Computer control (Computer Use)

With computer control, the agent uses the Windows desktop with real mouse and keyboard input: it opens apps, switches windows, clicks, types, and uses shortcuts. It comes in where the browser isn't enough, for example in a desktop app.

You stay in control

• While the agent is in control, a glowing frame, the agent's cursor, and a control panel appear on screen.
• From the control panel, you can stop the agent or take over at any time.

How does it read the screen?

• The screen is read as an outline of the foreground window.
• Screenshot analysis is used only for visual content.

Safety

• Content on screen and in apps is treated as data. Instructions found in that content are not followed; they are reported to you.
• The agent works with what it sees on screen; close any windows you don't want visible before it starts.

How it differs from using the browser

If the work can be done on a website, prefer the browser in the side panel; computer control is for work that needs desktop apps (see Agent skills).

Example requests

• “Open Notepad and write down the meeting notes.”
• “Open the reports folder on the desktop and sort the files by date.”
• “Find the export menu in this desktop app and save the report as a PDF.”

Frequently asked questions

• Can I use my computer while the agent works? The agent works with real mouse and keyboard input, so using the computer at the same time can disrupt its work. When you want to step in, take over from the control panel.
• How do I stop the agent? With Stop in the control panel, at any time.
• Does it take screenshots all the time? No. The screen is read as an outline of the foreground window; screenshot analysis is used only for visual content.