Engram is an AI company developing models that learn from the context of a person or organization and build a deeper understanding of their work. Rather than relying only on public internet data or repeatedly searching through documents, Engram aims to train models on the knowledge contained in shared workspaces, helping them retain useful information and improve over time.

The company’s first product is an API for agents designed to learn from large knowledge workspaces. In partnership with Notion, Engram is building custom agents that understand extensive Notion workspaces. With Harvey, it is developing models that internalize a firm’s knowledge and help find precedents across client matters. Engram is also piloting its models inside Microsoft 365 to support cost-efficient, customized agents for enterprise customers.

Engram’s approach builds on research into areas including context compression, retrieval, long-context systems, memory architectures, and model forgetting. Its broader goal is to create models that continually absorb relevant data and become more useful as work and knowledge evolve.