Graphiti, built by Zep, is an open-source Python framework for constructing and querying real-time knowledge graphs tailored to AI agent memory. Unlike static RAG pipelines, it continuously integrates new episodic and semantic data, maintains temporal awareness of when facts were true, and supports fast, precise retrieval without batch recomputation. Apache-2.0 licensed, 29,000+ GitHub stars.
Graphiti is an open-source Python framework from Zep for building and querying knowledge graphs that serve as memory for AI agents. You feed it episodes, meaning conversations, documents or events, and it extracts entities and the relationships between them into a graph you can query. The part that separates it from a normal index is time. Facts carry temporal information, so the graph records not just that something is true but when it became true and when it stopped being true, which lets an agent answer questions about the present without losing the history behind them. Ingestion is incremental, so new information updates the graph as it arrives rather than triggering a rebuild of the whole store. It is a library you run inside your own application against a graph database backend, not a hosted product, which means the memory layer stays in your infrastructure.
This is for the point where an agent needs to remember across sessions and a vector store has started to hurt. The symptoms are recognizable. Contradictory facts both come back as relevant because embeddings cannot tell which one is current, user preferences change and the old ones keep resurfacing, and relationships between people, accounts and objects are flattened into unrelated chunks of text. Graphiti suits assistants with long-running user relationships, support agents that track an account over months, and anything where the question is not just what is true but what changed. It is more machinery than a short-lived agent or a one-shot document question needs.
Graphify builds graphs as well, but over a codebase, deriving structure from parsed source so an agent can answer questions about how a repository fits together. Graphiti is about accumulated knowledge rather than one static artifact. Claude Mem targets memory for a specific coding agent workflow, which is narrower and much less work to adopt than a framework you build into an application. LLM Wiki is the lightest of the group, a pattern where the model keeps a set of linked markdown files current instead of maintaining a database. Pick Graphiti when memory is a product requirement rather than a convenience.
Compare with: Graphify, Claude Mem, LLM Wiki
The catalog records this as free, and the project is Apache-2.0 licensed open source. There is no vendor tier to compare. Running it in production is not free of cost, though. You supply the graph database it writes to, and ingestion calls a language model to extract entities and relationships, so the real bill is model usage plus whatever the database costs to host. Zep, the company behind it, sells a managed memory service separately, so check that offering if you would rather not operate the pieces yourself.
No notes have been added for this tool yet.
No prompting tips have been added yet.
No reviews yet. Be the first!