A desktop app and a CLI for running coding agents.
Each session runs in a host process on the machine the work is on — your laptop, or a server over ssh. Closing the app does not stop a turn, and the same session opens again from a browser or a phone.
Builds are unsigned and say so. Nothing is uploaded: this page talks to your own machine.
Screenshots of a real run, taken by the test suite. Click any of them.
Close the app mid-run, open the same session on another machine, restart the box — all three are the same arrangement, not three features.
Two machines on one session see one transcript. A prompt raised on either is answerable from the other.
Which agent ran, under which model, and who approved each thing it needed.
Group them and they reach each other — a message that wakes one, or a read of what it has been doing.
The decomposition is written down and reviewed in a diff, so an agent can propose one without running it.
A plain tunnel brings a dev server here. A remote window shows a few frames a second, with nothing installed there.
A page for what is tested, what was only reasoned about, and what is missing.
On your laptop or on a server you reach by ssh. It prints a link with a token in it; that link is what the client wants.
npx agbrte web .