Browser and computer use

The agent can check its own work. It opens pages in Velork's built-in browser, and with Computer Use on, it can control your Mac's screen or a cloud computer that Velork runs for you.

Use the built-in browser

The built-in browser opens as a tab in the agents panel. Press Cmd + Option + 1 or run Open Browser to open it. You browse in the same tab the agent uses, so you can watch it work. See Built-in browser for the browser itself.

Ask the agent to open, click, fill in, or test something, and it works in that tab:

Open the signup page on localhost:3000, sign up with a test email, and check that the welcome screen shows.

The agent can navigate, click and type, take screenshots, read the console and network requests, run scripts on the page, emulate devices, and inspect storage. After it changes UI code, it opens the running app to check the result.

Point at an element

To show the agent exactly what you mean, click Pick an element in the composer, then click an element in the browser. The element is added to your message with its page, selector, and markup.

The same button picks a layer when you're on the canvas.

Annotate a page

To point out several fixes at once, click Annotate in the browser toolbar or press Cmd + Shift + D. Click anything on the page and type what to change in the note, then press Enter to send it to the chat.

Turn on Batch edits to collect notes first: Enter adds a note to the batch, and Send all (Cmd + Enter) sends them together. Each note shows its status on the page: In batch, Queued, Working, Done, or Failed (click to retry). Press Escape to leave Annotate.

Control your Mac

Turn on Computer Use from the + menu in the composer, or press Cmd + Shift + U. The agent can then take screenshots of your screen, move the mouse, click, scroll, type, press keys, open apps and URLs, and record a flow as a video. It works with any app, not only Velork.

If macOS blocks it, the agent stops and asks you to allow Velork under Accessibility and Screen Recording in System Settings.

Computer Use is available in Develop and Debug mode.

Use a cloud computer

Turn on Cloud computer in the + menu to have the agent use a computer in the cloud instead of your screen. It's a Linux desktop with Google Chrome, a terminal, and a file manager. It is created the first time an agent uses it. Files, installed apps, and browser sign-ins persist between chats, and all your chats share it. Each chat works on its own screen there, so several can use it at once.

Open the cloud computer with Open Cloud Computer in the command palette, or Cmd + Option + 7 in the Agents layout:

• Watching shows what the agent is doing.
• Take control lets you use the computer yourself, for example to sign in to a site.
• Screen lets you pick which screen to follow. Auto follows the chat that is using the computer.

The agent can record a flow on the cloud computer and send you the video in chat, so you can see a result instead of reading a diff.

Settings → Cloud computer shows its state, specs, and disk use, with Start or Stop, Watch live, the minutes used this month, and Reset the computer, which deletes its disk for a fresh start. You can also run Stop Cloud Computer from the command palette; it resumes where it left off.

Cloud computer time counts against your plan's minutes while it runs, and it stops itself after a while without use. See Plans and usage.

Capture a site as inspiration

To reference how another site looks, ask the agent to capture it. It loads the page in the built-in browser and pastes it onto the canvas. See Capture websites.

Next steps