Graphiti is an open-source framework with 32,000+ stars on GitHub for building real-time, temporal knowledge graphs that give AI agents accurate long-term context.
Most retrieval systems flatten the world into document chunks and forget that facts change. Graphiti, built by the team behind Zep, does the opposite: it constructs a knowledge graph that evolves with every interaction, tracking what is true now and what was true before, with full provenance back to the source data. Agents can then query across time, meaning and relationships using hybrid retrieval that combines semantic search, keyword search and graph traversal. It ingests structured and unstructured data continuously, making it a strong memory and context layer for production agents that must stay correct as users, businesses and facts change. The framework is Apache-2.0 licensed and actively developed, updated in October 2026.
Graphiti’s temporal model solves a subtle but serious failure mode of agent memory: when a user changes jobs, moves cities or revises a decision, naive memory systems keep asserting the old fact forever. Graphiti instead records when each fact was true, so agents can answer in the present tense correctly and still explain the past. It accepts everything from chat transcripts to JSON feeds, extracts entities and relationships automatically, and has become a reference design for the fast-growing field of agent memory infrastructure.
Graphiti is free and open source under the Apache License 2.0. Zep, the managed memory platform built on Graphiti, is a paid cloud service — see getzep.com for plans.
Developers building agents that need trustworthy, up-to-date memory — assistants, support agents and enterprise copilots where stale or contradictory facts are costly.