Skip to main content
Every agent has a real web browser it can drive: navigating pages, clicking, filling forms, and pulling data. Use it when a task can only be done inside a website. Cove using a browser

​
What it can do

Navigate & read

Open pages, follow links, and extract content and data.

Interact

Click buttons, fill forms, select options, submit.

Log in

Sign into sites, with your help for credentials when needed.

Capture

Take screenshots to show you what it sees.

​
The shared browser

In web chat, the agent’s browser is shared with you. It appears in the Computer panel. Watch it work, or take over whenever a human is faster:
  • Logins: sign in yourself instead of handing over credentials.
  • Captchas & 2FA: solve the human-verification step the agent can’t.
  • Payments & confirmations: take over for anything sensitive.
You drive the same browser the agent does. Use Back, Forward, and Reload. Type a URL or a search into the address bar. Open or close tabs. A tab you open is the agent’s tab too: it shows up in the agent’s tab list, so you can line up a page and hand the task back. The agent picks up where you left it. On Slack, email, or text, the same browser is still yours to reach. When a step needs your own login or a code, the agent names that step and links you to the web conversation. Open the shared browser there and enter it yourself. A credential that has to be yours is never a dead end.

​
Terminal

The Computer panel also has a Terminal tab. Open it to get a live shell into the agent’s sandbox, useful for inspecting what the agent built, running a quick command, or debugging something without asking the agent to do it for you. The terminal is only shown when you have edit access on the agent.

​
Search vs. browse

The agent uses the lighter tool first. For reading public information it uses web search and fetch. It opens the full browser when it needs to interact: clicking, filling, logging in, or when a page needs a real browser to render.

​
Good uses

  • Pulling data from a dashboard that has no API
  • Filling out a web form on your behalf
  • Checking a page that requires a login
  • Capturing a screenshot of how something looks
  • Researching across sites that block simple fetching
Ask for it in plain language:
For anything requiring authentication, a captcha, or a payment, the agent hands off to you in the shared browser.

Connect accounts for repeat access

Authorize a service once so the agent doesn’t need the browser to log in every time.