Cognee
An open-source AI agent memory platform that runs locally on CPU, connects to Claude Code and Codex, and supports self-hosting or managed cloud hosting.
Cognee gives AI agents persistent memory across sessions, connecting documents, code, and conversations in a searchable knowledge graph. It's for developers who want agents to retain project context and teams whose knowledge sits across tickets, discussions, and repositories. The Python package is open source under Apache 2.0.
Memory can run locally on your CPU without an API key. Cognee uses GLiNER to extract entities and relationships, plus a local embedding model; both download on first use. This path processes content on your machine. Optional OpenAI models and embeddings send processing and answer-generation requests to that provider. You can self-host in Docker, on-premises, or in your own cloud, while Cognee Cloud provides managed hosting.
Cognee combines graph, vector, and relational retrieval so agents can find related facts as well as relevant text. Code memory captures symbols and dependencies. Agents can retrieve passages or use an optional local or hosted LLM to generate answers, with citations attached to supporting facts.
Claude Code and Codex plugins preserve context across agent runs. Cursor and other MCP clients can read and write memory through its MCP server, and LangGraph and OpenClaw have integrations. Slack, GitHub, and Linear connectors bring team knowledge into shared memory.
Custom data models and ontologies describe domain entities, relationships, and rules. Session distillation saves accepted lessons as durable knowledge, feedback adjusts later recall, and deletion controls let you remove individual items or datasets.
