topoteretes/cognee is a free, open source AI memory platform that gives agents long-term memory that persists beyond a session. It is written in Python and released under the Apache-2.0 license. The latest release is v1.6.5.
What this project does is let you reuse something that passed through a conversation once, later on. It converts documents, code and conversations into a self-hostable knowledge graph (knowledge graph) that can be searched and reused. A knowledge graph is a data structure that stores multiple entities and the relationships between them in graph form. The README explains that you can store and search text with local models, without an API key. The repository description also notes that it can build an agent's long-term memory with small models. Four operations handle memory: remember, recall, improve, forget. The repository was created in August 2023 and has collected 32,597 stars.
Still, the verifiable range is narrow. The provided information contains no changelog or release notes for v1.6.5, so there is no way to know what this release changed. There are also no comparison figures showing search accuracy, response speed, or how much memory accumulates. The evaluation scores fill in only the Momentum item, at 6.7, and the remaining three items are not in the material. It remains a candidate worth reviewing for agents that need long-term memory. To decide on adoption, you have to check the release notes yourself and validate in your own environment.