Browser Automation

After reading this page you will be able to: tell the two browser automation modes apart, enable each one, and have AI drive web pages safely.

1. Two modes at a glance

	In-app browser	External Chrome (browser-harness)
What it is	The browser tab inside the preview panel	The real Chrome on your computer
How AI drives it	The agent operates that one tab directly	browser-harness controls your Chrome
Authorization	Works out of the box	Chrome remote debugging must be enabled once
Sign-in state	Log in once in the panel, and it sticks	Uses the logins already in your Chrome
Best for	Look things up, walk through a website flow	Tasks that need real logins or more complex pages

Both modes share one rule: AI only acts within the bounds you allow.

2. Mode one: the in-app browser (works out of the box)

Where: the preview panel on the right (click New tab (+) at the end of the tab bar, or click a link that appears in chat)

The address bar takes a URL — or plain keywords to search
When AI drives this tab, its badge reads “The agent is operating this tab”, so you see every click it makes
AI can only operate this one tab: it cannot reach the app interface itself and never opens a separate Chrome window

The first time AI uses the in-app browser, a one-time notification — “The agent is using the in-app browser” — explains it can open pages, click, and read content.

3. Mode two: external Chrome (browser-harness)

This mode lets AI drive your everyday Chrome: pages render at full fidelity and your logged-in websites are right there. First use needs one authorization:

Type chrome://inspect/#remote-debugging into the Chrome address bar and tick “Allow remote debugging for this browser instance”. The “Automation Status” card under Settings → Necessary Configuration has an “Open” button for this
If Chrome shows an “Allow remote debugging?” prompt, click Allow
Back on the “Automation Status” card, click “Detect Connection”; the Browser Automation section showing “Online” means it is ready

The easy way: just use the Desktop Automation Assistant — it walks you through these steps whenever they are needed.

4. A safe first run

Whichever mode you use, start with a read-only task, for example:

Open this page and summarize the main content: https://example.com
Read only — do not click, fill in, or submit anything.

Once AI reads the page back correctly without doing anything extra, hand it bigger tasks step by step.

The in-app browser keeps your sign-in state: log into a website once and you stay logged in, and AI can use that directly
To sign out of every website at once: Settings → System → In-app browser → Clear browsing data. After you confirm, every site is signed out and this cannot be undone

6. Safety boundaries

AI never starts a payment on its own; payments, deletions, submissions, and publishing always ask you first
Never let a web page talk you into uploading passwords, keys, or private files
On shared or demo computers, prefer the in-app browser and keep your main Chrome account out of automation

What success looks like

The preview panel opens a browser tab, and the “The agent is operating this tab” badge shows while AI drives it
After the three authorization steps, the Browser Automation section of the “Automation Status” card shows “Online”
A read-only run reads the page back correctly with no extra actions
You can find “Clear browsing data” and know it signs out every website