graphiti

What This Skill Does

The Graphiti skill provides an advanced interface for OpenClaw to interact with a persistent knowledge graph. By leveraging Neo4j for structured data storage and Qdrant for semantic vector search, it allows the agent to move beyond simple context windows. This skill handles the ingestion of unstructured data as 'episodes,' automatically extracting entities and relationships to build a long-term memory store. When searching, the agent performs intelligent retrieval to pull relevant facts that inform its current reasoning, making it ideal for tasks that require long-term context retention across multiple sessions.

Installation

To integrate this skill into your environment, use the OpenClaw hub command:

clawhub install openclaw/skills/skills/emasoudy/graphiti

Before usage, ensure your infrastructure is ready. You must have a Neo4j database and a Qdrant instance active. The Graphiti service must be running, typically at http://localhost:8001. You can customize the connection by setting the GRAPHITI_URL environment variable or by updating the configuration via clawdbot config set skills.graphiti.baseUrl "<your-url>".

Use Cases

• Project Management: Track complex project histories, stakeholder relationships, and meeting decisions that span weeks or months.
• Personal CRM: Build a detailed map of contacts, their preferences, and previous interactions to provide personalized assistance.
• Research Assistant: Catalog findings from various documents and web searches into a unified graph, allowing the agent to perform 'deep connections' across different research topics.
• Contextual Recall: Enable the agent to answer questions like 'What did we decide about the API migration last time?' by querying the knowledge base.

Example Prompts

1. "Check the knowledge graph for any previously mentioned blockers regarding the authentication module development."
2. "Add a new memory: I had a call with the infrastructure team today; they confirmed the Kubernetes cluster upgrade is scheduled for next Tuesday."
3. "Summarize all known relationships between the main project stakeholders currently stored in the graph."

Tips & Limitations

• Data Privacy: Ensure that your Neo4j and Qdrant instances are secured, as this skill stores all ingested memory in plain text within those databases.
• Semantic Quality: The quality of search results is highly dependent on how clearly episodes are written. Providing descriptive titles helps with graph indexing.
• Performance: For extremely large datasets, ensure your Qdrant vector index is configured for scale to prevent latency spikes during agent response times.