OpenClaw Setup Guide
Knowledge Management

QMD Local Search

Semantic and keyword search across all your markdown files

QMD is a local search engine for markdown files. Three search modes:

  • BM25 keyword search (qmd search) -- fast, no model needed
  • Semantic vector search (qmd vsearch) -- finds conceptually related content even without keyword overlap
  • Hybrid search (qmd query) -- both combined with reranking, best quality

My agent uses it constantly -- finding related notes for cross-linking, searching memory, answering questions about vault content.

Setup

bun install -g github:tobi/qmd

Setup Prompt

Set up QMD collections for searching my files:

1. A "workspace" collection pointing to the OpenClaw workspace directory
2. A "memory" collection pointing to the memory subdirectory
3. A "vault" collection pointing to my knowledge vault

Run `qmd update` to index all files, then `qmd embed` to generate
embeddings for semantic search.

Commands reference:
- qmd search "query" -- keyword search (fast)
- qmd vsearch "query" -- semantic search (needs embeddings)
- qmd query "query" -- hybrid search + reranking (best quality)
- qmd update -- re-index after changes
- qmd embed -- update embeddings after changes

Set up a nightly cron job at 2 AM to run `qmd update && qmd embed`
so the index stays fresh.

Models

QMD downloads embedding models on first use (~329MB). The query expansion and reranker models (~1.9GB total) download on first qmd query or qmd vsearch. First run will be slow -- after that it's fast.

On this page