Review a page with your agent

On this page

In Browser, you can select elements on a web page and attach comments to them. Sending those comments to your agent includes the page URL and details identifying each element, so the agent can find what you mean. Browser runs on the selected devbox or Template box, and you and the agent share its tabs and signed-in sessions.

Open the shared browser

In the desktop app, select the thread where you want to send feedback. To review a page on the Template box, select a thread running there. If the machine is sleeping, wake it from the machine status control in the titlebar.
To review your own app, make sure it's running on that machine: ask the agent to start it, or start it yourself in Terminal.
Open Browser from the titlebar.
Enter your app's address or another website's URL. For an app running on port 3000 on the selected box, enter http://localhost:3000.

In Browser, localhost means the selected cloud machine. To test the app in your laptop's own browser instead, follow Preview your app. You can open several Browser tabs or expand the pane for more room.

You can also open Browser from a thread in the mobile app. It shares the same tabs and resizes the browser to fit your phone while you use it; see Use Browser from your phone.

Before signing in to a website, remember that Browser's cookies, site data, and saved sessions live on the selected box. Signing in will also give software and agents on that machine access to the session.

Websites see Browser's traffic coming from a cloud data center, and some services, such as npm, treat sign-ins from data center addresses as unusual and may restrict access. If a command-line tool such as npm login prints a sign-in link, you can also open that link in your laptop's own browser. If you don't want Browser in a project at all, you can turn it off.

When an agent needs you in Browser

Some steps need you, such as signing in to a website or approving a passkey. When an agent hands you one of these, it can put a small blue dot on the Browser button in the titlebar, the same dot that marks unread threads in the sidebar. The agent's reply says what to do there. The dot goes away when you open Browser on that box or send your next message in that thread. The mobile app doesn't show it.

Comment on a page

Select Review page elements in the Browser toolbar to enter Reviewing mode.
Hover over the element you mean, then click it.
Write your comment, then select Add comment (the up-arrow button).
Repeat for other elements, then select Done to finish selecting. Done doesn't send anything: your comments wait in the Browser review attachment in the thread's message composer.
Open the attachment to check your comments, then select Send in the composer. You can send the review with or without a separate message.

For example, suppose a button in your app shows "Sub…" because its label is clipped at narrow widths. In Reviewing mode, click that button and write, “Make the full button label visible when the window is 600 pixels wide.” The agent will receive your comment, the page URL, and the element's selectors, visible text, and position, which help identify the button you mean.

Change or discard unsent comments

Each unsent comment has a numbered marker on the page. Select a marker to edit or delete its comment, or open the Browser review attachment to inspect the comments and remove individual ones. To discard the whole review, select the trash button in the Reviewing toolbar and confirm with Discard all.

Unsent comments stay with their thread when you navigate elsewhere or switch threads. Markers appear only on the page where you selected the elements. If that page changes and an element can no longer be identified, its marker may disappear; the unsent comment will remain in the attachment.

If a website in Browser asks for a passkey, you can use Touch ID, Windows Hello, or another authenticator on your laptop. Enable Desktop app settings → Browser → Allow sites to use a passkey on this computer first; this is off by default. A passkey request after your own input in Browser may also offer to turn it on.

Complete the sign-in yourself in Browser. A passkey request goes through only after your own recent input in Browser, and each request opens a separate confirmation on your laptop. Read that confirmation carefully: a compromised devbox could misrepresent the requesting site. The private key will stay on your laptop, but the signed-in browser session will be shared with your agent. Agents can't request or complete the passkey flow themselves.

If the setting is off or a passkey isn't available, the website may offer another sign-in method, such as a password or emailed code.

Recover a browser problem

Choose Restart browser in the Browser toolbar and confirm if Browser can't start or becomes unavailable. Browser will open one blank tab and keep your sign-ins and saved browser data. To reopen your current tabs instead, check Restore open tabs before confirming. The choice starts unchecked each time. On machines awaiting the browser update, the confirmation will say that open tabs will reload and won't offer this choice yet.

You can also ask your agent to restart an unavailable Browser. The agent's restart will reopen the remaining tabs and keep your sign-ins and saved browser data. Reloading tabs can lose unsaved form entries.

Use Reset browser profile, also in the Browser toolbar, only when you want to clear the selected box's tabs, cookies, site data, and saved sessions. Confirming Reset profile will remove that data, and websites there will treat you as signed out.

If Browser can't connect

Browser streams the remote page as live video. It connects straight to the box when it can and otherwise goes through a relay, which also works on networks that block most video traffic. If the stream can't connect after a second try, Browser will show Couldn't connect to Browser with a Try again button. Check your network connection, then choose Try again.

If Browser says the box needs an update, boxes.dev will update the box in the background; choose Try again after a few minutes. If the page stops updating while you work, Browser will show Browser stopped updating; choose Try again to restart the stream.

Show Chrome's own window

To use Chrome's own tabs, address bar, and menus in place of Browser's, open Account from the settings menu and turn on Advanced → Chrome window. The next time Browser connects, it will restart the selected box's browser, reopen your tabs, and show the whole Chrome window. Your sign-ins and saved browser data will stay. The mobile app shows the same switch under Settings → Advanced, and it changes Browser in both apps.

In Chrome window, a sign-in pop-up or another window that a site opens will appear as a tab in the same window. When a sign-in pop-up finishes, Chrome will return to the page that opened it. If a pop-up still opens as a separate window, select Close pop-up at the bottom of Browser. Review page elements, downloads, and the More menu stay in the Browser toolbar. If the selected box can't show Chrome's window yet, Browser will keep its standard view and show a notice.

Upload and download in Browser

Browser supports website file pickers, downloads, permission prompts, and dialogs. Files downloaded from a website stay on the selected box. Use Files to copy them to your laptop, or to upload a file to that machine before choosing it in a website's file picker.

In the desktop Browser toolbar, select Show downloads, then Open beside a completed file to view it. Browser keeps downloaded files on the box even after an agent's browser connection ends. If the agent uses another download folder, Browser will keep a separate copy; large files may stay in progress while that copy finishes. A failure to save Browser's copy will appear in the download entry, even if the agent saved its own file successfully.

These copies use disk space and aren't automatically deleted when they leave the recent-download list. After an agent disconnects, later downloads can also stay in the box's Downloads folder. Use Open to find the retained file, then delete it in Files when no longer needed; delete the other copy too if one remains in Downloads.

Turn off Browser for a project

Open Project settings → Browser and turn off Use the Browser. Browser will be hidden on every box in that project, in both the desktop and mobile apps, and agents there won't be able to use it. Each box will keep its tabs and sign-ins for when you turn Browser back on.