All pages
Lesson 6.1 · The browser and your desktop · 12 min
The built-in browser
Preview your app, pick elements, and read the console and network.
TRMNLY has a browser in the right sidebar. Use it to watch the app you are building next to the terminal that runs it, point at an element and hand it to the agent, and read the page's console and network traffic. The agent can drive the same tab: navigate, click, type, take screenshots and run checks.
Time: 12 minutes · Works on: macOS app. The cloud workspace has its own browser tools on a different engine.
Open a page
- Show the right sidebar with
⌘⇧\. - Click the Web tab. It is one of the five tabs shown directly in the strip.
- Click the address bar, which reads URL — try http://localhost:3000, type an address and press Return.
The browser has its own tab strip. Click + to add a tab and the × on a tab to close it. Tabs come back after you relaunch TRMNLY.
Try it with your own project:
- Start your dev server in a terminal pane, for example
npm run dev. - Open its URL in the Web tab.
- Edit a file and watch the page reload next to your terminal.
Pick an engine
The chip beside the address bar picks the tab's engine:
- WebKit: the default, the same engine as Safari.
- Chrome — TRMNLY profile: the Google Chrome on your Mac, with a separate profile that TRMNLY owns. An agent driving this tab can't touch your real browsing.
- Chrome — copy of your profile: a copy of your Chrome profile, with its extensions and logins, taken the first time you choose it. Logins you make in Chrome afterwards are not in the copy.
If Chrome isn't installed, the menu shows Chrome — not installed instead. Switching engines reloads the page, and cookies and logins don't carry over between engines.
Use Chrome when a site only works in Chromium, when you need a Chrome extension, or when you want to check your app in the browser most people use.
Point at an element for the agent
Design mode lets you show the agent exactly which element you mean, instead of describing it.
- Click the crosshair button next to the address bar. Its tooltip reads Design mode — click an element to hand it to the agent.
- Move the pointer over the page. Whatever is underneath gets outlined.
- Click to select it. The page's own click handlers don't fire while you are picking, so selecting a "Delete" button picks it and doesn't press it. Press Escape to leave without choosing.
- A strip under the address bar shows what you picked: its tag, its size, and buttons to Copy the selector or Forget this selection.
- Switch to the Agent tab and ask, for example: "Make the selected button match the primary button style."
The agent receives the element's selector, the styles that differ from a plain element of that type, its HTML, and a cropped screenshot. If the page navigates, the selection is marked as stale.
Read the console and network
- Click the DevTools button. Its tooltip reads Toggle DevTools (Console / Network / Elements).
- The panel has six panes: Console, Network, Elements, Storage, Performance and Egress.
- Turn on Preserve to keep entries when the page navigates.
- In Network, choose Export network log as HAR… to save the traffic to a file.
The agent reads the same logs. Ask "What errors are in the browser console?" and it calls browser_console.
The browser's More menu
The browser has its own More button beside the address bar. It is separate from the sidebar's More menu. It holds:
- Open in Safari and Open in Chrome: hand the page to your regular browser.
- Copy URL.
- Send Page Text to Terminal: types the page's visible text into the focused terminal pane through the command queue. Nothing is submitted. You review it and press Return.
- Clear Browsing Data…
- Import from Browser…: brings your bookmarks bar over from Chrome, Arc, Brave or another Chromium browser as one-click shortcuts.
- Restart Browser: rebuilds this tab's page engine. Use it when a page has frozen and Reload doesn't help. Restart All Tabs does the same for every tab.
- Search Engine: the engine used when you type words instead of an address.
- Zoom In, Zoom Out and Actual Size.
Reader mode, a button in the address bar, strips navigation and ads and shows only the article. The find bar searches the current page, with Next match on ⌘G and Previous match on ⇧⌘G.
Let the agent drive
The agent has a large set of browser_* tools, shaped like Playwright. It can navigate, click, type, fill forms, wait for elements, take screenshots, record a GIF, and check text or URLs. Try:
- "Open localhost:3000, sign up with a test email, and tell me if anything fails."
- "Take a screenshot of the pricing page at phone width."
- "Check the network log for requests that returned 500."
Navigation follows the app's network allowlist. If a site is blocked, the agent asks you to allow that one host.
Saved logins
The browser can store site logins so the agent signs in without seeing the password. Use the address bar's Saved accounts for this site button to manage them. The agent can look up which account matches a site, but the password is filled inside the browser and never returned to it.
To bring in passwords from another browser, export a CSV from it, then use the accounts panel's Import from 1Password / Chrome CSV button.
Check yourself
- Your local dev server loads in the Web tab and reloads when you save a file.
- After picking an element in design mode, the agent can describe the element you picked.
- The Console pane shows a message you log from your page with
console.log.
Go further
- Browser: engines, design mode, every browser tool, sessions and recordings
- Agent safety: the network allowlist the browser respects
- Workflows and tab monitors: replay recorded browser actions with one click