Liminality MCP is in open beta. The first 3 solves on any key are free. Connect the MCP →

About

A neuroscientist started a software lab.

Physea Labs is small and technical. We started from the brain: how memory settles overnight, how a skill sharpens with practice. Then we asked why almost no software works that way.

The honest answer is the business model. Most AI is rented. The model is generic, your context is shallow, and every correction you give it is gone by the next session. That is fine for casual use and wrong for serious work. We build the other thing: a model that arrives shaped to you and keeps adapting on your own machine.

Adapting is the whole point. A model trained once and frozen isn't really learning. It's reciting. The systems we build change in small ways from what you correct, accept, and ask about, the way a brain tunes itself with practice. Use it more, fit it closer.

The part we won't move on: it's yours.

Local
Where it matters, the model and your data sit on hardware you own. No shared inference layer to negotiate, no leakage surface to audit.
Yours
You buy the model once and keep it. If Physea disappeared tomorrow, your install, your memory, and everything it learned stay put.
No token meter
Inference runs on your machine, so every prompt is free and every conversation runs as long as your hardware will carry it.

Investors

Margin that doesn't scale with use.

The customer's hardware runs the model. There's no inference bill after we ship, so margin grows with the number of people we sell to, not the amount they use. We sell the model, not the compute.

Generic AI got very good and still left a gap: it isn't yours. A model that ships already shaped to one person, keeps learning locally, and carries no ongoing cost is a category the rental players are structurally slow to build. That's the wedge.

Where we are

Incorporated. Sia is running. Beta is opening. The lab builds end-to-end today, and the first models are close. We're founder-led and deliberately lean.

We're raising a pre-seed round to take the first models to market.

Round size, valuation, and the full deck go to qualified investors on request. If that's you, write to [email protected] and we'll send the details.

Reach the founder.

Replies come from a person, not a queue. Tell us what you're trying to build and we'll answer plainly, even if the answer is "not us, try this instead."

Or just email [email protected]

Goes straight to the founder. No list, no autoresponder.