A local, high-performance hybrid search and memory engine designed to eliminate API embedding costs for AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install qmd-memory
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install qmd-memory using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is QMD Memory Skill?
The QMD Memory Skill is a powerful infrastructure tool within the Openclaw Skills ecosystem that replaces expensive cloud-based vector searches with a local, private alternative. By utilizing GGUF models for embeddings and re-ranking, it allows your agents to maintain a deep, searchable history of workspaces, logs, and research without recurring API fees. This skill effectively turns your local machine into a high-speed knowledge base, ensuring that context retrieval is both fast and cost-effective.
Built for developers who demand privacy and efficiency, it integrates seamlessly with the existing memory architecture. By moving memory search operations to your local hardware, you can achieve professional-grade semantic retrieval while saving between $50 and $300 per month on token costs associated with external embedding providers. It is an essential component for any serious implementation involving Openclaw Skills.
QMD Memory Skill Use Cases
- Reducing monthly operational costs by moving embedding and search operations from paid APIs to local models.
- Centralizing knowledge across multiple AI agents using a shared MCP memory server.
- Automatically organizing and indexing developer documentation, ADRs, and technical snippets for instant retrieval.
- Maintaining private, local trade journals or research repositories that require semantic search without data leakage.
How QMD Memory Skill Works
- The skill performs an initial scan of your workspace to identify core agent files, logs, and project documentation.
- It automatically downloads optimized local models, including embedding gemma and qwen-reranker, to handle all vector operations locally.
- Collections are established with specific semantic context descriptions so the agent understands the purpose of each data silo.
- Search queries are processed through a hybrid engine that combines keyword matching with semantic similarity and query expansion.
- Pre-configured cron jobs ensure the local index is updated nightly, keeping the agent memory synchronized with your latest file changes.
QMD Memory Skill Setup
To integrate this capability into your workflow, follow these steps using the CLI:
# Install the skill from the hub
clawhub install asabove/qmd-memory
# Execute the setup to download models and configure collections
openclaw skill run qmd-memory setup
# Optional: verify your potential API savings
openclaw skill run qmd-memory calculate-savings
QMD Memory Skill Data Schema & Taxonomy
The skill manages data locally within the ~/.cache/qmd/ directory, ensuring all your Openclaw Skills data remains under your control.
| Component | Purpose | Technical Details |
|---|---|---|
| index.sqlite | Main search index | SQLite database storing text and metadata |
| models/ | Local AI models | ~2GB of GGUF models for embeddings and reranking |
| mcp.pid | Server management | PID file for the optional multi-agent MCP server |
Data is logically partitioned into collections such as workspace for agent identity, daily-logs for session history, and intelligence for research and analysis documents.
QMD Memory Skill Advanced Features
- Multi-Agent MCP Server: Allows a collective knowledge base to be queried by multiple agents simultaneously.
- Automated Query Expansion: Uses a local 1.7B parameter model to improve search accuracy by expanding user queries.
- Specialized Workspace Templates: Instant configuration for trading, content creation, and developer-centric environments.
- Smart Reranking: Implements a local cross-encoder to ensure the most relevant context is always fed to the agent.
- Custom Collection Hooks: Add any local directory to the search index with custom semantic descriptions.
SKILL.md
Loading
Related Openclaw Skills
qmd
A high-performance local search engine designed for indexing and retrieving Markdown notes and documents using keyword, vector, and hybrid search methods.
Clawdbot (Snibet Edition)
A high-performance tweet engine for builders that generates sharp, human-sounding X content without the AI fluff.
Weave
A powerful CLI tool for creating crypto invoices, generating cross-chain payment quotes, and monitoring settlement status.
TikTok Slideshow Creator
Generate high-converting TikTok image carousels with automated text overlays and seamless draft integration via the ViralBaby API.
Hardcover
A GraphQL-based skill for AI agents to query and manage personal reading libraries and book metadata from Hardcover.app.
FFmpeg CLI
A comprehensive media processing toolset for automated video transcoding, editing, and manipulation via the command line.
Featured*
