Cognee: Berlin Startup Raises $7.5 million, Builds Long-Term Memory for AI Agents
Berlin-based AI startup Cognee has closed a funding round of $7.5 million. Lead investor is Pebblebed, with 42CAP joining as a follow-on investor. Existing and new business angels are also participating. The company develops a structured memory layer for AI systems and agents – a technology that addresses a fundamental problem of today’s AI: missing long-term memory.
“AI systems today don’t fail because they aren’t powerful enough,” says Vasilije Markovic, founder and CEO of Cognee. “They fail because they don’t remember. We’re building the memory layer that allows AI to understand context, not just retrieve text.” Most AI systems rely on stateless approaches such as file-based retrieval or short context windows.
Cognee transforms unstructured data into a persistent, structured memory layer based on knowledge graphs and semantic representations. This is intended to enable AI systems to maintain context over time, process networked information, and significantly reduce hallucinations in production environments.
Thousands of developers, over 70 companies in use
Cognee emerged from an open-source project and is experiencing rapid adoption in the global developer community. Thousands of engineers are already using the technology to build AI-native applications that require reliable memory and context handling. In parallel, Cognee is running at more than 70 companies, primarily in knowledge-intensive and regulated industries. The fresh capital flows into further development of core memory and graph technology.
European DeepTech infrastructure for the global AI stack
Founded in 2024, the company has its roots and headquarters in Berlin but targets a global market. The team operates from Europe and a second location in San Francisco and works closely with developers and enterprises worldwide. “Europe has a strong tradition in building deep infrastructure,” says Markovic. “Our ambition is to contribute a core building block to the global AI stack – not another application, but infrastructure that many AI systems can rely on.”
The positioning is clear: Cognee wants to be not an AI application, but the layer beneath it – the infrastructure on which others build. With structured memory, persistent context, and semantic networking, the startup addresses a gap that becomes increasingly important as AI workflows grow in complexity.
