GRAPHYou commented this keyword
Graphiti
An open-source Python framework for building knowledge graphs that AI agents can query, with an MCP server so assistants can use it.
- Python 3.10+
- A graph database (Neo4j 5.26 or FalkorDB 1.1.2)
- OpenAI API key
- Setup
- 30 min
- Level
- Advanced
- Stars
- 31.5k
- License
- Apache-2.0
What it does
Graphiti builds a context graph: a map of people, products and ideas, the facts that connect them, and when each fact was true. When something changes, the old fact is marked out of date instead of deleted, so you can ask what is true now or what was true before.
It is a developer framework, made by the team behind Zep. It includes an MCP server, which is a connector that lets AI assistants add to and search the graph.
Install it
- Install Python 3.10 or newer.
- Set up a graph database. The simplest is Neo4j Desktop from neo4j.com/download. Or run FalkorDB in Docker with the command in the second block.
- Set your OPENAI_API_KEY in your environment. Graphiti uses OpenAI by default. Anthropic, Gemini and Groq are also supported.
- Install the Python package with the first command below.
- Work through the Quickstart example in the repo's examples/quickstart folder.
Install Graphiti
pip install graphiti-core
Optional: start FalkorDB with Docker
docker run -p 6379:6379 -p 3000:3000 -it --rm falkordb/falkordb:latest
Use it
- Follow the Quickstart to connect to your database, add episodes (raw text or JSON) and search the graph.
- To connect it to an AI assistant, use the MCP server in the repo's mcp_server folder. Its own README has the setup steps; it can run with Docker and Neo4j.
Good to know
The main README does not give Claude-specific install steps. Those are in the MCP server README.
Graphiti works best with model providers that support structured output, such as OpenAI, Anthropic and Gemini. If you hit rate-limit (429) errors, lower the SEMAPHORE_LIMIT setting, which defaults to 10.
Want the next one first?
One new guide a week, straight to your inbox.
