Give a bot a computer

A bot can work without a computer, use the computer running Brainwrite, or work inside an isolated desktop.

Built-in browser

Bots can use a built-in browser tab that you can watch and take over. It is on unless you turn it off, and each bot also has its own browser switch.

To turn it off for this installation, open Settings → Computers → Built-in browser (in Advanced mode, Settings → Local VM).

Watching and taking over

Open the Browser panel to watch the bot's browser live. Hovering over it changes nothing. A click, a key or a scroll in the panel takes the browser from the bot: its browser tools pause while you use it, and a few seconds after you stop, control goes back to the bot and the panel says Control is back with the bot's name. To keep the browser for longer, for example while you sign in, choose Take control, then Hand back when you're done.

If one of the bot's browser actions stopped answering, you can still take control: Brainwrite first checks whether the browser responds. If it does, you get control; if it doesn't, Brainwrite asks you to restart the browser first. If another window or device already has control, the panel says Another window or device controls this browser. Hand control back there to use it here.

When a page asks its own question (a confirmation or a box to type into), the panel shows it with Dismiss and Accept, and a text box when the page asks for text. Whoever controls the browser answers it; anyone else watching sees Take control to respond. When the page didn't take something you clicked or typed, the panel says The page did not take some of your input. Check the page, then try again. and you carry on. When a page stops responding to your input altogether, Brainwrite restarts the browser for you and every view reconnects: The page stopped responding, so the browser was restarted. Your sign-ins are kept; open the page again.

Restarting the browser

Restart browser…, in the Browser panel's menu, closes the profile's open tabs and keeps its saved sign-ins. Stop the bots using the browser first. You can restart without stopping them while the browser is interrupted, or while every bot using it is waiting for you to sign in.

Sign-ins

A bot's browser keeps the sites you sign in to in it, and restarting the browser keeps them too. A bot's own browser, and each shared browser profile, runs on its own Chrome profile on disk, so cookies and what sites store in the browser (where apps such as Slack, Discord and WhatsApp keep their sign-ins) survive a restart, the browser closing after sitting idle, Brainwrite restarting, and a crash. A temporary browser keeps nothing.

Attaching files

A bot can attach a file from this computer to a web page, for example an image to a post, wherever the page has a file upload. It can use any file your account can open, so give it the path of the file you mean. Attaching a file follows the bot's approval level, like its other browser actions: in Ask mode it asks you first, and with Full access it attaches without asking.

When a browser sits unused

After 10 minutes without use, a bot's browser closes to free memory. Its tabs close; a browser profile keeps its saved sign-ins, while a temporary browser loses them too. The next time the bot uses its browser, it opens again and the bot is told that its earlier pages are gone, so it opens the page it needs again.

A browser never closes this way while a bot's turn is using it, while you have control of it, or while its Browser panel is open. A bot that uses your own Chrome never has it closed this way.

Page errors

After each click, typing or other action in the built-in browser, the bot is told about failed requests, uncaught errors and console errors the page reported, as page output rather than instructions. This helps it notice when a page did not work.

Recording browser steps

To keep a record of what a bot did in its browser, open the bot's settings (in Simple mode, choose All settings first), go to Access, and under Browser turn on Record browser steps. The browser must be on for that bot first.

While it is on, each click, typing and page change is saved as a numbered step, with a screenshot and a short note, and any errors the page reported. Steps are kept with the conversation for 30 days. No screenshot is taken while a password or other secret is shown unmasked on the page, for a minute after a one-time code was typed, or while you have control of the browser. If the bot uses your own Chrome, the screenshots show your Chrome.

Workspace backups leave recorded steps and their screenshots out.

Auto, local, and cloud selection

Computer selection is explicit per bot. Auto may reuse a ready, verified backend, but it does not silently provision every backend or turn on host control.

Preview and takeover

The Computer panel can show live frames while an agent works. Supported cloud desktops also provide an interactive viewer. Brainwrite opens that viewer in the app when possible, with a browser fallback for external join flows.

When a bot needs you to sign in to a website, it asks with a Sign in needed card instead of asking for a password in chat. Open goes to the computer or browser the bot is working in, with Take control; Done hands it back and resumes the bot. See Website sign-in.

In the built-in browser (and in My Chrome) a bot can also sign in with a login you saved, from the card's Save a login or Settings → Logins, or with one from a connected 1Password vault: it calls fill_login on the site's login page, and Brainwrite types the username, password and any one-time code into the page itself and submits the form, only on that login's own site. The bot never sees the password. See Saved logins, including what a screenshot can still show.