Deep Search Semantic Memory
Enhances AI codebase understanding by providing local semantic search and long-term document retrieval for oversized projects.
Deep Search acts as an external long-term memory for Claude, integrating with tobi/qmd to provide 100% local BM25 and vector search capabilities. It automatically detects when a project exceeds standard context limits and suggests indexing documentation and source files, allowing the AI to query historical specs, meeting notes, and large-scale architecture docs that would otherwise be too bulky for the standard context window. By bridging working memory with semantic retrieval, it ensures Claude remains highly intelligent even in massive codebases.
Key Features
01Zero-config detection for oversized codebases and documentation folders
0245 GitHub stars
03Seamless MCP server integration for Claude, Cursor, and Antigravity
04Local semantic search using BM25, Vector, and LLM re-ranking
05Staleness prevention with incremental embedding and Git hook support
06Tiered memory architecture bridging short-term continuity with long-term retrieval
Use Cases
01Retrieving specific context from massive documentation folders containing over 50 markdown files
02Searching for historical architectural decisions across projects with hundreds of source files
03Maintaining project intelligence across multiple sessions by indexing PRDs, specs, and meeting notes