qmd

Local hybrid RAG search combines BM25 keyword matching with semantic vector search to find relevant markdown notes and meeting transcripts, then applies LLM reranking to improve result relevance. This handles both exact terms and conceptual queries across your knowledge base.

To index your markdown knowledge base for local semantic search, run the command `qmd collection add <path-to-notes> --name notes` to register the directory, then execute `qmd embed` to build the BM25 and vector search index.

Yes, you can retrieve specific documents from your knowledge base using an MCP server tool. QMD exposes search tools via MCP for agent workflows and also provides direct CLI access for interactive document retrieval by ID or path.

The best way to search meeting transcripts for past project decisions is using hybrid retrieval with LLM reranking. You can query concepts across projects and scope results using collection and context metadata to locate exactly where a decision was made.

Yes, local document retrieval supports scoping by collection and context metadata. You can index document directories as individual collections and add per-collection context descriptions to boost search accuracy and scope your queries.