Skip to main content

AI Agents

The agent's browser

Updated October 1, 2026 · 3 min read

The agent's browser

Some work lives on websites that have no connection to Claos. For that, an agent can open a real web browser: it reads pages, clicks, scrolls and fills in forms, while you watch beside the chat and step in when a person is needed.

Asking an agent to browse

Just ask: "Open the Wikipedia article on tea and summarise the history section", or "Check our pricing page and list anything that looks out of date." When the agent starts browsing, the Browser panel opens beside the conversation and you see the page live. The agent's cursor, labelled with its name, shows where it's working.

You can also open the browser yourself with the globe button at the top of the conversation (Open the browser). It starts on a search page, ready for you or the agent.

Using the browser yourself

The panel works like a small browser: tabs, back, forward, reload, and an address bar that takes an address or a search. You can click, scroll and type in the page at any time, even while the agent is working. The agent reads the page fresh at every step, so it picks up from wherever you left it.

When the agent hands over to you

The agent doesn't fill in passwords, one-time codes or card details itself (unless you've saved a login for the site, below). When a page asks for one, or shows a CAPTCHA, it pauses and a card appears in the chat, such as Sign in needed. Do what the page asks in the browser panel, then press Continue on the card. Cancel tells the agent to go on without it. Typing a message in the chat counts as Continue too.

Before a click that can't be undone, such as Send, Place order, Pay, Delete or Publish, the agent pauses with Approval needed. Choose Approve to let that one click through, or Deny.

Connected sites

Once you sign in to a site in the agent's browser, Claos remembers the sign-in, so next time the agent opens that site already signed in. It's saved encrypted, for you alone.

To see and manage these, press the key button at the top of the browser panel to open Connected sites. For each site you can:

  • Add login. Save a username and password so the agent can sign in by itself next time, without handing over to you. The agent uses the login but never sees it.
  • Available to. Leave All agents ticked, or choose which agents may use this site.
  • Add a note, such as which account this is.
  • Forget this site signs the browser out of it. A saved login stays.

Add a login for another site saves a login for a site you haven't visited yet, and Forget every site signs the browser out everywhere.

The browser in a room

A room has one browser, opened with the globe button in the room's header. The person whose message started it can use it; everyone else in the room watches. If someone else sends the next message, the browser switches to them.

When the browser closes

The browser closes after a few minutes without activity between messages. It isn't lost: the panel shows Browser paused, and one click reopens it on the same pages, for up to seven days. Pages reload when it reopens, so anything typed but not submitted is gone.

If an agent has been browsing for an hour with nothing from you, it stops and asks whether to carry on.

Good to know

  • The agent's browser can't open private or internal network addresses.
  • A workspace can have a few browsers open at once, one per conversation. If you reach the limit, Claos closes your own idle ones first.
  • Browsing time counts a little toward your usage, on top of the conversation itself.
  • On a phone you can watch the browser; to click or type in it yourself, use a larger screen.
  • People with the viewer role can't use the browser.

Was this article helpful?

Back to AI Agents

Support

Still need help?

Can't find what you're looking for? Start a conversation and our support team will help.

Open your laptop.The work is already done.

Free to start. No credit card required.