-litellm status check to async with 5-min cache
- Maintain backward compatibility with legacy rotation config
Frontend:
- Add "Embedding Pool" tab in API Settings
- Auto-discover providers when target model selected
- Show provider/key count with include/exclude controls
- Increase sidebar width (280px → 320px)
- Add sync result feedback on save
Other:
- Remove worker count limits (was max=32)
- Add i18n translations (EN/CN)
- Update .gitignore for .mcp.json
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Added functions to get and update CodexLens embedding rotation configuration.
- Introduced functionality to retrieve enabled embedding providers for rotation.
- Created endpoints for managing rotation configuration via API.
- Enhanced dashboard UI to support multi-provider rotation configuration.
- Updated internationalization strings for new rotation features.
- Adjusted CLI commands and embedding manager to support increased concurrency limits.
- Modified hybrid search weights for improved ranking behavior.
- Added token estimation and batching functionality in LiteLLMEmbedder to handle large text inputs efficiently.
- Updated embed method to support max_tokens_per_batch parameter for better API call management.
- Introduced new API routes for managing custom CLI endpoints, including GET, POST, PUT, and DELETE methods.
- Enhanced CLI history component to support source directory context for native session content.
- Improved error handling and logging in various components for better debugging and user feedback.
- Added internationalization support for new API endpoint features in the i18n module.
- Updated CodexLens CLI commands to allow for concurrent API calls with a max_workers option.
- Enhanced embedding manager to track model information and handle embeddings generation more robustly.
- Added entry points for CLI commands in the package configuration.
- Added JSON-based settings management in Config class for embedding and LLM configurations.
- Introduced methods to save and load settings from a JSON file.
- Updated BaseEmbedder and its subclasses to include max_tokens property for better token management.
- Enhanced chunking strategy to support recursive splitting of large symbols with improved overlap handling.
- Implemented comprehensive tests for recursive splitting and chunking behavior.
- Added CLI tools configuration management for better integration with external tools.
- Introduced a new command for compacting session memory into structured text for recovery.
- Updated embedding_manager.py to include backend parameter in model configuration.
- Modified model_manager.py to utilize cache_name for ONNX models.
- Refactored hybrid_search.py to improve embedder initialization based on backend type.
- Added backend column to vector_store.py for better model configuration management.
- Implemented migration for existing database to include backend information.
- Enhanced API settings implementation with comprehensive provider and endpoint management.
- Introduced LiteLLM integration guide detailing configuration and usage.
- Added examples for LiteLLM usage in TypeScript.
- Create ccw-litellm Python package with AbstractEmbedder and AbstractLLMClient interfaces
- Add BaseEmbedder abstraction and factory pattern to codex-lens for pluggable backends
- Implement API Settings dashboard page for provider credentials and custom endpoints
- Add REST API routes for CRUD operations on providers and endpoints
- Extend CLI with --model parameter for custom endpoint routing
- Integrate existing context-cache for @pattern file resolution
- Add provider model registry with predefined models per provider type
- Include i18n translations (en/zh) for all new UI elements
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Introduced a new CodexLens Manager item in the dashboard for easier access.
- Implemented GPU management commands in the CLI, including listing available GPUs, selecting a specific GPU, and resetting to automatic detection.
- Enhanced the embedding generation process to utilize GPU resources more effectively, including batch size optimization for better performance.
- Updated the embedder to support device ID options for GPU selection, ensuring compatibility with DirectML and CUDA.
- Added detailed logging and error handling for GPU detection and selection processes.
- Updated package version to 6.2.9 and added comprehensive documentation for Codex Agent Execution Protocol.
- Fix model installation detection using fastembed ONNX cache names
- Add embeddings_config table for model metadata tracking
- Fix hybrid search segfault by using single-threaded GPU mode
- Suppress INFO logs in JSON mode to prevent error display
- Add model dropdown filtering to show only installed models
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Updated the Chunker class to adjust the window movement logic, ensuring proper handling of overlap lines.
- Introduced a new smart search tool with features including intent classification, CodexLens integration, multi-backend search routing, and index status checking.
- Implemented various search modes (auto, hybrid, exact, ripgrep, priority) with detailed metadata and error handling.
- Added support for progress tracking during index initialization and enhanced output transformation based on user-defined modes.
- Included comprehensive documentation for usage and parameters in the smart search tool.
- Updated the dashboard template to hide the Code Graph Explorer feature.
- Enhanced the `executeCodexLens` function to use `exec` for better cross-platform compatibility and improved command execution.
- Changed the default `maxResults` and `limit` parameters in the smart search tool to 10 for better performance.
- Introduced a new `priority` search mode in the smart search tool, replacing the previous `parallel` mode, which now follows a fallback strategy: hybrid -> exact -> ripgrep.
- Optimized the embedding generation process in the embedding manager by batching operations and using a cached embedder instance to reduce model loading overhead.
- Implemented a thread-safe singleton pattern for the embedder to improve performance across multiple searches.
Previously, embeddings were only generated for root directory files (1.6% coverage, 5/303 files).
This fix implements recursive processing across all subdirectory indexes, achieving 100% coverage
with 2,042 semantic chunks across all 303 files in 26 index databases.
Key improvements:
1. **Recursive embeddings generation** (embedding_manager.py):
- Add generate_embeddings_recursive() to process all _index.db files in directory tree
- Add get_embeddings_status() for comprehensive coverage statistics
- Add discover_all_index_dbs() helper for recursive file discovery
2. **Enhanced CLI commands** (commands.py):
- embeddings-generate: Add --recursive flag for full project coverage
- init: Use recursive generation by default for complete indexing
- status: Display embeddings coverage statistics with 50% threshold
3. **Smart search routing improvements** (smart-search.ts):
- Add 50% embeddings coverage threshold for hybrid mode routing
- Auto-fallback to exact mode when coverage insufficient
- Strip ANSI color codes from JSON output for correct parsing
- Add embeddings_coverage_percent to IndexStatus and SearchMetadata
- Provide clear warnings with actionable suggestions
4. **Documentation and analysis**:
- Add SMART_SEARCH_ANALYSIS.md with initial investigation
- Add SMART_SEARCH_CORRECTED_ANALYSIS.md revealing true extent of issue
- Add EMBEDDINGS_FIX_SUMMARY.md with complete fix summary
- Add check_embeddings.py script for coverage verification
Results:
- Coverage improved from 1.6% (5/303 files) to 100% (303/303 files) - 62.5x increase
- Semantic chunks increased from 10 to 2,042 - 204x increase
- All 26 subdirectory indexes now have embeddings vs just 1
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Implement tests for migration 005 to verify removal of deprecated fields in the database schema.
- Ensure that new databases are created with a clean schema.
- Validate that keywords are correctly extracted from the normalized file_keywords table.
- Test symbol insertion without deprecated fields and subdir operations without direct_files.
- Create a detailed search comparison test to evaluate vector search vs hybrid search performance.
- Add a script for reindexing projects to extract code relationships and verify GraphAnalyzer functionality.
- Include a test script to check TreeSitter parser availability and relationship extraction from sample files.
