Cognee Gives AI Agents the Memory They've Been Missing
Founded in Berlin in 2024, Cognee built what it calls a memory engine
Berlin-based startup Cognee has raised a $7.5 million seed round to solve a deceptively simple problem: AI agents have no memory. The round was led by Pebblebed, the fund run by Pamela Vagata, co-founder of OpenAI, and Keith Adams, founder of Facebook AI Research Lab, with participation from 42CAP and Vermilion Ventures, plus angel investors from Google DeepMind, n8n, and Snowplow. The company shared the news in a press release.
Think of it this way: every time you start a new session with an AI agent, it wakes up with complete amnesia. Teams have been patching this gap with makeshift combinations of vector stores, RAG pipelines, and rule engines, and still ending up with hallucinations and shallow outputs. Cognee takes a different approach.
Founded in Berlin in 2024, Cognee built what it calls a memory engine. Its ECL pipeline (Extract, Cognify, Load) pulls in data from over 38 sources, organizes it into a structured knowledge graph, and makes it searchable. A feedback layer called "memify" then sharpens that graph over time based on how agents use it, so the system genuinely learns and improves with each interaction.
The results so far are hard to ignore. Pipeline volume jumped from 2,000 to over one million runs in a single year, a 500x increase. More than 70 companies are now running Cognee in production, including Bayer for scientific research workflows and the University of Wyoming for policy document analysis. The open-source project has cleared 12,000 GitHub stars.
With the new funding, the team plans to launch a cloud platform, build a Rust-based engine for edge devices, and ship 30+ new data connectors in the first half of 2026.
Thinking about launching a business, setting up premises, or creating partnerships in Berlin?
We invite you to contact us.
Our team is ready to help you with our free consultation services.