Graphiti

**Judgement:** Graphiti is a focused and technically solid skill for interacting with a local/remote Graphiti service backed by Neo4j and Qdrant. It is well-suited for users who already operate a knowledge graph stack and want an agent to search and persist information into that graph. **What it does well:** - Offers two clear capabilities: fact search (`/facts/search`) and episode creation (`/messages`). - Integrates with existing infrastructure via REST and supports dynamic endpoint discovery (Clawdbot config → `GRAPHITI_URL` env var → `http://localhost:8001` fallback). - Documentation includes concrete curl examples, making it straightforward to validate the service and debug outside the agent. **Key risks / limitations:** - Requires a correctly deployed Graphiti service plus Neo4j and Qdrant; setup is non-trivial and not handled by the skill itself. - No mention of authentication, access control, or encryption; if Graphiti is exposed beyond localhost, there is risk of unauthorized reads/writes to the knowledge graph. - Persistent storage of “episodes/memories” means sensitive data could be retained longer than intended if the agent is not carefully configured. - Reliability and performance are completely dependent on the external Graphiti/DB stack; the skill adds minimal resilience. **Recommended scenarios:** - Teams already running Neo4j + Qdrant + Graphiti who want their agents to: - Search an internal knowledge graph for grounded facts. - Persist conversation events, project updates, or decisions as structured episodes. - Experimental or research setups exploring graph-augmented LLMs where the environment is controlled and security can be managed at the infrastructure level. **Less suitable for:** - Users without existing graph infrastructure or who need a turnkey, hosted knowledge base. - High-security or multi-tenant environments unless Graphiti is deployed behind strong network and auth controls.