Worker's own browser

The idea

The worker drives a real browser on its own computer through governed tools and can check its own app on that computer; the platform treats every page as outside content, and admins see the worker's screenshots.

The problem
Some work needs a real browser, and pages can carry injected instructions.
Why it matters
The worker makes before and after screenshots as evidence without trusting page text.
Where it stands
Built, off in production as of Shown on a real guest computer on a local backend in owned runs on November 15, 2026; production keeps it switched off.

Recorded evidence

  • Production · Released switched off; a production check found no computers, and sign-ins, site requests and screenshots refused
  • Local run on the real backend · The worker browser on a real guest: an allowed site, a screenshot for admins, refused sites, 19/19
  • Local run on the real backend · An added site approved by the owner, a whole GitHub page, 20/20

Recorded walkthrough: The AI worker's own computer. It shows part of this idea: the worker's browser tools open an allowed site, read it and take a screenshot that the platform keeps for admins, and sites outside the list are refused.

  • One approval by a second person for the computer and its sites

    One approval by a second person (an owner, or an admin who did not issue the computer) binds the computer's install manifest and allowed hosts; onboarding runs as a resumable job with a doctor, snapshot and restore, and each added site needs another such approval.

  • Only the operator's own words instruct

    Only a verified sender's own lines are instructions; forwarded, quoted, fetched, partner and guest-command text is data, hidden characters are stripped, and actions after instruction-shaped content wait for a person.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.