Engram

Engram is a San Francisco-based artificial intelligence company developing a learned memory layer to enhance the efficiency and reduce the operational costs of enterprise AI. The company addresses the significant expense and inefficiency of current AI models, which must re-process an organization's entire context with every query. Engram's architecture separates an AI's reasoning and inference capabilities from its memory, allowing it to study a client's specific data—such as documents, workflows, and institutional knowledge—in advance. This information is compressed into a compact, reusable, and continuously improving memory layer, which can reduce token consumption by 90-99% and cut AI-related costs by a factor of 10 to 100. Founded in October 2023 by a team of AI researchers from Stanford, Berkeley, and Cornell, Engram emerged from stealth in June 2026 with a $98 million funding round at a $600 million valuation. The investment was led by General Catalyst, Kleiner Perkins, and Sequoia Capital, with participation from prominent angel investors like OpenAI co-founder Andrej Karpathy and Wiz CEO Assaf Rappaport. The founding team includes CEO Dr. Dan Biderman, who conducted postdoctoral work at Stanford's AI lab; CTO Sabri Eyuboglu, a Stanford PhD; influential Stanford professor Chris Ré; and researchers Jessy Lin and Jack Morris, both of whom have experience at Meta's FAIR lab. Engram's business model is centered on providing customized models that adapt autonomously to each customer's needs. The company targets large enterprises struggling with high AI expenditures and has already established strategic partnerships to test its models with companies including Microsoft, Notion, and the legal AI firm Harvey. The collaboration with Microsoft involves integrating Engram's models within the Microsoft 365 ecosystem, backed by a commitment for GPU capacity on the Azure cloud infrastructure, aiming to bring more efficient, context-aware AI to knowledge workers. Keywords: learned memory layer, AI memory, continual learning, online learning, token reduction, inference cost, enterprise AI, organizational knowledge, AI cost savings, Dan Biderman, Sabri Eyuboglu, Chris Ré, Jessy Lin, Jack Morris, Kleiner Perkins, General Catalyst, Sequoia Capital, Andrej Karpathy, AI efficiency, knowledge retention, memory compression, AI agents, large language models, model fine-tuning, Microsoft 365, Notion partnership, Harvey AI