qmd

qmd is a specialized local search engine designed for developers and knowledge workers to navigate Markdown-heavy projects efficiently. By leveraging hybrid search techniques—combining fast BM25 keyword matching with semantic vector similarity—it allows users to pinpoint relevant information within large documentation sets or codebases before performing full file reads. This approach provides significant token savings during Claude Code sessions, often reducing context overhead by up to 90% by enabling targeted retrieval of specific file segments.

• Performs rapid keyword-based discovery using BM25 for immediate results.

• Offers semantic similarity search via vsearch as a fallback for conceptual queries.

• Supports hybrid retrieval and reranking for high-precision information extraction.

• Enables filtering by project collections, file path discovery, and JSON output for automated pipelines.

• Provides deep-link retrieval via file paths or unique document IDs, allowing seamless integration into AI-assisted coding workflows.

• Monitors index health and supports incremental updates, ensuring search results stay synchronized with changing documentation.

• Always prefer qmd search over reading files directly to optimize token usage when exploring new repositories.

• Use the --files flag to identify relevant files without loading full content into context.

• Utilize specific queries or target collections with the -c flag to improve result accuracy.

• Rely on semantic vector search for exploratory or 'how-to' questions when keyword matching fails to capture intent.

• Integrate with other Simba components to maintain consistency across project documentation and rule enforcement. Suitable for complex system documentation, API reference maintenance, and large-scale codebase knowledge navigation.