Browser
The agent can use a browser for a web page, form, download, or visual check that an API cannot provide.
Compare the current pricing pages for these five products. Record the plan name, monthly price, usage limit, and source URL. Use only the vendors’ own pages.
What the browser can do
Section titled “What the browser can do”- Open and navigate public websites.
- Fill forms and click page controls.
- Read rendered content that search or an API does not expose.
- Upload a file from the conversation workspace.
- Download a result and return it through the conversation.
- Check a site at desktop and phone widths.
Browser work runs in an isolated session. Do not assume that it contains your personal login or cookies. Use a supported connection for account data.
Control page actions
Section titled “Control page actions”State the URL, objective, allowed actions, and required evidence. Require approval before the agent submits a form, publishes content, buys something, or changes external state.
Open the staging signup flow, create a test account with the supplied test address, and report each broken step. Do not use production or send an invitation.
Website content is external data. A page cannot change your request or grant new authority.