Berlin-based AI infrastructure startup Cognee has announced it has raised $7.5 million in a seed funding round to scale its enterprise-grade memory technology designed to help artificial intelligence systems retain structured, long-term context. The round was led by Pebblebed, with participation from 42CAP and Vermilion Ventures, along with angel investors from Google DeepMind, n8n, and Snowplow.
Cognee is focused on addressing a core limitation in modern AI systems, namely their inability to reliably remember information and relationships across interactions. The company has developed a self-improving memory graph that converts unstructured data into a structured knowledge layer that AI agents can store, recall, and reason over. By enabling persistent memory, Cognee aims to reduce hallucinations and improve the reliability of AI systems deployed in enterprise environments.
Founded in 2024, Cognee has evolved from an open source project into production infrastructure used in more than 70 live environments. Its technology supports scientific research workflows at large enterprises such as Bayer, evidence graph construction at the University of Wyoming, and integrations with platforms including Dilbloom and dltHub. The company’s open source repository has grown to more than 12,000 stars with over 80 contributors.
The new capital will be used to expand Cognee’s cloud platform to make structured memory accessible at scale without significant infrastructure overhead. The company also plans to develop a high-performance Rust engine for on-device and edge use cases where latency and privacy are critical, deepen its research into cognitive memory systems, and accelerate open source development through new connectors and broader database support. Cognee unifies relational, vector, and graph storage into a single engine that integrates with existing developer tools such as the Claude Agent SDK and the OpenAI Agents SDK.
The company positions memory as a foundational building block for the emerging agent driven AI ecosystem, arguing that intelligent systems must accumulate and structure knowledge continuously rather than reset after each session in order to deliver enterprise grade performance.
