qmd

What problem does it solve? Finding information across large personal collections of markdown notes, meeting transcripts, and documentation is slow with plain grep and keyword search misses conceptual matches. This Skill provides a fully local, on-device search engine that combines keyword matching, semantic vector search, and LLM reranking without any cloud dependencies. ## Core Features & Use Cases - Hybrid Retrieval Pipeline: Combines BM25 keyword search (SQLite FTS5), vector embeddings, query expansion, and LLM reranking with Reciprocal Rank Fusion for high-quality results. - Collection Management: Index directories of markdown and text files as named collections, add context descriptions to improve retrieval accuracy, and generate embeddings with a single command. - MCP Integration: Exposes search tools via an MCP server in stdio or HTTP daemon mode, giving agents native search tools with warm models for fast queries. - Use Case: A user with hundreds of meeting transcripts asks "what decisions were made about the database migration" and gets reranked, contextually relevant excerpts in seconds, all processed locally. ## Quick Start Ask the agent to search your notes for a topic, for example: use qmd to find everything in my notes about the API redesign decisions.