tobi/qmd

mini cli search engine for your docs, knowledge bases, meeting notes, whatever. Tracking current sota approaches while being all local

What it solves

QMD (Query Markup Documents) provides a fully local, on-device search engine for personal knowledge bases, markdown notes, and documentation. It solves the problem of finding specific information across fragmented local files by combining multiple search strategies and LLM-powered refinement, ensuring high-quality results without sending data to the cloud.

How it works

QMD indexes local directories (collections) and employs a hybrid search pipeline:

  1. Query Expansion: An LLM generates variations of the user's query to improve recall.
  2. Parallel Retrieval: It simultaneously performs BM25 full-text search (keyword) and vector semantic search (embeddings) using local GGUF models via node-llama-cpp.
  3. Fusion: Results are combined using Reciprocal Rank Fusion (RRF) with a top-rank bonus to preserve exact matches.
  4. Re-ranking: A local LLM reranker scores the top candidates to refine the final order.
  5. Contextualization: Users can add descriptive metadata to paths, which helps LLMs make better contextual choices when selecting documents.

Who it’s for

  • Knowledge Workers: People with large collections of markdown notes or meeting transcripts who need a fast, private way to search them.
  • AI Agent Developers: Developers building agentic workflows that need a reliable tool for retrieving local context (supported via CLI, SDK, and MCP server).
  • Privacy-Conscious Users: Anyone who wants LLM-powered search capabilities without using external APIs.

Highlights

  • Local-First: All embeddings, reranking, and query expansion happen on-device using GGUF models.
  • Hybrid Search: Combines keyword (BM25) and semantic (vector) search for balanced precision and recall.
  • Agent-Ready: Includes a Model Context Protocol (MCP) server and a Node.js/Bun SDK for easy integration into AI agents.
  • Contextual Metadata: Allows adding tree-based context to collections to improve retrieval accuracy.
  • Flexible Retrieval: Supports retrieving documents by path, docid, or glob patterns with line-range controls.

Related

  • Project
    Raudaschl/rag-fusion

    A search methodology and implementation that uses LLM-generated query variations and Reciprocal Rank Fusion to improve document retrieval, especially when user terminology differs from the indexed text.

  • Project
    shinpr/mcp-local-rag

    A local RAG system that indexes private documents (PDF, DOCX, Markdown) and provides hybrid semantic and keyword search via MCP or CLI without using external APIs.

  • Project
    vercel-labs/lat.md

    A knowledge graph system for codebases that uses interconnected markdown files and source code annotations to provide AI agents and humans with scalable, searchable context.

  • Project
    flyhunterl/flymd

    A lightweight, local-first Markdown editor featuring AI-assisted writing, RAG-based semantic search, and high-precision PDF parsing.

  • Project
    PromtEngineer/localGPT

    A private, on-premise document intelligence platform that uses hybrid search and local LLMs to let users query and summarize their files without data leaving their machine.