Browser Use

Coming soon: give agents a browser for forms, dashboards, OAuth flows, and web testing.

Browser Use is coming soon.

The planned capability covers the most common case where an agent needs more than its CLI: driving a real browser.

Info:

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

Use it for forms, dashboards, OAuth flows, scraping internal docs, or testing your own web UI. The agent will navigate a browser and report back what it sees.

Agent driving a browser to fill a form

Browser Use plugin running in a workspace session, showing the agent's live browser view alongside the chat transcript.

Why a separate plugin

Browser Use is planned as a strict subset of Computer Use: it will only drive a browser, not your whole machine.

That's the point. When the work is "navigate a website," you shouldn't have to grant your agent click-anywhere access to your desktop. Browser Use is meant to give you the narrower capability with the narrower permission scope.

Planned availability

Browser Use will ship as a plugin. Like other plugins, you'll enable it once and it becomes available to sessions in the workspaces you choose.

The first time the agent navigates somewhere new, you'll see a permission prompt, the same plugin-scoped approval model other capabilities already use. Approvals are scoped per session.

Planned capabilities

  • Open URLs and follow links.
  • Fill form fields and submit forms.
  • Click buttons and other interactive elements.
  • Read page content (DOM, visible text).
  • Take screenshots that stream back into the transcript.
  • Handle multi-step flows like OAuth or pagination.

Sessions and credentials

Info:

Browser sessions will be isolated per workspace by default. If you need the agent to use a logged-in session (your GitHub, your Linear), you'll grant it explicitly via a credential binding on the plugin.

This keeps work-account credentials out of unrelated sessions and out of agents that don't need them.

Planned uses

  • Filling out forms in admin tools.
  • Running through OAuth or auth flows.
  • Pulling data from internal dashboards that lack APIs.
  • Smoke-testing your own web UI as part of a workflow.
  • Verifying claims from a documentation page by actually loading it.

On this page