Browser Use Cloud gives you four approaches: AI Agent Tasks, Direct Browser Control, Skills, and the Open Source Library with cloud browsers.
When to use what
| Use case | Approach | Best for |
|---|
| ”Do this for me” | AI Agent Tasks | Most users. Natural language in, structured data out. |
| Repeated extraction, production API | Skills | High-volume, deterministic. $0.02/call. |
| Custom Playwright/Puppeteer scripts | Browser Infrastructure | Developers who want raw CDP access with stealth infra. |
| Already using the open-source library | Open Source + Cloud | Keep your agent code, swap in cloud browsers. |
Start with Tasks — it’s the fastest path to value. Move to Skills when you need a repeatable, production-grade endpoint.
AI Agent Tasks
Give natural language instructions — the AI operates the browser for you.
from browser_use_sdk import AsyncBrowserUse
client = AsyncBrowserUse()
result = await client.run("Search for top 10 Hacker News posts")
print(result.output)
Direct Browser Control
Full programmatic access via Chrome DevTools Protocol. Write your own automation with Playwright, Puppeteer, or Selenium.
from browser_use_sdk import AsyncBrowserUse
client = AsyncBrowserUse()
browser = await client.browsers.create(proxy_country_code="us")
# Connect via browser.cdp_url with Playwright/Puppeteer/Selenium
Skills
Describe what you need in plain text — get a production-ready API endpoint you can call repeatedly.
from browser_use_sdk import AsyncBrowserUse
client = AsyncBrowserUse()
result = await client.skills.execute(skill_id, parameters={"X": 10})
print(result)
Open Source Library
Already using the open-source browser-use Python library? Point it at cloud browsers with one flag — get stealth infra, proxies, and CAPTCHA solving without changing your agent code.
from browser_use import Agent, Browser
from langchain_openai import ChatOpenAI
# Just set use_cloud=True — everything else stays the same
browser = Browser(use_cloud=True)
agent = Agent(
task="Find the top 3 trending repos on GitHub",
llm=ChatOpenAI(model="gpt-4o"),
browser=browser,
)
result = await agent.run()
export BROWSER_USE_API_KEY=your_key
cloud_proxy_country_code and cloud_profile_id for geo-routing and persistent login state. See the open-source docs for full options.
Core concepts
- Task — a natural language prompt the AI agent executes. The main thing you’ll use.
- Session — a stateful browser environment. Auto-created by default, or create one manually for multi-step workflows.
- Profile — persistent browser state (cookies, localStorage). Login once, reuse across sessions.
- Skill — a website interaction turned into a deterministic API endpoint. Create once, call forever.
