Engram: Giving Enterprise AI a Memory

The models we use everyday are brilliant strangers. They synthesize vast amounts of information and solve genuinely hard problems, yet they remember almost nothing about our preferences and the organizations we are a part of. On every query, these models tend to reread the same documents and relearn the same context. As companies put agents into every function, that repeated work becomes one of their largest budget line items. Those re-read documents are tokens, and tokens are money. When a model reads a 70,000-word contract, its internal memory of that document can balloon past 100 gigabytes, roughly 250,000 times the size of the original file, regenerated from scratch every time.

Engram’s founding insight is that this work doesn’t have to happen over and over in the moment. Their product studies an organization’s world ahead of time and compresses what it learns into a compact, reusable memory. Appropriately, Engram’s name comes from neuroscience. An engram is the trace a memory leaves in the brain. The result is models that match or outperform frontier systems while using as little as 1 to 10% of the tokens, and that keep getting better the more they’re used. The more an organization uses Engram, the more proprietary its memory becomes. 

The ambitious scope of solving the large model memory challenge requires a team with uniquely deep expertise. I’d been following several of the founders’ work for years before I ever met them, so when we first sat down together, I expected to be impressed. However, I didn’t expect that the team would prove even more impressive than the body of work that had first put them on my radar.

Engram’s founders Dan, Sabri, Jessy, Jack, Scott, and Chris have collectively spent many years working on many of the technical foundations underlying machine memory, including continual learning, knowledge retention, information retrieval, memory compression, state space models, and large-scale machine learning systems. What stood out the most to me about the team wasn’t any single credential, but how naturally their ideas fit together. They had each already been thinking about complementary pieces of this problem long before the rest of the industry started paying attention to it.

Just as importantly, Engram quickly assembled a superstar team with the same intellectual depth and curiosity as the founders, drawing from neuroscience, systems, and frontier AI research. After spending time with them all, it became clear to me that if anyone was going to crack long-term memory for AI systems, it would be this team.

Our conviction in Engram is already shared by their serious commercial partners. Microsoft is testing Engram inside Microsoft 365. Notion and Harvey are also testing our memory layer into their platforms. These partners are where a large portion of the world’s AI-assisted knowledge work happens, and their partnerships with Engram underscore the importance of Engram’s technology for personalized, dynamic memory.

We’re so excited to partner with the entire Engram team by leading their Series A, as they come out of stealth with $98M raised.