Graphiti is a framework for building temporal knowledge graphs that power AI agents.
Graphiti addresses the limitation of static knowledge graphs and traditional retrieval-augmented generation by creating context graphs that track how facts evolve over time. Rather than storing flat document chunks, the framework represents entities, relationships, and facts with temporal validity windows—recording when information became true and when it was superseded. Every derived fact maintains provenance back to source episodes, the raw data that produced it. The framework continuously integrates user interactions, structured and unstructured enterprise data, and external information into a coherent, queryable graph. It supports incremental updates without requiring complete recomputation, enabling efficient retrieval and precise historical queries through hybrid search combining semantic, keyword, and graph traversal methods.
Graphiti suits teams building interactive, context-aware AI applications that operate on evolving, real-world data where understanding how facts change matters as much as knowing current state. It is particularly valuable for agents that need to maintain rich, structured context across multiple interactions and data sources. The framework uses Pydantic models for developer-defined entity and relationship types, allowing customization of ontology rather than enforcing a prescribed schema. This approach differs from traditional knowledge graphs by making temporal tracking and provenance central rather than optional features.
The project shows consistent development activity with regular commits and ongoing refinement of core functionality. The maintainers actively engage with the community through documentation and are responsive to issues raised by users. Development focuses on improving the efficiency of graph operations and expanding the capabilities for handling diverse data sources and query patterns.
