Computer Use

Coming soon: let agents drive desktop apps with explicit approval.

Computer Use is coming soon.

The planned capability will let agents control desktop apps the way you do- moving the cursor, clicking, typing, and switching apps.

Info:

Coming soon. This page previews the planned Computer Use experience; details may change before launch.

Use it when the work spans tools that don't expose APIs: legacy admin UIs, design tools, internal apps, anything you'd otherwise have to drive by hand.

Agent driving a desktop app via Computer Use

Computer Use plugin approving an action mid-session.

Planned availability

Computer Use will ship as a plugin. Once available, enable it in the workspace settings; it will become available to any session you launch in that workspace.

When the agent first tries to take a Computer Use action, you'll see a permission prompt. Approvals are scoped per session- a session can be allowed to click and type without being allowed to access files outside the workspace.

Planned capabilities

When enabled, the plugin will give the agent these capabilities:

  • Read the screen (or a region of it).
  • Move the cursor and click.
  • Type text and send keyboard shortcuts.
  • Switch focus between windows and apps.
  • Take screenshots that stream back into the transcript.

The screenshots are how you watch the agent work in real time- you see exactly what it sees.

Workspace boundary

Warning:

Computer Use will run against your real machine. Permission prompts are your last line of defense before the agent acts on a real UI.

Computer Use will be one of the most powerful plugins Proliferate ships, and one of the most sensitive. Apply the same care you would to anything that touches your machine outside a sandboxed workspace:

  • Only enable it on workspaces where you actually need it.
  • Keep approvals tight. Approve specific actions, not blanket sessions.
  • Watch the screenshot stream. If the agent goes somewhere unexpected, stop the session.

For lighter-weight needs- driving a website, filling a form, navigating a dashboard- Browser Use will provide a narrower browser-only scope.

Planned uses

  • Driving internal admin UIs that don't have APIs.
  • Working in design tools (Figma, etc.) the agent doesn't have a plugin for.
  • Running through onboarding flows or installers as part of a setup script.
  • Reproducing a bug that requires real desktop interaction.

On this page