cognee adds skills system, knowledge graph memory

Standard RAG is showing its age — cognee's bet that knowledge graphs beat top-k chunk retrieval for agent memory is gaining serious traction with 13,600+ stars and a $7.5M seed.

• –The `cognify` pipeline extracts entities and relationships via LLM, commits them to a graph store (Kuzu/Neo4j), and keeps vector embeddings in sync — enabling 14 retrieval modes including structural graph traversal
• –The new skills system (v0.5.4–0.5.5) lets agents define self-improving workflows with LLM-based scoring and automatic amendment loops — a step toward agents that optimize their own memory strategies
• –Supports 29+ vector databases and 12+ framework integrations, making it drop-in for most existing stacks
• –70+ companies in production including Bayer and dltHub, running over 1 million pipelines/month — this isn't vaporware
• –MCP server support means cognee can plug directly into Claude, Cursor, and other MCP-compatible tools out of the box