Founded by Leading AI Researchers from Stanford, Berkeley, and Cornell, Engram Launches with Microsoft, Notion and Harvey as Early Partners
SAN FRANCISCO, June 23, 2026 /PRNewswire/ — Today, Engram, the company building the learned memory layer for AI, emerged from stealth with $98M in funding from General Catalyst, Kleiner Perkins, Sequoia Capital, Factory, Modern, Amplify Partners, Neo and notable angels and advisors including Assaf Rappaport, co-founder and CEO of Wiz, Andrej Karpathy, co-founder of OpenAI, and Pieter Abbeel, AI and robotics pioneer and co-director of the Berkeley AI Research Lab.
The announcement comes as enterprises are confronting a growing and costly problem: the AI their employees use every day is a brilliant stranger. It can synthesize vast amounts of information and solve complex problems, but it knows next to nothing about their organization. It rereads the same documents, relearns the same context, and rediscovers the same institutional knowledge on every query, every time. As businesses deploy AI agents across every function, those wasted tokens are quietly becoming one of the biggest financial pressures in enterprise technology.
Engram trains models to study an organization’s world and anticipate its questions in advance, forming a compact, continuously improving memory (also known as an “engram”, a neuroscience term meaning the trace of memory in the brain) that’s unique to each customer. The result is models that get smarter the longer they’re used, and that match or outperform frontier models using up to 100x fewer tokens.
“Whatever the AI knows about you is improvised on the spot — a sticky note about your past, a document pulled mid-conversation,” said Dan Biderman, CEO and co-founder of Engram. “If we can anticipate your interactions, we can prepare memories ahead of time instead of pasting them on the fly.”
Engram’s approach is grounded in years of foundational academic research. Biderman completed his postdoctoral work at Stanford under Chris Ré — one of the most influential figures in modern machine learning and a co-founder of Engram — where he focused on making AI agents cheaper to run, after earning a PhD at Columbia’s Center for Theoretical Neuroscience. His co-founders span the leading labs working on how AI learns and remembers. Engram’s CTO, Sabri Eyuboglu, a Stanford PhD under Chris Ré, created Cartridges, one of the canonical recent methods for turning a large body of documents into a small, reusable memory. Jessy Lin, a Berkeley PhD who researched at Meta’s FAIR lab, developed Active Reading, a way of training models to study material deeply rather than just store it. Jack Morris, a Cornell PhD also from FAIR, is known for his work on retrieval and memorization of LLM. And Scott Linderman, a tenured Stanford professor of statistics and neuroscience, is a leading researcher on state space models, a fast-growing alternative to the transformer designed to handle long stretches of information efficiently.
“When an AI reads a 70,000-word legal contract, which is roughly 400 kilobytes of text, its internal memory of that document can swell past 100 gigabytes. That’s 250,000 times larger than the original file, and a huge part of what makes AI slow and expensive to run,” said Eyuboglu. “We do that studying once, ahead of time, training the model to compress everything it learns into a compact memory it can reuse on every query.”
Engram enters the market with meaningful early commercial traction, anchored by a partnership with Microsoft. The two companies are working together to test Engram’s models within Microsoft 365, with the goal of making enterprise AI more efficient and more attuned to the specific context of each organization. The partnership also includes a commitment to GPU capacity across Dapple and Azure, giving Engram the infrastructure to train its models at scale. The work explores how a learned memory layer could one day bring organizational knowledge directly into the tools hundreds of millions of people rely on every day.
“Our customers have built up extraordinary knowledge inside Microsoft 365, and we’ve only begun to tap what it can do for them,” said Jason Graefe, Corporate Vice President, AI Partner Catalyst, at Microsoft. “Engram’s approach could turn that knowledge into a kind of memory each organization owns and controls, while making AI efficient enough to power the long-running, proactive agents we believe every knowledge worker will eventually rely on. It’s the sort of frontier bet we want to be making.”
Engram is also partnering with Notion and Harvey to bring its memory layer into their platforms.
“Our enterprise customers are running long-lived agents across their Notion workspaces, and that kind of always-on work can burn through tokens fast, even for something as simple as triaging a task,” said Simon Last, cofounder of Notion. “We’re testing Engram’s models inside our new custom agents, and we’re already seeing them approach frontier quality while using an order of magnitude fewer tokens, because the agent already knows the workspace instead of rediscovering it on every query.”
“Law firms and enterprises hold a lot of unique knowledge. Soon every employee will rely on agents that are adding millions of tokens per day of new context — faster than context windows or search can keep up,” said Gabe Pereyra, cofounder and President of Harvey. “We’re working with Engram to build learned enterprise memories that are secure, cost-efficient, and turn unstructured context into compounding agentic knowledge bases.”
“Memory is the missing ingredient in AI,” said Hemant Taneja, CEO of General Catalyst. “We see enormous potential for Engram’s technology across the companies we’re building and transforming in healthcare, legal, and financial services, where the institutional knowledge is deep and the cost of running AI against it is only growing. The ability to improve the speed, independence, and cost efficiency of agents is one of the most important things any company can deliver.”
“Most of the conversation around enterprise AI has focused on making models generally smarter. But for the companies actually deploying AI at scale, that was never the hard part. Getting a model to truly remember a specific organization and its unique ways of working is the problem nobody had convincingly solved,” said Leigh Marie Braswell, partner at Kleiner Perkins. “Dan, Sabri, Jessy, Jack, Scott and Chris have spent years on the research that finally makes persistent organizational memory possible, and they are now working to bring this to every AI-native company.”
In a world where model providers accumulate the value generated by every enterprise interaction, Engram offers a model where companies own the intelligence they build. The more an organization uses Engram, the more specialized and proprietary its models become, creating a form of AI that is sovereign to the enterprise and not dependent on or extractable by any model provider.
“Today, even if you wanted to make your AI better, there’s almost nothing you can do,” said Biderman. “Your AI gets better when the model behind it gets better. How you use it has almost no effect. We are building towards a different future: the more you work with a model, the more it learns your world and the better it becomes for you.”
About Engram:
Engram is building AI that actually knows your organization. Its models study an organization’s world in advance, forming a compact, continuously improving memory unique to each customer. Founded by leading AI researchers from Stanford, Berkeley, and Cornell, Engram compresses that knowledge into reusable model memory that matches or outperforms frontier models while using 1-10% of the tokens. The AI gets smarter the more you use it, and you own the memories. Engram is headquartered in San Francisco and backed by General Catalyst, Kleiner Perkins, Sequoia Capital, and others. Learn more at engram.com.
SOURCE Engram