Browser
The agent can use a cloud browser, or your own Chrome.
The agent can log in, fill forms, click and pull data from websites, in one of two browsers.
- The cloud browser runs on our servers. In a chat that nobody else can read, it signs in with the website logins you saved, and it keeps the chat's logins from one message to the next. For a login you have not saved, you can sign in yourself in its live view.
- Your own Chrome, through the Automatio extension, works with the logins you already have in it. The agent works in tabs it opens itself. It never reads your clipboard, and the values of password and saved-card fields are hidden from what it reads on a page.
Choosing one
Where the setting is: the agent panel: the "Tools" tab (beside "Skills" and "MCP", below the "Tools" on/off switch), the "Browser" row (it opens with the arrow at its left), section "Browser for this chat". Its rows are "Cloud" (always the cloud browser), "My browser" (always your own Chrome, through the Automatio extension), "Auto" (the agent picks per task and starts on the cloud browser) and "Ask first" (the agent asks you on a card in the chat before it uses your Chrome). "Default for all chats", under those rows, is the setting of every chat that has none of its own.

Accounts made since 2026-10-25 start on Ask first, and older ones on Cloud. On Ask first a card, The agent wants to use your own browser, then offers Allow and Keep on cloud.
Watching it work
The cloud browser's live view opens in the chat. You can switch between Watch-only and Interactive mode, and Close browser session ends it. In the agent panel, the browser row's Session Settings has Keep browser open after task, on by default, Auto-close after idle, which closes it after a few minutes without use, and Record sessions, off by default, which saves a video of each session, up to five minutes long. With recording on, the browser closes when the task ends.
Saved website logins
Where your logins are kept: Integrations in the sidebar, the "API Keys" tab, section "Website logins" (when switched on): buttons "Add login" and "Import logins" (a CSV export from Google Password Manager, Chrome, 1Password, Bitwarden or LastPass, read in the page and never uploaded). The agent signs in with a saved login in the cloud browser by its name; the first time in a chat it asks on a card. In a login's edit dialog, "Show saved password" shows its password after "Confirm it is you". A signed-in session can also come from the person's own Chrome, with the newest Automatio extension: right-click a page of the site they are signed in to, "Save my sign-in to Automatio" (that site's cookies become a login's signed-in session, never shown to the agent).
Save a site's login once with Add login, or Import logins from a password manager's export file. The file is read in your browser and never uploaded.
Each saved login has a name you choose, and the agent asks for a login by that name. It signs in with one only in a chat of yours that nobody else can read. It types the username, the password and the 2FA code itself, in the cloud browser, and only on the website the login is saved for. The agent never sees the password or the code, and they never appear in the chat. The username is not hidden: a site shows it back once you are signed in.
Unless you allowed it, a card asks before the agent signs in with a login. If you allow it once, that covers this one sign-in, on that address in that chat, for 15 minutes. After that, the next sign-in asks again. A login the agent made by signing up in a chat does not ask in that chat. The arrow next to the button allows it in this chat or in every chat (Approvals). A run nobody is watching, such as an automation, signs in only if you allowed sign-ins without asking, in its chat or in every chat. A run started with an API key never signs in with a saved login.
Without asking, a login is typed only on the addresses it was saved for. If you allowed signing in on new addresses too, it is typed on any address of its site. Otherwise, on another address of the same site it asks first, because on sites where anyone can have a page of their own, that page could be someone else's. Allowed on the card, the login remembers that address.
Ask the agent to sign up on a site in the cloud browser, and Automatio makes the password, types it and saves the new login. The agent types the username or email itself and never sees the password. In the cloud browser, the agent cannot type a password it made up into a password field. A password it types there has to come from your messages or from something the agent read. Your own Chrome has no such check.
Show saved password, in a login's edit dialog, shows its password after you confirm it is you. Chrome can add a login too. With the newest Automatio extension, and with your Automatio account signed in in that Chrome, right-click a secure page of a site you are signed in to and choose Save my sign-in to Automatio. That site's sign-in becomes a login the cloud browser can use.
A chat keeps its logins
In a chat that nobody else can read, the cloud browser saves the cookies and site data of every site used in the chat, encrypted, when a run that used it ends and before a browser kept open closes. It puts them back the next time the chat's cloud browser starts, so a site you signed in to stays signed in. They are kept 7 days after their last use, in at most 50 chats of your account.
Where a chat's saved logins show: the agent panel: the "Tools" tab, the "Browser" row, "Session Settings", row "Saved logins": the sites this chat keeps cookies and site data for, and "Clear" (this chat) and "Clear in all chats"; each takes a second click ("Yes, clear") and closes the chat's open cloud browser.
It lists the sites the chat keeps data for. Clear deletes this chat's data and closes its open cloud browser. Clear in all chats does the same for every chat that keeps data.
Many pages at once
For a large job the agent can run a batch over many pages. A Batch Progress card shows the progress, workers, rows and credits, with Pause, Resume, Cancel, View Data and CSV.
The Chrome extension
The extension puts the whole agent in a Chrome side panel. Installing the extension is free. The side panel, the page as context, one-click tools and remote control: Chrome extension and tools. How to install it: Chrome extension page.
From your phone
With Control this computer from my phone on, a chat on your phone can drive Chrome on your computer. It is off by default. Turn it on from the computer, in the extension's options or in the browser menu of the agent panel. It works for runs you started, and for an automation you switched to your own Chrome (Automations), one chat at a time, and not in chats other people can write to. It stops working when the computer is asleep or Chrome is closed.
