Deep Search Semantic Memory
Empowers Claude with local semantic search and long-term document retrieval for managing oversized codebases and extensive documentation.
cm-deep-search acts as an external semantic memory layer for Claude, specifically designed for projects that exceed standard context windows. By integrating with the qmd engine, it enables local BM25 and vector-based search across thousands of files, allowing the AI to instantly retrieve relevant documentation, meeting notes, and historical PRDs. It functions as a semantic disk search that complements short-term working memory, ensuring Claude remains accurate and context-aware even in massive enterprise-scale repositories without compromising data privacy.
주요 기능
0145 GitHub stars
02Incremental indexing to prevent stale results during active development
03Seamless MCP server integration for Claude and Cursor
04Local vector embeddings and BM25 search for 100% data privacy
05Cross-repository semantic queries with LLM-powered re-ranking
06Automated detection of oversized projects requiring external search
사용 사례
01Navigating massive codebases with 200+ source files without losing context
02Indexing meeting notes and PRDs to provide Claude with historical project context
03Retrieving specific architectural decisions from hundreds of legacy documentation files