qmd-memory

// **Author:** As Above Technologies

$ git log --oneline --stat

stars:1,933

forks:367

updated:March 4, 2026

SKILL.mdreadonly

QMD Memory Skill for OpenClaw

Local Hybrid Search — Save $50-300/month in API Costs

Author: As Above Technologies Version: 1.0.0 ClawHub: [Coming Soon]

💰 THE VALUE PROPOSITION

API Costs You're Paying Now

Operation	API Cost	Frequency	Monthly Cost
memory_search (embedding)	$0.02-0.05	50-200/day	$30-300
Context retrieval	$0.01-0.03	100+/day	$30-90
Semantic queries	$0.03-0.08	20-50/day	$18-120
TOTAL			$78-510/month

With QMD Local

Operation	Cost	Why
All searches	$0	Runs on your machine
Embeddings	$0	Local GGUF models
Re-ranking	$0	Local LLM

Your savings: $50-300+/month

One-time setup. Forever free searches.

🚀 QUICK START

# Install the skill
clawhub install asabove/qmd-memory

# Run setup (installs QMD, configures collections)
openclaw skill run qmd-memory setup

# That's it. Your memory is now supercharged.

WHAT YOU GET

1. Automatic Collection Setup

Based on your workspace structure, we create optimized collections:

✓ workspace     — Core agent files (MEMORY.md, SOUL.md, etc.)
✓ daily-logs    — memory/*.md daily logs
✓ intelligence  — intelligence/*.md (if exists)
✓ projects      — projects/**/*.md (if exists)
✓ documents     — Any additional doc folders you specify

2. Smart Context Descriptions

We add context to each collection so QMD understands what's where:

qmd://workspace    → "Agent identity and configuration files"
qmd://daily-logs   → "Daily work logs and session history"
qmd://intelligence → "Analysis, research, and reference documents"

3. Pre-configured Cron Jobs

# Auto-update index (nightly at 3am)
0 3 * * * qmd update && qmd embed

# Keep your memory fresh without thinking about it

4. OpenClaw Integration

Memory search now uses QMD automatically:

memory_search → routes to QMD hybrid search
memory_get → retrieves from QMD collections
Results include collection context

5. Multi-Agent MCP Server (Optional)

# Start shared memory server
openclaw skill run qmd-memory serve

# All your agents can now query collective memory
# Forge, Thoth, Axis — shared knowledge base

📊 SEARCH MODES

Mode	Command	Best For
Keyword	`qmd search "query"`	Exact matches, fast
Semantic	`qmd vsearch "query"`	Conceptual similarity
Hybrid	`qmd query "query"`	Best quality (recommended)

Example Queries

# Find exact mentions
qmd search "Charlene" -n 5

# Find conceptually related content
qmd vsearch "how should we handle customer complaints"

# Best quality — expansion + reranking
qmd query "what decisions did we make about pricing strategy"

# Search specific collection
qmd search "API keys" -c workspace

🔧 CONFIGURATION

Add Custom Collections

openclaw skill run qmd-memory add-collection ~/Documents/research --name research

Add Context

openclaw skill run qmd-memory add-context qmd://research "Market research and competitive analysis"

Refresh Index

openclaw skill run qmd-memory refresh

💡 TEMPLATES

Trading/Investing Workspace

openclaw skill run qmd-memory template trading

Creates:

intelligence — Trading systems, dashboards, signals
market-data — Price history, analysis
research — Due diligence, reports
daily-logs — Trade journal

Content Creator Workspace

openclaw skill run qmd-memory template content

Creates:

articles — Published content
drafts — Work in progress
research — Source material
ideas — Brainstorms, notes

Developer Workspace

openclaw skill run qmd-memory template developer

Creates:

docs — Documentation
notes — Technical notes
decisions — ADRs, architecture decisions
snippets — Code snippets, examples

📈 COST SAVINGS CALCULATOR

Run this to see your estimated savings:

openclaw skill run qmd-memory calculate-savings

Output:

Your Current API Memory Costs (estimated):
  memory_search calls/day:     ~75
  Average cost per call:       $0.03
  Monthly API cost:            $67.50

With QMD Local:
  Monthly cost:                $0.00

YOUR MONTHLY SAVINGS:          $67.50
YOUR ANNUAL SAVINGS:           $810.00

ROI on skill purchase:         40x (if skill was $20)

🛠️ TECHNICAL DETAILS

Models Used (Auto-Downloaded)

Model	Purpose	Size
embeddinggemma-300M-Q8_0	Vector embeddings	~300MB
qwen3-reranker-0.6b-q8_0	Re-ranking results	~640MB
qmd-query-expansion-1.7B-q4_k_m	Query expansion	~1.1GB

Total: ~2GB (one-time download)

System Requirements

Node.js >= 22
~3GB disk space (models + index)
~2GB RAM during embedding (then minimal)

Where Data is Stored

~/.cache/qmd/
├── index.sqlite      # Search index
├── models/           # GGUF models
└── mcp.pid           # MCP server PID (if running)

🤝 SUPPORT

Questions?

GitHub Issues: github.com/asabove/qmd-memory-skill
Discord: As Above community
Email: support@asabove.tech

Found it valuable?

Star us on ClawHub
Share with other OpenClaw users
Subscribe to our newsletter for more agent optimization tips

📜 LICENSE

MIT — Use freely, modify as needed.

QMD itself is created by Tobi Lütke (github.com/tobi/qmd). This skill provides easy OpenClaw integration.

"Stop paying for memory. Start compounding knowledge."

As Above Technologies — Agent Infrastructure for Humans