Cognee
used in
0
stacks
Open-source memory platform that gives AI agents persistent long-term memory via combined knowledge graphs and vector embeddings.
Apache 2.0 licensed Python framework that ingests documents, relational data, and system context, then models them into a knowledge graph with vector embeddings so agents can retain memory across sessions. Exposes core operations for remembering, recalling, forgetting, and improving memory, and combines vector search with graph-based retrieval. Integrates with Claude Code, LangGraph, CrewAI, n8n, and others, including via an MCP server. Offered as self-hosted open source or as paid Cognee Cloud plans starting at $35/month.
Strengths
- Combines graph and vector storage for relationship-aware retrieval beyond similarity search
- Provenance tracking on ingested context from files, APIs, and databases
- First-party integrations with agent tooling (Claude Code, LangGraph, MCP server)
- Pluggable backends for graph, vector, and relational storage
Tradeoffs
- LLM-driven graph construction adds cost and latency to the ingestion pipeline
- Younger project with a fast-moving API surface compared to mature retrieval stores
- Graph-plus-vector architecture introduces more moving parts than a plain vector memory