agent-memory

What This Skill Does

The agent-memory skill provides an enterprise-grade cognitive architecture for OpenClaw agents, integrating a multi-layered storage stack. It combines Mem0 for semantic, long-term memory (backed by Qdrant for vector search and Neo4j for graph-based entity relationships) with a robust SQLite database for structured data management. This skill transforms a stateless agent into a persistent entity capable of recalling past interactions, understanding complex entity relationships, and maintaining organized business data like tasks, project statuses, and bookmarks. By including Langfuse integration, it also provides complete observability over the agent's cognitive processes, ensuring developers can monitor and debug how memory is retrieved and synthesized over time.

Installation

To integrate this brain into your agent, use the ClawHub CLI command: clawhub install openclaw/skills/skills/aiwithabidi/agent-memory-pro. After installation, ensure your environment is configured for the required backends (Qdrant, Neo4j). Run bash {baseDir}/scripts/setup_brain.sh inside your container to install all necessary Python dependencies. For initial bootstrapping, you can run python3 {baseDir}/scripts/seed_mem0.py to populate your agent with core facts and organizational knowledge immediately upon deployment.

Use Cases

This skill is ideal for agents acting as long-term personal assistants or business analysts. Use it to store client preferences, manage project lifecycles within SQLite, or track the evolution of a business strategy across multiple sessions. It is perfect for teams requiring data persistence that survives container restarts, or for agents that need to perform complex reasoning across interconnected pieces of information, such as linking a contact to a specific project and an upcoming deadline.

Example Prompts

1. "Store in memory that my primary project is 'Project Phoenix' and the current status is 'in-progress'."
2. "Search through all my past notes and tell me what the client mentioned about budget constraints for the voice AI project."
3. "List all active projects currently stored in the structured database and provide a summary of their current status."

Tips & Limitations

To maintain optimal performance, regularly audit the knowledge graph to remove deprecated relationships that may confuse the retrieval model. While the system handles automated deduplication, manual cleanup via memory_engine.py is recommended for high-volume environments. Note that this skill requires local or cloud instances of Qdrant and Neo4j; ensure these services are active before triggering memory-intensive operations to avoid latency or failed lookups. Always use clear, distinct facts when storing information to improve semantic recall accuracy.