A frontier model that runs inference on your own hardware, not in someone else's data center. No account, no setup, no per-token bill — pull the weights from Hugging Face and run them yourself. The claim is one you can check rather than one you have to take on faith.
ENIAC — 1,800 square feet of vacuum tubes.
The microchip — fits on a fingertip.
The smartphone — a computer in your pocket.
Cloud AI — back to a warehouse of GPUs.
Watt-1 — a frontier model, back in your pocket.
"The next data center is the phone in your pocket."
Closed labs don't publish parameter counts for their current frontier models — listed as undisclosed rather than estimated.
No server. No per-token bill. Weights open on Hugging Face — verifiable, not just claimed.
1 The model ships inside the app — nothing to provision.
2 Every token runs on the phone's own chip — nothing sent anywhere.
3 Works with no connection — same weights, same result, offline.
Cloud inference bills you because someone else's GPU did the work. Run it on hardware you already own, and the meter stops.
Does anything leave my device?
No — inference runs entirely on-device.
Does it work offline?
Yes, once installed, with no connection at all.
How do I run it today?
Download the weights from Hugging Face and run them on your own hardware. A native app is in progress — request access to hear when it ships.