qmd

What This Skill Does

qmd is a high-performance local search and indexing engine designed for OpenClaw agents, providing a comprehensive solution for managing local knowledge bases. It leverages a multi-layered retrieval architecture that combines BM25 keyword-based search with vector semantic search and sophisticated reranking capabilities. By utilizing local resources, qmd ensures that your private documents remain secure while providing AI agents with the ability to perform deep context retrieval across large file repositories. Whether you are managing technical documentation, personal research notes, or codebases, qmd provides an interface to query, retrieve, and inspect your data with granular control through both CLI and MCP (Model Context Protocol) modes.

Installation

To integrate this tool into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/steipete/qmd Ensure that you have an Ollama instance running locally at the default address (http://localhost:11434) as it acts as the primary provider for the embeddings and reranking processes that power the search intelligence.

Use Cases

• Personal Knowledge Management: Index your entire Obsidian or Markdown vault to query concepts across years of notes using semantic search.
• Project Context Loading: Index specific project folders to allow the agent to reference implementation details or architectural decisions during coding tasks.
• Offline Documentation Retrieval: Quickly access information from local man pages or downloaded technical specifications without needing an internet connection.
• Automated Research: Configure the MCP mode to allow the AI agent to dynamically pull relevant context from your private drive during complex reasoning workflows.

Example Prompts

1. "qmd collection add ~/Projects/MyDocs --name project-docs --mask "**/*.md" and then run qmd update to index everything."
2. "qmd query "How do I implement authentication in the legacy module?" and provide the file contents."
3. "qmd get docs/architecture.md:20 -l 50 to pull the context for the system design section."

Tips & Limitations

• Performance: For massive repositories, the initial qmd update may take time as embeddings are generated. Use the file masking feature to exclude build artifacts or large non-text assets to optimize speed.
• Ollama Dependency: Ensure your local Ollama server has sufficient VRAM to handle the embedding models, otherwise retrieval times may spike.
• Privacy: Since indexing occurs locally, your data does not leave your machine, making this a highly secure solution for sensitive documents.
• Updates: If you add new files to a collection, remember to trigger qmd update to ensure the vector index remains synchronized with the file system state.