Engram emerges from stealth with $98M to build a learned memory layer for AI

Engram , a San Francisco-based startup building a learned memory layer for AI, has emerged from stealth with $98M in funding at a reported $600M post-money valuation. The company tackles a core enterprise-AI inefficiency: models having to relearn an organisation's context on every query, a major driver of cost and latency.

Investors. The round was backed by General Catalyst, Kleiner Perkins, Sequoia Capital, FactoryDealroom has a profile for this one. Try Dealroom →, ModernDealroom has a profile for this one. Try Dealroom →, Amplify Partners and Neo, alongside notable angels and advisers including Assaf RappaportDealroom has a profile for this one. Try Dealroom → (co-founder & CEO of Wiz),Andrej KarpathyDealroom has a profile for this one. Try Dealroom → (co-founder of OpenAI) and AI/robotics pioneer Pieter Abbeel.

What it does. Rather than bolting an external memory database onto the context window, Engram bakes new and evolving context directly into model weights via adapter fine-tuning. It trains per-team models that internalise a company's documents, workflows and institutional knowledge — so a model "just knows" things the way a long-tenured employee does, instead of re-reading files at test time. Engram says this lets models answer in ~100 tokens what frontier models would burn ~100,000 tokens doing.

Use of funds. The capital will scale model training and further develop the technology, which compresses institutional knowledge into compact, reusable memory.

Founders. Founded in 2026 by researchers from Stanford, Berkeley and Cornell — CEO Dan Biderman, CTO Sabri Eyuboglu, Jessy Lin, Jack Morris, Scott Linderman and Chris Ré. Early partnerships include Notion, Microsoft and Harvey.

Sources: Quartz · The SaaS News