Graphiti
Temporal knowledge graph memory system for agents that need facts, entities, and time-aware context.
Graphiti overview
Graphiti is an open-source temporal knowledge graph system from Zep for building agent memory that tracks entities, relationships, and changing facts over time.
Temporal knowledge graph model
Graphiti represents memory as entities and relationships with time-aware context.
This helps agents reason about changing facts instead of flattening everything into chunks.Agent memory orientation
The project is built for agent and assistant memory workflows.
It focuses on recall structures that matter to agents, not just generic graph storage.Open-source implementation
The repository and docs provide a concrete path for evaluation.
Graph memory is easier to trust when the implementation is inspectable.When to use Graphiti
Customer or user memory
Track evolving relationships, preferences, and events across time.
Research knowledge graphs
Represent entities and links discovered during research workflows.
Agent context retrieval
Retrieve facts by relationship and recency rather than only semantic similarity.
How it compares
Vector search is useful for fuzzy recall; Graphiti is more appropriate when relationships and time matter.
Questions
What should I check before using Graphiti?
Test Graphiti with repeated sessions. Add facts, update them, ask for recall, inspect retrieval behavior, and verify deletion or scoping controls before storing sensitive user or project memory.
Is Graphiti open source?
Graphiti is listed with Apache-2.0 based on the official source links in this profile. Re-check the repository, model card, or docs before production use.
Who should evaluate Graphiti?
Graphiti is most worth evaluating for builders who need graph memory instead of simple chat history.