We believe in a future where every business, every institution and eventually every individual runs its own AI system. Built for them, trained on what they know, and owned outright rather than rented by the token from someone else's cloud.
That future is better for the people using it, because nothing they type ever leaves the building. It is also better for the planet. A machine that sits in an office is air-cooled, needs no water and plugs into a standard wall socket.
Compress a state-of-the-art model far enough and it becomes the box in the corner of your office.
Locai Labs is a UK lab, based in London, working on the problems that stand in the way of that. Making a state-of-the-art model small enough and fast enough to serve a whole organisation from a single machine, and letting it keep learning on private data once it is there without losing what it already knew.
An AI appliance that sits inside your own network. Air-cooled, plugged into a standard wall socket, and sized to serve a whole team without a per-token bill.
The models that run on it. Open weight, tuned to the exact hardware they ship on, and yours to keep and inspect.
Efficient inference, so a state-of-the-art model fits on one machine, and continual learning, so it keeps improving on your data once it is deployed.
Your data belongs to you. Everything we build is designed so the model comes to your data rather than your data going to a vendor: deployed inside your own perimeter, with no harvesting, no hidden tracking, and no third party reading what your team asks.
Being open by design means building AI that is accessible, auditable, and created in collaboration with the world. We're committed to publishing open-weight models, research, and methods so anyone can inspect, improve, and build upon them.
AI doesn't have to cost the earth. We build efficient models, and hardware to match: Locai One is air-cooled, uses no water for cooling and plugs into a standard wall socket.