Embgrep

Gives Claude four semantic search tools over your local codebase without needing API keys or vector databases. Uses fastembed with ONNX Runtime to generate embeddings and stores them in SQLite. The index_directory tool chunks code by functions and docs by headings, semantic_search runs natural language queries with cosine similarity ranking, and update_index does incremental reindexing based on SHA-256 hashes. Supports 15+ file types including Python, JavaScript, TypeScript, Rust, Go, and Markdown. Useful when you want Claude to find relevant code or documentation by meaning rather than exact string matches, especially across large projects where keyword search falls short.