> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cogniagent.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# The live browser
> Your AI coworker uses websites the way you do — logging in, filling forms, clicking through pages — and hands you the keyboard when it hits a password.
It uses websites the way you do — and hands you the keyboard when it hits a login.
## What it does
Some things simply aren't available any other way. A supplier portal with no export button. A form on a site nobody has ever built an integration for. A dashboard that only exists as a dashboard.
The live browser gives your coworker a real browser it can drive: navigating, typing, clicking, scrolling. You can watch it happen. And when it reaches something only you can get past — a password, a two-factor code — it stops and hands the browser over to you rather than guessing.
## When you'd use it
* Pull a report out of a supplier portal that has no export button
* Submit a form on a site with no integration
* Check stock levels on a partner's dashboard every morning
* Page through search results on a site that blocks automated readers
* Download a statement that's only available after signing in
* Fill in the same tedious web form for forty records
## The handoff, which is the good part
This is the bit worth understanding, because it's what makes the capability usable on real sites.
A live view appears inside the task. You can watch every page load and every click as it happens.
A password, a one-time code, a "prove you are human" check — anything it shouldn't or can't get past on its own.
A card appears in the conversation. The browser is now yours, in the same window, on the same page.
Enter the password, approve the prompt, then give control back. Your coworker picks up exactly where it stopped.
A named browser remembers where it's logged in. The next task on the same site doesn't ask again.
That last step is what turns this from a party trick into something you'd use daily. You sign in once, not once per task.
## Turning it on
Open your coworker, choose **Capabilities**, and find **Live browser** in the Built-in tools list. It's on by default.
> **Live browser** — Drive a real cloud browser: log in to sites, fill forms, click through pages. You can watch it live, and logins are remembered.
## Good to know
**Reach for the simpler tool first.** If a page is public, [Reading web pages](/cowork/capabilities/reading-web-pages) is faster and cheaper. The browser is for pages that need signing in to, clicking through, or filling in.
The browser runs in the cloud, not on your computer. Closing your laptop doesn't stop it, and nothing it does touches your own browser or your own sessions.
You type your own passwords, during the handoff, into the site itself. Never put a password in the chat — your coworker doesn't need it and shouldn't have it.
## Common questions
Yes — that's the point of the live view. You see the actual browser as it navigates, and you can step in at any moment, not just when it asks.
No. A named browser keeps its sessions, so once you've signed into a site, later tasks find it already signed in. You'll only be asked again if the site signs you out.
It waits. The task pauses at the handoff rather than failing or guessing, and picks up whenever you get to it. If you'd rather it kept busy elsewhere, see [Self-scheduling](/cowork/capabilities/self-scheduling).
You control which sites it ever visits by what you ask for, and you can require approval before it acts using [Permission rules](/cowork/control/permission-rules). Credentials you enter go into the site, never into the conversation.
Your coworker reads the page each time rather than following a fixed script, so a redesign that would break a recorded macro usually doesn't trouble it. If it genuinely gets stuck, it tells you instead of pretending it worked.
## Next steps
The lighter option for public pages.
Check for a proper integration first.
What pauses for your yes.
Run that portal check every morning.