Public preview of GitHub Copilot computer use is appearing in Copilot CLI and the desktop app on Windows and macOS. Copilot can read accessible content and visual context, click, type, scroll, drag, and navigate across desktop applications. This may help with legacy or GUI-only software that has no API, CLI, or MCP interface. It requires user permission; it is not silent background control. Read the GitHub Changelog post.
How to enable it
- Copilot CLI: Run
/computer onto enable,/computer showto check status, and/computer offto disable. - Copilot app: Open Settings → Computer Use → Enable Computer Use.
- macOS: Follow the prompts for Accessibility and Screen Recording permissions.
Copilot asks before controlling an app, and organization settings can disable the feature. You can review or reset apps marked as always allowed. Check permissions before testing, especially if private windows or sensitive information may be visible.
Write a bounded instruction
State the outcome, the apps involved, and constraints. For example: “Review this week’s browser notifications and list unresolved items only. Do not delete or send anything; sort the list by date.” This is easier to review than “organize this for me.” Check the actual interface and result when it finishes rather than relying only on Copilot’s confirmation.
Start with a reversible task
- Choose a workflow that can be undone, such as organizing notifications or filling a test form.
- Use an account without payment permissions and approve app control as it is requested.
- Keep before-and-after states and verify the selected window, fields, and controls.
- If a CAPTCHA, permission dialog, or unexpected page appears, pause for a person instead of asking it to bypass safeguards.
GUI automation is sensitive to layout, scaling, load times, and pop-ups. Keep human confirmation and rollback for payments, account settings, or external email. The announcement does not establish a universal additional fee.