How it works
ElaNa is built to work even when a person has no connection at all — but it isn’t static. When a connection is available, ElaNa syncs with the larger, server-side versions of the Okran LLM: the server pushes updated, compressed weights directly down to the offline model. It’s a weight-sync mechanism, not distillation and not live retrieval — the offline model actually absorbs what the bigger models have learned, rather than just borrowing their answers over a network call.
Why it matters — availability
This is what “AI that’s easy for people” looks like when you take it seriously. Not everyone has fast, constant internet. An AI that only works when you’re online isn’t actually accessible — it’s just convenient for people who already have the infrastructure. ElaNa is Ulter’s attempt to close that gap.
Why it matters — privacy
But availability isn’t the only reason ElaNa matters to me — privacy is the other half of it. I sit on both sides of this: I’m the CEO of Ulter, and I’m also a user of AI products myself, so I understand the pull from both directions. Learning from users’ chats is a legitimate, real method for making a model better — I’m not going to pretend otherwise. But learning in order to improve a model is a different thing entirely from using that same data for advertising. Those are two different purposes, and they deserve two different products.
That’s why Ulter will offer both the Okran LLM and ElaNa rather than forcing everyone into one model. ElaNa is built specifically for the people and organizations for whom privacy isn’t optional: individual users who simply want that peace of mind, corporations working on projects they can’t afford to have leaked, and even governments. For all of them, running fully offline — with no dependency on a live connection to a server that could see their data — is the actual requirement, not a nice-to-have.
The one dependency
Project Quantum Reduction isn’t a side project sitting next to ElaNa — it’s the lifeline ElaNa depends on. Making ElaNa real, at a size that can genuinely run offline while still being useful, only happens if that reduction succeeds. If it doesn’t work, ElaNa doesn’t exist. Everything about ElaNa’s future comes down to that one dependency.
The research this depends on: Project Quantum Reduction →
The privacy position in full: Privacy →
← Back to the research log