qmd

What This Skill Does

qmd is a powerful, local-first search and indexing engine designed for developers and power users who need to navigate vast local knowledge bases. It leverages a three-tiered retrieval architecture: BM25 for traditional keyword matching, vector embeddings for semantic similarity, and reranking for precision. By operating locally, qmd ensures that your private files remain on your machine while providing enterprise-grade search capabilities directly within the terminal or through an MCP (Model Context Protocol) interface.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/instant-picture/qmd-1-0-0 Ensure you have Ollama running at the default address (http://localhost:11434) to power the embedding and reranking models. The index is stored locally in ~/.cache/qmd.

Use Cases

• Project Documentation Search: Index sprawling repositories of Markdown documentation to find specific technical details across hundreds of files.
• Knowledge Management: Create a personal wiki index where semantic search retrieves notes by intent rather than just keywords.
• System Auditing: Use the MCP mode to allow OpenClaw agents to read and retrieve context from your local source code folders, enabling smarter code assistance.
• Offline Research: Analyze large datasets or text archives without relying on cloud-based search APIs, ensuring data privacy and speed.

Example Prompts

1. "Index the current project directory using the collection name 'dev-docs' and include only markdown files."
2. "Search for the semantic meaning of 'authentication middleware' across my notes using hybrid search."
3. "Retrieve the contents of the 'api-ref.md' file starting at line 10 for 40 lines to help me debug the request header issue."

Tips & Limitations

• Performance: For large directories, use the --mask parameter effectively to ignore node_modules or binary files, which will significantly speed up indexing and reduce noise.
• Ollama Dependency: qmd relies on Ollama. If search latency is high, ensure the model loaded in Ollama is efficient (e.g., a lightweight embedding model).
• Limitations: As a local tool, index updates are not automatic. Remember to run qmd update whenever significant changes are made to your source files. The accuracy of semantic search is dependent on the quality of the embedding model configured in your environment.