Graphiti Free

Graphiti is an open-source engine that builds a time-aware knowledge graph from your data, so AI agents can track how facts change instead of overwriting them with the latest vector.

What is Graphiti?

Graphiti comes from the team behind Zep. It is a Python framework that ingests text and structured data straight into a graph of entities and relationships, without batch reprocessing. Every edge and node carries validity timestamps, so a question like “what did the price used to be?” has a real answer.

It is a library, not a SaaS product. You run it yourself and back it with Neo4j or FalkorDB.

Graphiti features that matter

• Bi-temporal model: track both when an event happened and when you learned about it.
• Incremental updates: new data merges into the graph in place, no full rebuilds.
• Hybrid retrieval: combines semantic embeddings, BM25 keyword search and graph traversal.
• Graph backends: Neo4j by default, with FalkorDB and Kuzu support.
• MCP server: plug the graph into Claude or Cursor as a tool.

Product highlights

Most memory layers flatten everything into embeddings and lose the timeline. Graphiti keeps the timeline, which is why it is popular for agents that need to reason about history — CRM assistants, ops copilots, anything where “when” changes the answer.

The cost is operational: you manage a graph database, and extraction quality depends on the LLM you point it at. Cheap models produce sloppy graphs.

Who should use Graphiti

Best for backend engineers comfortable running Neo4j who need temporal reasoning. Not for no-code users, and overkill if your app just needs to remember a username.

Common use cases

• Sales copilots that track how a deal’s status changed over months.
• Personal assistants that remember evolving preferences.
• Incident-response bots linking past outages to current symptoms.
• Research agents building citation graphs from papers.
• Compliance tools that need an audit trail of who knew what, when.

Pricing and limits

Graphiti itself is free and Apache-2.0 licensed. Your real costs are Neo4j hosting and the LLM calls used for extraction and search. Zep’s hosted product is the paid alternative if you don’t want to run it.

Quick Summary

• Graphiti is a free, open-source temporal knowledge graph library from the Zep team.
• It tracks fact validity over time, unlike plain vector stores.
• Requires a graph DB (Neo4j or FalkorDB) and an LLM for extraction.
• Hybrid retrieval blends embeddings, keyword search and graph hops.
• Best for engineers who need historical reasoning, not casual users.