Browser
Starting & interaction
Start the agent's browser, watch it work, and take control to sign in.
The agent's browser is in beta and not switched on for workspaces yet. This page describes how it works once it is.
Each agent has one browser. Everyone who works on the agent shares it, with the same tabs and the same sign-ins. The agent uses it to work on websites, and you watch it live.
Start the browser
- Open the agent's Browser tab in the Computer panel.
- Click the play button beside Browser. The live view opens in the Computer panel, and the play button becomes a stop button.
- Watch the page, or ask the agent in chat to open a site.
You can also just ask the agent, for example "Open our supplier portal in the browser and find last month's invoice total." It starts the browser when it needs it.
The + beside Browser opens a blank tab and gives you control, ready for an address. On a stopped browser it starts the browser instead.
Tabs
The Browser tab lists the browser's open tabs. Click one to bring it to the front of the view. The agent's own tab carries the agent icon. Looking at another tab doesn't move the agent, which keeps working in its own.
A stopped browser still lists its last tabs, dimmed. They come back when you start it again.
Watch and take control
While the agent works, the page is locked to you. Clicking it asks Take control?, and the agent pauses until you hand it back.
The bar above the page has Back, Forward, Reload, the address, and Take control. Once you have control, you can click, type and enter an address. Text that isn't an address searches Google.
- Click Take control.
- Use the page as you would any browser.
- Click Hand back when you're done. The agent carries on.
Sign in to a site
The agent never types into a password field. When a site needs a sign-in, take control, sign in yourself, and hand back.
The browser stays signed in, and everyone with Editor or Full access on the agent uses the same browser. Sign in only with an account everyone on the agent may use.
Approve a submit
When the agent fills in a form, it pauses before it submits. A panel at the top of the page names the step it wants to take. Click Approve to let it continue, or Reject to stop it.
Stop the browser
Click the stop button beside Browser. Closing the view doesn't stop the browser or the agent's work in it. A browser left with no activity for a while stops by itself.
If the view freezes, it says "The browser stopped responding." Click Reconnect, or Restart to stop and start it again.
Who can use it
Starting the browser and using it need Editor or Full access on the agent. With Viewer access you can't start it or watch it. See Roles & permissions.
The agent can't use the browser from a Telegram or Slack chat. Ask it in the app instead.