Agents can use your saved browser from their first step

We made the browser one of the tools every agent starts with, like running code or reading files. Until now, an agent had to load the browser guide before it could open a browser at all. Most runs did, but a scheduled workflow with a saved browser would sometimes skip that step, decide it had no browser, and stop without opening a page, even though the browser and its saved logins were fine. That failure is gone: a workflow that follows a script now goes straight to opening the browser.

How it works

• Workflows with a saved browser have the browser from their first step, whether they run on a schedule, from a trigger, or by hand.
• Conversations with a specialist always have the browser, for any site behind a login or without an API.
• Workflows without a saved browser don't see the browser at all, since they could never open one. To give a workflow a browser, attach one where you connect its other accounts.
• The browser guide is still there for agents to read before they drive an unfamiliar site by hand: keeping track of what was asked, checking each step, handling sign-in pages, and recovering from errors.