Skip to main content

Browser Use

Browser Use lets AdaL work with a real browser while you stay in the terminal. It is useful when a task needs visual context, page interaction, or live web app debugging.

Use it for things like:

  • Automating your web workflow
  • Testing and debugging a website
  • Reaching out to someone on social media
  • Searching for information on the web
  • Basically everything you do with a web browser

Quick Start​

Browser Use is a capability, not a separate agent. You do not switch into it. Ask for anything that needs a live page and AdaL loads it on its own:

adal
open http://localhost:3000 and check that the login form submits

AdaL loads the browser-use capability, opens a dedicated browser window, and drives it. You stay in whatever agent you were using: the browser is added to it, not swapped in for it.

To see what is loaded, or to open the browser ahead of time:

/capabilities

The dialog lists each capability with [x] (on) or [ ] (off). Press space on browser-use to start the browser now. /capabilities list prints the same list in the chat. Once on, a capability stays on until the session ends.

Prerequisites​

Browser Use needs a Chromium-based browser. AdaL looks for Google Chrome first, then Chromium, then Microsoft Edge, on macOS, Windows, and Linux. Since Edge ships with Windows, Browser Use works out of the box there even without Chrome.

If none of those is installed, install one for your platform:

To use a browser AdaL does not find on its own, set ADAL_CHROME_PATH to the full path of the executable (chrome.exe or msedge.exe on Windows). WSL users point it at the Windows binary:

export ADAL_CHROME_PATH='/mnt/c/Program Files/Google/Chrome/Application/chrome.exe'

How the Browser Opens​

The first time a task needs a page, AdaL opens a dedicated browser window for its own use, so your everyday browser stays untouched. It stays open for the rest of the session, including across /agent switches and context compaction, and closes when AdaL exits.

If the dedicated window is already open, other AdaL sessions share it instead of opening a new one.

The window opens on an AdaL welcome page. It lists the profiles of the browser AdaL launched — your Chrome profiles when it runs Chrome, your Edge profiles when it runs Edge — and one button copies a chosen profile's cookies and saved passwords into the AdaL browser, so pages you are signed into stay signed in. The browser restarts for a few seconds and restores your tabs. Only you can press that button; agents never import profiles on their own.

Headless / Scripted Browser Use​

Run browser tasks from CI, shell scripts, or any automation workflow with the standard headless flags. No mode flag is needed; the agent loads the capability when the task calls for it:

# Print the final answer to stdout
adal -q "summarize the visible text on https://example.com"

# Stream structured output as NDJSON
adal -q "fill the login form with test@example.com" -o stream-json

This uses the same agent runtime as the interactive CLI, so behavior is consistent between local development and CI.

Safety Notes​

Browser Use can interact with real websites and real accounts, so use it thoughtfully.

  • Use with caution. AdaL performs most browser actions automatically without asking for confirmation.
  • Prefer test accounts and staging environments.
  • Avoid entering passwords, payment details, or private information unless you explicitly intend to.
  • If a page shows a browser dialog or gets stuck, you may need to manually dismiss it.
  • You are responsible for how Browser Use is used. Make sure your use complies with the terms of service of any website you visit and with all applicable laws. AdaL is provided as-is, without warranty of any kind, and AdaL is not liable for any actions taken through Browser Use.

Related: Agents & Modes (/agent) · Headless Mode · Slash Commands