Compute
An AI that moves the mouse.
Agents see the screen, click, and type to finish real UI work: with your consent.
What you get
Screenshots feed the model's vision; a zoom verb magnifies small targets before clicking.
Click, double-click, drag, scroll, type, hotkeys, clipboard, window management: the whole desktop.
Computer use stays off until a human enables it on that computer: the cloud can't flip it remotely.
An exclusive control lease means two chats can never fight over one mouse.
A screen border shows when an agent has control, with an always-on-top Stop button.
Screenshots and actions land in the chat and the computer's Computer Use history.
In action
Who it's for
Walk through real UI flows: file the expense, click through the admin panel, verify the release.
Hand off the fiddly desktop task: exports from legacy tools, batch settings, form marathons.
Give an agent eyes and hands on a cloud desktop and watch it work, step by step.
How it works
- 1Allow it once
Enable 'Allow computer use' on the machine: cloud desktops come ready.
- 2Describe the job
Tell the agent what to get done, like you'd brief a colleague.
- 3Watch it work
The screen border shows control; every click and screenshot streams into the chat.
- 4Stop any time
The on-screen Stop button interrupts instantly without killing the whole run.
Frequently asked
Ready to try it?
Try 200+ models free for 7 days, then pick the plan that fits. Cancel anytime.