qmd

qmd is an on-device hybrid search engine designed specifically for Markdown files. It was created by Tobi Lütke, CEO of Shopify, and is used as a memory and search backend for agents working within the LLM Wiki pattern. [llm-wiki.en.md]

qmd combines three retrieval techniques into a single pipeline:

• BM25 — a classical keyword-based ranking algorithm
• Local vector embeddings — semantic similarity search performed on-device
• LLM re-ranking — a language model re-orders results for relevance

This hybrid approach gives it stronger retrieval quality than any single method alone. [llm-wiki.en.md]

Within the LLM Wiki pattern, navigation at small-to-moderate scale is handled by a plain index.md catalog — no external infrastructure required. However, as a wiki grows beyond hundreds of pages, the index-first approach reaches its limits. qmd fills that gap, providing real search infrastructure once the flat index file alone is no longer sufficient. [llm-wiki.en.md]

It is part of the broader tool stack that has emerged around the LLM Wiki pattern:

[llm-wiki.en.md]

• On-device — search runs locally, keeping data private
• Markdown-native — built for the file format the LLM Wiki uses
• Agent-accessible — designed to serve as a memory layer for LLM agents, not just human users

[llm-wiki.en.md]

qmd is not a full RAG pipeline — it does not replace the compiled, interlinked wiki pages that define the LLM Wiki pattern. Instead, it augments the pattern's navigation layer when the wiki grows large enough that a simple index file is insufficient. [llm-wiki.en.md]

• GitHub: https://github.com/tobi/qmd [llm-wiki.en.md]