Browser Use
Browser Use Overview
Gamut agents can drive a full web browser: navigate pages, click, fill forms, take screenshots, and extract data. Watch live and step in when a task needs a human.
Browser hosts
Gamut offers three browser hosts, selected in Settings > Browser.
| Browser Host | Description | Best For |
|---|---|---|
| Built-in Browser | A headless Chromium browser that runs inside the agent's container. Works out of the box with zero configuration. | Quick tasks, web scraping, form automation, testing. |
| Google Chrome | Connects to your local Chrome installation and can use your existing profiles, cookies, and logged-in sessions. | Tasks that need access to your authenticated accounts without re-logging in. |
| Browserbase | A cloud browser service that runs sessions on remote infrastructure with anti-detection and proxy support. | Scalable automation, avoiding IP blocks, stealth browsing. |
The built-in browser is the default. Changing the host applies to all new browser sessions.
When to use browser automation
Browser automation fits when a site offers no API, or when the task is inherently visual. Common scenarios:
- Researching information across multiple websites.
- Filling out web forms.
- Monitoring a web page for changes.
- Extracting data from sites that only render content in a browser.
- Testing a web application's user interface.
- Navigating internal tools behind a login.
When a service offers a dedicated API or MCP integration, prefer it: APIs are faster, more reliable, and less fragile than UI-based interaction.
The browser panel
When an agent opens a browser, a panel slides open on the right side of the chat. It shows a live view of the browser and supports direct interaction.
Live preview
The panel renders a real-time screencast of the browser viewport. Frames stream over a WebSocket connection and draw to a canvas element, so the view matches what the agent sees with minimal delay. The preview scales to fit the panel width while preserving the browser's aspect ratio.
Tab bar
When the agent has multiple tabs open, a tab bar appears at the top of the panel. Each tab shows its title, and the agent's active tab is marked with a blue indicator dot.
- Click a tab to switch the preview to it.
- Right-click a tab to close it (except the agent's active tab).
- Toggle auto-follow with the eye icon to keep the preview on whichever tab the agent is working in.
Activity log
Below the preview, an activity log lists every browser tool call in the current session. Each entry shows the tool name (such as "Click", "Fill Input", or "Screenshot") with a brief summary of its parameters. Expand an entry to see the full result text.
Controls
A floating control pill at the bottom of the panel provides:
- Pause / Resume: pause the agent's execution, interact with the browser directly, then resume.
- Stop: close the browser entirely. A confirmation dialog appears if the agent is actively running.
- Expand / Collapse: widen the panel for a larger preview, or return it to the default width.
Human-in-the-loop input
Some steps need a human: logging into a site, solving a CAPTCHA, completing two-factor authentication. When the agent encounters one, it calls the request_browser_input tool, which:
- Shows an overlay on the browser preview with a pulsing "Your input needed" indicator.
- Posts a message card in the chat explaining what the agent needs.
- Pauses the agent until you click Done (after completing the step in the browser) or Decline. A decline can include a reason, which steers what the agent does next.
Resizable panel
Drag the left edge of the panel to resize it; the width persists across sessions. The minimum width is 320px and the maximum is 800px.