Made for the way you work
A browser the agent can use
Move from code to the result people see. The agent can work in a real browser inside your editing workflow.
In practice
A useful place to start
After changing a page, ask the agent to load it, try the flow, inspect errors, and capture the result. Open the same page yourself to check what it saw.
Test real behavior
Browser tools can navigate, click, type, select, scroll, and inspect page content. Screenshots, console output, and network details help explain a problem.
Keep visual evidence
Capture screenshots or bounded recordings when a bug depends on what the page does over time. The agent can inspect recorded frames and use them as evidence.
Control the browser
Open tabs beside the chat or in a separate window. Clean session profiles are the default; using a persistent profile with existing logins needs an explicit setting.
Get started
Enable the agent browser in Settings, then ask it to open the page and verify a concrete behavior. Take over the tab when a step needs you.
Good to know
Browser actions can change real accounts and send data to websites. Access to your logged-in profile is a separate choice. Some operations, such as file upload, have workspace-host restrictions.