What is the Tallos built-in browser?
The Tallos built-in browser is a Chromium-based browser that lives in a pane inside each workspace. It has an address bar with history, back, forward, reload and hard reload, find in page, devtools and its own tabs.
Tabs are scoped to the workspace. Each parallel workspace runs its own dev server on its own branch, so each gets its own browser tabs. Switch workspaces and Tallos restores that workspace's tabs and scroll positions. You never mix up the preview of one agent's branch with another's.
Drag the browser next to the agent's terminal and a diff, and you have the whole loop on one screen: the agent edits, the page reloads, you look.
How do I send a page element to an agent?
Use the browser's design tools. Instead of describing a bug in words or pasting a full-page screenshot, you point at the exact element and Tallos packages the context an agent needs.
- 1
Grab page element
Turn on grab mode and hover or click an element. Press
Cto copy its context for your agent, orSto take a screenshot of it. - 2
Annotate page element
Click an element, mark it as a Change or a Question, and write what the agent should do. Repeat on as many elements as you like, then Send feedback to an agent.
- 3
Draw on screenshot
Capture the page and mark it up with pen, arrow, rectangle, ellipse, highlighter or text, then copy the markup into your agent's prompt.
| Context | Details |
|---|---|
| Markup | An HTML snippet of the element, its CSS selector and element path, attributes and nearby elements |
| Computed styles | Display, position, size, margin, padding, colors, border, radius, font family, size and weight, line height, text alignment, z-index |
| Accessibility | Role and accessible name, including ARIA labels |
| Source hints | React component names and the source file when the page exposes them |
| Page | URL, title, viewport size and scroll position at capture time |
| Image | A cropped PNG screenshot of the element |
How do I fix a UI bug with the built-in browser?
- 1
Open the page
Open the workspace's browser and go to the page with the bug, usually your local dev server.
- 2
Point at the element
Annotate the broken element and write the fix: "this padding is too tight, match the cards above."
- 3
Send it
Send the feedback to the agent working in that workspace. It gets the markup, styles and screenshot of exactly what you pointed at.
- 4
Check the result
The agent edits the source and your dev server reloads the page. Point again if it is still wrong.
- 5
Review and ship
When it looks right, read the change in diff review and commit.
This loop is why a browser belongs inside the environment. It is the fastest way to steer an agent on front-end work, and a big part of how people practice vibe coding without losing control of the result.
Point, send, ship
Open a workspace, start your dev server and send your first element to an agent.
What are browser profiles?
A browser profile is a separate browser identity with its own storage partition: cookies, local storage and cache. Profiles never leak into each other, so you can test as an admin in one tab and as a regular customer in another.
- Create profiles in Settings → Browser, optionally seeded with cookies and a viewport size.
- Import cookies from Chrome, Edge or other browsers so you and your agents reuse existing logins. Tallos only replaces cookies for the domains in the import.
- Google sign-in is not imported; sign in to Google directly inside Tallos.
- Agents inherit the profile. Browser commands an agent runs use the tab's active profile.
Can AI agents use the built-in browser themselves?
Yes. Agents drive the same browser you see through the `tallos` CLI. The loop is snapshot, act, snapshot again: tallos snapshot returns an accessibility tree with element refs, and the agent clicks, fills and presses keys by ref.
tallos goto --url http://localhost:3000/login
tallos snapshot # returns refs like e1, e2, e3
tallos fill --element e1 --value "[email protected]"
tallos click --element e3
tallos set device --name "iPhone 12"
tallos screenshot --format pngTallos installs the tallos-cli skill into your detected agents, so they already know these commands. For native apps outside Tallos, agents use the separate computer-use skill instead.
What else does the built-in browser do?
- Viewport emulation. Set a custom viewport on a tab to test responsive layouts; the page sees the emulated size in
window.innerWidthand media queries. - Link routing. Choose whether links from the terminal, Markdown and editor open in Tallos or your system browser. Hold
Shiftwith the modifier-click to flip it for one link. - New tabs stay inside. Links with
target=_blankand ordinary popups open as new Tallos tabs with the same profile. - Downloads shelf. Active and recent downloads appear under the toolbar with open, show in folder and cancel actions.
- Remote workspaces. In remote workspaces, you choose whether browser traffic goes through the SSH host or this computer.
Your everyday browser
- Your extensions, bookmarks and personal sessions
- One tab list for everything you do
- Describe the bug or paste a screenshot to the agent
Tallos built-in browser
- Tabs scoped to each workspace and branch
- Point at an element; the agent gets markup, styles and an image
- Agents can drive it themselves from the terminal
Frequently asked questions
Does Tallos have a built-in browser?
Yes. Every workspace has its own Chromium-based browser pane with tabs, history, find in page and devtools. Tabs are scoped to the workspace and restore when you switch back.
How do I show an AI agent which element to fix?
Use Grab or Annotate in the Tallos browser. Click the element and Tallos sends its HTML, computed styles, accessibility info and a cropped screenshot to the agent with your note.
Can I test logged-in pages in the Tallos browser?
Yes. Import cookies from Chrome, Edge or other browsers into a browser profile. Google logins are not imported, so sign in to Google directly inside Tallos.
Can Claude Code or Codex click around the page on their own?
Yes. Through the tallos CLI an agent can navigate, take an accessibility snapshot, click, fill inputs, press keys, emulate a device and take screenshots in the same browser you see.
Can I test mobile layouts in the built-in browser?
Yes. Set a custom viewport size on a tab, or have an agent run tallos set device --name "iPhone 12" to emulate a device before taking screenshots.
Do browser profiles share cookies?
No. Each profile has its own storage partition for cookies, local storage and cache, so logins in one profile never leak into another.
