Cary Greenwood
Back to notes
QMD Search-Before-Read Pattern
QMD Search-Before-Read Pattern

QMD Search-Before-Read Pattern

Source: QMD skill installation Date: 2026-01-29 Tags: #qmd #tokens #optimization #clawdbot

Summary

QMD (Quick Markdown Search) enables targeted retrieval from markdown files, reducing token usage by 70-95% compared to loading full documents into context.

Key Points

  • Full document load: ~3,500 tokens for a 1,500-word doc
  • QMD search snippet: ~150 tokens for the relevant section
  • Savings: 95% fewer tokens on retrieval tasks

Collections Configured

  • obsidian → Cary’s Obsidian vault (227 docs)
  • clawd → Abel’s workspace (54 docs)

Usage Pattern

# Always search first
qmd search "topic" -n 5

# Then fetch only what's needed
qmd get "path/to/file.md"

When to Use

  • Cary asks about something in his notes
  • Cron jobs need context from Obsidian
  • Looking up past decisions, plans, research
  • Any time you’d otherwise load a full markdown file
  • qmd search → Files on disk (Obsidian vault)
  • memory_search → Abel’s conversation memory

Real Cost Impact

  • 10 searches/day with QMD vs full loads = ~$0.50/day saved
  • Compounds across cron jobs, research, conversations

Action Items

  • Installed QMD via Bun
  • Indexed Obsidian vault and clawd workspace
  • Added mandatory rule to AGENTS.md
  • [[AGENTS.md]] - QMD Search Before Read section
  • [[TOOLS.md]] - QMD configuration details