Deep Search & Semantic Memory
Empowers Claude with local semantic search and long-term memory for navigating massive codebases and documentation.
cm-deep-search acts as a high-capacity semantic memory bridge for large-scale projects that exceed standard LLM context windows. By integrating with the local qmd engine, it provides vector-based search, BM25 indexing, and LLM re-ranking to help Claude accurately retrieve relevant documentation and source code snippets. The skill features intelligent auto-detection that triggers only when project complexity reaches a specific threshold, ensuring Claude remains performant while gaining the ability to 'remember' and find information across thousands of files without manual indexing overhead.
主要功能
01Local semantic search using BM25 and vector embeddings
02Incremental indexing to ensure search results stay fresh after code changes
03MCP server support for seamless integration with Claude and IDEs
04Tiered memory architecture bridging working and long-term storage
0545 GitHub stars
06Zero-config auto-detection for oversized codebases and docs
使用场景
01Navigating legacy codebases with hundreds of source files and complex dependencies
02Maintaining context across long-term development cycles in large SaaS projects
03Retrieving specific details from massive technical documentation folders and PRDs