# The agent browser

> Watch the browser an agent uses on a ticket, take it over when it needs a hand, and review the screenshots it leaves behind.

This page shows you how to watch the browser an agent works in, and how to take it over when the agent needs a
person: to sign in somewhere, get past a captcha or show it the right page. By the end you'll know where the
browser is on a ticket, what each control does, and where its screenshots end up.

Each ticket gets its own browser, a Chrome that runs on the machine the agent runs on. The agent opens it to try
a change the way a user would, or to read a page it needs. It's the ticket's task browser, never anyone's personal
browser, so what you do in it stays with that ticket.

## Open it

Open the ticket. Above the agent's work is a row of views: **Feed**, **Terminal**, **Browser**, **Desktop**,
**Code**, **Changes** and **Processes**. Choose **Browser**.

Two other ways in:

- On the board, a card whose agent has a browser open shows a small live preview of it. Click the preview to open
  the ticket on **Browser**.
- **Split** shows two views side by side, and starts with **Feed** beside **Browser**. That's the easiest way to
  read what the agent says while you watch what it does.

Until the agent opens a page, the view says **Waiting for the agent to open a browser…**. If it says **No live
browser**, By Dawn can't reach the machine the ticket runs on, for example because that machine is offline.

**Desktop** is the same idea for the agent's whole screen rather than one browser. It only appears when the
ticket's machine can share its screen.

## Live view

The **Live view** switch at the top of the browser turns the picture on and off. It starts on, and By Dawn remembers
your choice for each ticket. Turning it off stops the stream to you but not the browser: the agent carries on
either way. You need it on to take over.

## Take over

Choose **Take over**, or click anywhere on the page where it says **Click to take control**. The button changes to
**You're driving** and the page gets a coloured border. Your clicks, typing and scrolling now go to the browser,
and the agent waits behind you until you hand it back. Choose **You're driving** again to give control back.

Only one person drives at a time. If someone else takes over while you're driving, By Dawn tells you **Another viewer
took control**.

Copy and paste work between your computer and the browser. If your own browser blocks clipboard access, By Dawn asks
you to allow it, or copies inside the remote page instead.

## The controls

| Control | What it does |
|---|---|
| The address field | Type a URL and press `Enter` to go there. |
| **Back** and **Refresh** | Act on the page, as in any browser. |
| **Remote resolution** | The size the browser runs at, scaled to fit your screen: **Default · 1280×800**, **Desktop · 1920×1080**, **Laptop · 1440×900**, **iPad · 1024×768**, **iPhone · 390×844** or **Custom…** (320 to 4096 pixels a side). Pick a phone size to see how a change looks on a phone. |
| **Maximize** | Fills the window with the browser. `Esc` restores it. |
| **End browser session** | Closes the browser. The agent opens a new one when it next needs it. |

Right-click the page for **Back**, **Forward**, **Reload**, **Copy**, **Paste**, **Select all**, **Copy current
URL** and **Inspect (DevTools)**. In the desktop app, **Inspect (DevTools)** opens the page's developer tools in a
new tab. In a web browser, it shows them inside the view instead, and **Close DevTools** takes you back to the page.

## Sessions and tabs

The browser can hold more than one **session**, each with its own cookies, sign-ins and tabs, so the agent can be
signed in as two different users at once. When there's more than one, a strip above the tabs lists them. **Main**
is the one every ticket starts with, and it can't be closed.

Under it is the tab strip for the session you're looking at. Click a tab to switch to it, or choose **New tab**.

## Browser evidence

When an agent reaches a step it wants to prove, it saves a screenshot of the browser. The ticket collects them
under **Browser evidence**, which shows how many there are, or **None yet**. Open it to step through them in order
with **Previous checkpoint** and **Next checkpoint**.

The screenshots stay on the ticket after the browser closes. When the ticket reaches review, **Browser evidence**
is badged **Review** and opens by itself, because it's often the fastest way to check the work.

In the desktop app, a ticket's browser can also run on your own computer, through [the local agent](https://docs.bydawn.com/local-agent/).

## FAQ

**Can the agent see what I do while I'm driving?**

It sees the page as you leave it, but it doesn't act while you're driving. It waits behind you and carries on from
wherever the page is once you hand control back.

**Is it safe to sign in to a site in the agent browser?**

The sign-in is saved in that ticket's browser session, on the machine the ticket runs on, so the agent can use it
for the rest of the ticket. Sign in with an account meant for the work, not a personal one.

**Why don't I see a Browser view on a ticket?**

The view only appears where By Dawn can reach the machines tickets run on. The desktop app always can. If you're using
By Dawn in a web browser and don't see it, use the desktop app. **Browser evidence** shows on every ticket either way.

**Does turning Live view off stop the agent?**

No. It only stops the picture coming to you. The agent and its browser carry on.

## Where to go next

- [Terminals](https://docs.bydawn.com/terminals/): the other way to watch an agent and step in.
- [Remote agents and machines](https://docs.bydawn.com/remote-agents/): the machines tickets and their browsers run on.

Source: https://docs.bydawn.com/agent-browser/