Every machine has one browser for its agents: a real Chromium on a virtual screen. Agents reach it through the playwright and chrome-devtools tools.
Give the agent a browser job
In a project with a dev server:
repose run -p "start the dev server, open the signup page \
with playwright, fill the form with a test address and \
screenshot the result"The first playwright call starts the browser on the machine.
Open the view
From another terminal on your laptop, or the same one after the agent has started:
$ repose browser
Watching todo-app's browser at http://localhost:6080/#p=5m2k8Q1p
(the view sleeps after 30 idle minutes).The command returns at once and keeps its forward running in the background; run it again for the same link.
Take over
Clicks in the view go to the agent's browser. Use it for the steps only a person can do:
- A login. Sign in to the site the agent needs. The session stays in the browser's profile, so the agent's next call is logged in, and so is the next agent tomorrow.
- A captcha, a passkey, a 2FA prompt. Do it, then tell the agent to carry on.
The agents' guide to the machine tells them to ask you for repose browser when a page needs a human.
Close it
Closing the tab leaves the view running until it sleeps; repose browser stop ends it and the forward now.
Things worth knowing
- One browser, one profile. Anything you log in to is there for every agent on that machine. Treat it like a browser on a shared computer: don't log in to more than the job needs.
- Your own Chrome instead. When the job needs your laptop's logins or extensions, lend the agent your Chrome.