Engramme has opened its memory API to the public, marking what the company describes as the arrival of a new vertical in AI. The API is live for beta access and is designed not as a product to install, but as infrastructure to be embedded invisibly into any app or service, surfacing a user's relevant memories at the moment they are needed, without any search query or prompt from the user.
Engramme
The technology behind this is what Engramme calls Large Memory Models, a purpose-built AI architecture that does not follow the transformer paradigm underlying most of today's language and reasoning systems. Where transformers generate plausible outputs from learned patterns, Engramme's models retrieve actual memories tied to real events, conversations, documents, and interactions. According to the company, this makes hallucinations structurally impossible within its memory layer: the system cannot surface something that did not happen. The architecture is designed to integrate across Gmail, Zoom, WhatsApp, Slack, Google Docs, and wearables, including Meta glasses, operating across devices and surfaces without requiring the user to initiate anything.
Engramme
The intended audience is product teams rather than end consumers directly. Companies integrating the API would be able to offer users persistent, proactive memory across their products, with Engramme acting as the invisible memory layer underneath. Samsung, Superhuman, Dropbox, and Microsoft have each expressed interest in embedding the system. AI has already progressed through language, vision, and audio as core challenges. Engramme bets that memory is the next frontier, and one that no existing transformer-based system can address in the way their architecture is designed to do.
Test out Gateway for yourself!
Behind the product are Gabriel Kreiman and Spandan Madan. Kreiman is a professor at Harvard Medical School and a neuroscientist with over 160 publications spanning Nature, Nature Neuroscience, and ICLR, who closed his academic lab to found the company. His decades of research on how neuronal circuits encode and retrieve memories form the scientific foundation of Engramme's architecture. Madan holds a Harvard PhD in computer science with prior experience at MIT CSAIL, Google DeepMind, Meta, and Adobe. The company has raised a $3M pre-seed round and is now accepting beta applications through its website.
