Browser Use

Browser Use turns a task into browser actions. This example runs both the agent and Chromium inside one isolated microsandbox.

The browser image is large. Its first pull and Python package install can take a few minutes. Use a snapshot to avoid repeating that setup.

Run Browser Use

Create the agent

Save this as browser-use-task.py:

import asyncio
import os

from browser_use import Agent, BrowserProfile, BrowserSession, ChatBrowserUse


async def main() -> None:
    browser = BrowserSession(
        browser_profile=BrowserProfile(
            headless=True,
            executable_path=os.environ["CHROME_PATH"],
            args=["--no-sandbox", "--disable-dev-shm-usage"],
            enable_default_extensions=False,
            user_data_dir=None,
        )
    )
    agent = Agent(
        task="Return the title of https://example.com",
        llm=ChatBrowserUse(model="bu-2-0"),
        browser_session=browser,
    )
    history = await agent.run(max_steps=3)
    print(history.final_result())
    await browser.kill()


asyncio.run(main())

Start the sandbox

Export a Browser Use API key, then run the task. The secret stays on the host and can only be substituted into requests to Browser Use’s model endpoint.

export BROWSER_USE_API_KEY="your-key"

msb run --name browser-use-task --replace \
  --cpus 2 --memory 4G --root-disk 10G \
  --secret "BROWSER_USE_API_KEY@llm.api.browser-use.com" \
  --copy-file ./browser-use-task.py:/root/browser-use-task.py \
  mcr.microsoft.com/playwright/python:v1.61.0-noble -- sh -lc '
    apt-get update -qq && apt-get install -y -qq libnss3-tools &&
    mkdir -p "$HOME/.pki/nssdb" &&
    certutil -d sql:"$HOME/.pki/nssdb" -N --empty-password &&
    certutil -d sql:"$HOME/.pki/nssdb" -A -n microsandbox-egress -t "C,," -i /.msb/tls/ca.pem &&
    pip install --disable-pip-version-check browser-use==0.13.10 &&
    CHROME_PATH="$(find /ms-playwright -path "*/chrome-linux/chrome" -type f | head -n 1)" \
      python /root/browser-use-task.py
  '

$env:BROWSER_USE_API_KEY = "your-key"

msb run --name browser-use-task --replace `
  --cpus 2 --memory 4G --root-disk 10G `
  --secret "BROWSER_USE_API_KEY@llm.api.browser-use.com" `
  --copy-file ./browser-use-task.py:/root/browser-use-task.py `
  mcr.microsoft.com/playwright/python:v1.61.0-noble -- sh -lc '
    apt-get update -qq && apt-get install -y -qq libnss3-tools &&
    mkdir -p "$HOME/.pki/nssdb" &&
    certutil -d sql:"$HOME/.pki/nssdb" -N --empty-password &&
    certutil -d sql:"$HOME/.pki/nssdb" -A -n microsandbox-egress -t "C,," -i /.msb/tls/ca.pem &&
    pip install --disable-pip-version-check browser-use==0.13.10 &&
    CHROME_PATH="$(find /ms-playwright -path "*/chrome-linux/chrome" -type f | head -n 1)" \
      python /root/browser-use-task.py
  '

The agent prints its final result and exits. The Browser Use browser session uses the Chromium already included in the image, so it does not download a second browser.

--secret routes egress through microsandbox’s TLS-terminating proxy. Chromium reads its own NSS store, not SSL_CERT_FILE or NODE_EXTRA_CA_CERTS, so without the certutil step above every page load fails with net::ERR_CERT_AUTHORITY_INVALID and the agent burns steps clicking through the interstitial. Importing /.msb/tls/ca.pem into $HOME/.pki/nssdb is what makes navigation clean.

Clean up

Remove the stopped sandbox and unset the API key:

msb rm -f browser-use-task
unset BROWSER_USE_API_KEY

msb rm -f browser-use-task
Remove-Item Env:BROWSER_USE_API_KEY

Reference

⌘I

Overview

AI Agents

Dev Environments

CI/CD & Jobs

Browser & Web Automation

File Processing

Docker & System Tools

Data & Storage

Observability

Run Browser Use

Reference

​Run Browser Use

​Reference

Run Browser Use

Reference