Cognee

What is Cognee?

Cognee is an open-source memory platform for AI agents. It captures context from documents, data warehouses, chats, APIs, and other sources, converts that information into a recallable memory layer, and lets agents retrieve it across sessions. The memory engine combines graph, vector, and relational approaches so agents can preserve facts, relationships, decisions, and project history rather than starting from an empty context each time.

Developers can run Cognee locally through its open-source package or use Cognee Cloud for managed operation. The platform provides an SDK, MCP server, and integrations for coding and agent environments including Claude Code, Codex, Cursor, LangGraph, OpenClaw, and CrewAI. Teams can customize ontologies and data models and apply permission controls.

How to Use Cognee

1. Install the open-source Cognee package locally or create a Cognee Cloud workspace.
2. Connect the documents, warehouses, chats, APIs, or other sources the agent should remember.
3. Integrate the memory layer with an agent through the SDK, MCP server, or a supported first-party integration.
4. Store and recall context across sessions, then move to managed or private deployment as the workload grows.

Core Features

• Persistent agent memory: Retains context and makes it available across separate agent sessions.
• Graph, vector, and relational engine: Represents facts and relationships through multiple complementary memory structures.
• Data-source adapters: Connects documents, warehouses, chats, APIs, and common business tools.
• Single memory API: Gives different agents a consistent interface for reading and writing memory.
• Agent integrations: Supports MCP and integrations with coding agents and agent frameworks.
• Customization and governance: Provides custom ontologies, data models, and permission controls.

Use Cases

• Coding-agent memory: Agents can retain past implementation decisions, fixes, and project context.
• Research memory: Research agents can recall sources, entities, and connected facts across sessions.
• Sales intelligence: Teams can make deals, customer profiles, and account context queryable by agents.
• Technical knowledge bases: Organizations can organize manuals, documentation, and industrial knowledge for retrieval.
• Customer-facing agents: Product teams can give domain-specific agents long-term, source-aware context.

Pricing

Free costs $0 per month and includes one workspace, one million tokens, unlimited users, unlimited API calls, and agent integrations. Standard costs $2.50 per one million processed tokens, with additional workspaces at $5 each per month, and adds Slack, Notion, and Google Drive integrations. Enterprise uses custom pricing and adds dedicated support, bring-your-own-cloud deployment, and a support SLA.

Frequently Asked Questions

Is Cognee open source?

Yes. Cognee provides an open-source package that can run locally.

Which agents can use Cognee?

The homepage lists Claude Code, Codex, Cursor, LangGraph, OpenClaw, CrewAI, MCP-compatible agents, and custom integrations.

Can Cognee run in a customer's own cloud?

Yes. Bring-your-own-cloud deployment is listed for the Enterprise offering.