How Browser Core works

Private beta documentation

1. Choose an execution location

Local: a connected computer executes a browser with its own resources. BYO: a worker runs on a server you control. The commercial shared Cloud option is not available in this beta.

2. Give explicit access

For an existing Chrome/Edge tab, open the Browser Core extension and grant access only to the specific tab. Chrome keeps its cookies and passwords; the service does not export the browser profile. A new origin requires renewed permission.

3. Use the agent and verify outcomes

The agent may read a limited DOM snapshot and interact with page controls. Website content is untrusted. Check the current site, project and confirmation receipt before changing important settings or submitting forms.

Important usability limitation: Remote manual control is a temporary intervention tool for CAPTCHA, login/password entry, 2FA/OAuth, and exceptional corrections. Video, mouse and keyboard operations are transmitted over the network and can be less responsive than a local browser. Do not choose the remote handoff to browse or work for long periods. For regular interaction use the Local worker and your own Chrome window.

4. Take over a session

For login, CAPTCHA and multi-factor authentication, manually take over the live browser. Enter credentials directly into that browser, not into ChatGPT messages. The agent is paused until the user finishes. One-use handoff links expire.

5. Data and limitations

Persistent profiles are encrypted on your selected worker. A crash or sign-out may require re-authentication. This is a beta and cannot promise uninterrupted 24/7 sessions. No credit card is required for the current private pilot.

Troubleshooting and contact