Cognee raises $7.5M to build memory layer for AI agents
What's the deal? CogneeDealroom has a profile for this one. Try Dealroom →, a Berlin-based startup building memory infrastructure for AI agents, has raised $7.5M in seed funding. The round was led by PebblebedDealroom has a profile for this one. Try Dealroom →, whose founders include Pamela VagataDealroom has a profile for this one. Try Dealroom → (ex-founding engineer at OpenAI) and Keith AdamsDealroom has a profile for this one. Try Dealroom → (founder of Facebook AI Research Lab), with participation from 42CAPDealroom has a profile for this one. Try Dealroom →, Vermilion VenturesDealroom has a profile for this one. Try Dealroom →, and angel investors from Google DeepMind, n8n, and Snowplow.
The company tackles a core limitation of large language models: they're stateless. Once a session ends, context disappears.
Cognee's open-source engine transforms raw data into a structured knowledge graph that AI agents can query, update, and refine over time. Its "Cognify" pipeline extracts entities and relationships from data, while the "Memify" layer uses feedback loops to sharpen the graph with use.
The platform integrates with major AI frameworks including Claude Agent SDK, OpenAI Agents SDK, LangGraph, and Google ADK.
Why now? Cognee's traction has accelerated sharply. In 2025, the company grew its pipeline volume from roughly 2,000 runs to over one million — a 500x increase in a single year.
It now runs in more than 70 companies. Bayer uses Cognee for scientific research workflows, and the University of Wyoming built an evidence graph from policy documents with page-level provenance.
The open-source project has amassed over 12,000 GitHub stars and 80+ contributors.
What could go wrong? The AI memory space is getting crowded. Cognee must prove that its graph-based approach outperforms simpler alternatives at scale — and that enterprises will pay for what has long been an open-source tool.
Integration complexity could also slow adoption. Supporting 38+ data sources and multiple storage layers (relational, vector, and graph) means more surface area for bugs and maintenance headaches.
The signal: AI agents are moving from demos to production, and infrastructure is catching up. Memory is emerging as a distinct category — not just a feature bolted onto retrieval-augmented generation pipelines.
The investor profile reinforces this shift. Backing from founders of OpenAI and Facebook AI Research suggests that agent memory is being treated as foundational infrastructure, not a niche add-on.
Cognee plans to use the funding to launch a cloud platform, build a Rust engine for edge devices, and expand its open-source connectors.
Sources:
Cognee
Vasilije Markovic
David Myriel
B.S.
