Cognee
Knowledge-graph memory for agents: Claude Code and Codex plugins, an MCP server and a Python library
Install
claude plugin marketplace add topoteretes/cognee-integrations
claude plugin install cognee-memory@cogneeThen configure local or remote memory per the plugin guide.
This is third-party code. Review the repository files before installing.
What it does
Cognee turns documents, code and conversations into a graph of entities and relationships plus a vector index, and the agent retrieves context from it. The core operations are remember, recall with automatic search routing, improve from feedback and forget. Session memory acts as a fast cache and moves useful lessons into the permanent graph in the background. It connects to Claude Code and Codex through a plugin with hooks, to Cursor, Cline and other clients through an MCP server, and to your own apps via Python, TypeScript and REST APIs.
Who it is for. For agent builders and teams who need self-hosted memory that keeps relationships between facts.
Good fit when
- The agent must remember project decisions and fixes across sessions
- You want a team knowledge base from docs, tickets and code
- Relationships between entities matter, not just similar text chunks
Not a fit when
- A simple notes file for the agent is enough
- You cannot pay for LLM calls during processing or run a local model
Example request
Remember why we dropped Redis in this service and recall it when caching comes up againLimitations
Uses OpenAI for the language model and embeddings by default; other providers or local Ollama are alternatives. Processing and answers consume model calls. The Claude Code and Codex plugin lives in the separate cognee-integrations repository. Storing the graph in Postgres is marked as a demo, with the production version licensed separately.
How to disable. Uninstall with claude plugin uninstall cognee-memory or remove the MCP server from the client config. Delete data with the forget operation or by removing the storage directory.
Security check
- Sends ingested data to the chosen LLM provider
- Plugin hooks capture session content automatically
README in short
The README covers use cases: a company brain, agent memory across runs and memory grounded in a domain ontology. The quick start installs the package with uv and shows a keyless demo and a remember and recall example. The agent section gives plugin install commands for Claude Code and Codex and links the MCP guide for Cursor and Cline. It also covers Docker Compose deployment, a single-Postgres mode and BEAM benchmark results with methodology caveats. Apache 2.0 licensed.
FAQ
Can I try it without an API key?
Yes, cognee-cli demo builds a demo graph without a key.
How do I run the API, UI and MCP together?
From a source checkout with docker compose --profile ui --profile mcp up; MCP listens on port 8001.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
Anthropic's coding agent for the terminal, IDE and GitHub: understands your codebase, runs tasks and handles git
OpenAI's coding agent that runs locally in your terminal, with IDE and desktop versions