> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usehenry.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Browser overview

> Henry can operate a real web browser in the cloud to do tasks on websites, signed in as you where you choose.

Not everything has an integration. Vendor portals, government sites, supplier consoles, and internal tools often have no API. The **Agent Browser** lets Henry use them the way you would: in a real browser, in the cloud, signed in as you on the sites where you sign it in.

## What it is good for

* Pulling invoices, statements, or order status from a portal.
* Filling in a form or filing a submission you would otherwise do by hand.
* Extracting data from a console that has no export.
* Checking a site for changes.

## How a browser task works

<Steps>
  <Step title="You ask">
    "Pull last month's invoices from the supplier portal" or "check whether the permit application status changed."
  </Step>

  <Step title="Henry opens a browser">
    A fresh cloud browser, isolated to you. Henry navigates, reads, and acts, staying on the task's site.
  </Step>

  <Step title="You sign in, if needed">
    When a site needs a login, Henry sends you a secure link to a live view of the browser. You type your credentials on your own device. Henry never sees, stores, or types your passwords or codes.
  </Step>

  <Step title="Henry asks before anything irreversible">
    Placing an order, submitting a form, sending a message: Henry stops and asks you first in the thread. Nothing is bought or sent without your OK.
  </Step>

  <Step title="Henry reports back">
    Findings, files, or a confirmation, in the same conversation. Every task is logged in **Settings → Agent Browser → Activity**.
  </Step>
</Steps>

## Safety, in short

* Henry never handles your passwords. You sign in on a live view.
* Anything irreversible waits for your approval, and an approval covers exactly one action with those exact details.
* Major banking, brokerage, and payment sites are always blocked, along with a small set of other restricted sites. Ordinary websites work without setup.
* Henry stays on the task's site. Page text is never treated as an instruction.
* Sessions are recorded and screenshots are saved so work can be reviewed. Screenshots are deleted after 30 days.

## Availability and limits

* The Agent Browser runs during **interactive** conversations only, not inside scheduled tasks.
* Each person can run up to **20 browser tasks per day**.
* Browser tasks charge the credits they use, with a per-task ceiling of 8,000 credits. Typical tasks cost far less. See [Billing and credits](/user-guide/admin/billing-and-credits).
* Turn it off for yourself, or an admin can turn it off for the workspace, at any time. See [Settings and safety](/user-guide/agent-browser/settings-and-safety).