Memory & evals · Project
Graphiti
Graphiti builds temporal knowledge graphs from streams of changing information.
+ FollowBuilder facts
- Fact history. Edges carry validity windows and source episode provenance. An invalidated relationship can remain available for historical queries. Source ↗
- Retrieval. Episodes update the graph incrementally. Search combines semantic, BM25 and graph retrieval, and optional Pydantic schemas define custom entity and edge types. Source ↗
- Dependencies. The Python library supports Neo4j, FalkorDB and Amazon Neptune. Neptune needs OpenSearch, and the default inference and embedding configuration uses OpenAI. Source ↗
- Access. Graphiti is self-managed and uses infrastructure and model access supplied by the operator. Zep is a separate managed or bring-your-own-cloud service. Source ↗
Sources checked
Publisher update ·
Graphiti 0.31.2 ↗Removing an episode now preserves facts still supported by other episodes and rebuilds affected entity summaries from surviving facts.
Change timeline
0 storiesNo verified releases in this timeline yet.