GitHub Copilot Gains 'Computer Use' to Control Desktop Apps on macOS and Windows
GitHub launched a public preview of computer use for GitHub Copilot, letting it interact directly with desktop applications on macOS and Windows.
Details
- What it does: Copilot can autonomously read content, click controls, type text, scroll, and navigate across desktop applications, extending assistance to legacy software and GUI-only programs that lack APIs or command-line interfaces
- Approval model: Copilot requests user approval before controlling any application; users can review or reset permissions for apps marked always-allowed, and organizations can disable the feature entirely through managed settings
- macOS setup: the system walks users through required Accessibility and Screen Recording permissions
- Availability: public preview in GitHub Copilot CLI and the GitHub Copilot app on both macOS and Windows
- Getting started: enable via
/computer onin CLI (check status with/computer show, disable with/computer off), or through Settings → Computer Use in the app - Example use cases: summarizing notifications in a browser, updating content in a presentation, or moving information through a workflow in a desktop application
What happened next
This gives GitHub Copilot the same kind of autonomous desktop/computer-control capability OpenAI built into GPT-6 Astra and extended to its “dots” agents — a sign that computer-use agents are becoming a standard capability tier across major AI platforms rather than a single vendor’s differentiator, and it arrived the same day as Copilot’s new dynamic workflows feature.