Build personalized AI agents that learn from dynamic data
Graphiti is Knowledge Graph-based memory for AI agents. Automatically build rich graphs from changing business data & chat histories. Enable your Python agent with fast access to relevant, accurate data, even as it evolves over time. Visit our GitHub repo!
Graphiti tracks provenance: every entity and relationship traces back to episodes (raw ingested data). Users can audit why an agent made a decision by following the chain from derived fact to original source.
Graphiti provides temporal context graphs where every fact has a validity window (when it became true, when it was superseded). Agents can query what is true now or what was true at any point in history, enabling context-aware decisions based on evolving data.
Graphiti supports incremental graph construction. New data integrates immediately without recomputing the entire graph. The context graph evolves in real-time as episodes are ingested, suitable for interactive, production AI applications.
Graphiti builds entity-relationship graphs with summaries, facts, and episodes (raw source data). Agents retrieve rich structured context via hybrid search (semantic + keyword + graph traversal) instead of unstructured text, improving precision and relevance.
Graphiti supports both prescribed ontology (define types upfront via Pydantic models) and learned ontology (let structure emerge from data). Developers can start simple and evolve as patterns appear, avoiding rigid one-size-fits-all schemas.