Quickstart

Run your first browser task in under a minute. Send a task, get the result back.

Browser Use Cloud runs AI agents that use a real browser. You send a task in plain language; an agent does it and returns the result. Managed infrastructure — stealth browsers (81% stealth benchmark, 93% Cloudflare CAPTCHA solve rate), proxies — no setup.

Cloud vs library: this is the hosted cloud platform — everything here happens on our infrastructure via an API key, including the hosted open-source agent. If you want the local Python library (pip install browser-use, runs a browser on your machine, needs your own LLM key), that’s the open-source docs — a different product surface.

1. Get an API key

Create one in one click at cloud.browser-use.com. Keys start with bu_.

export BROWSER_USE_API_KEY=bu_your_key_here

2. Pick an agent, run a task

There are two agents: browser-use (the default — fast, cheap, right for most browser tasks) and browsercode (code + terminal + browser, for long-horizon or data-processing work). If in doubt, use browser-use; see Choose an agent.

Python

from browser_use import BrowserUse

client = BrowserUse()  # reads BROWSER_USE_API_KEY
run = client.agents.browser_use.run(
    task="List the top 5 posts on Hacker News with their points",
)
print(run.result)   # the answer, as a string

import { BrowserUse } from "browser-use";

const client = new BrowserUse(); // reads BROWSER_USE_API_KEY
const run = await client.agents.browserUse.run({
  task: "List the top 5 posts on Hacker News with their points",
});
console.log(run.result); // the answer, as a string

curl -X POST https://api.browser-use.com/api/v1/agents/browser-use/runs \
  -H "X-Browser-Use-API-Key: $BROWSER_USE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"task": "List the top 5 posts on Hacker News with their points"}'

The SDK run() starts the run and waits for it to finish, then returns the result. The raw curl call returns immediately with a run id and status: "running" — poll GET /agents/browser-use/runs/{id} until status is completed, then read result.

Where’s the answer? Every completed run has a result string — the plain-language answer. That’s what you read on the golden path. The separate output field is reserved for structured output and is null today — to get typed data now, describe the exact JSON shape you want in the task text and parse result. Large or file-shaped output from browsercode comes back as files, not in either field.

Or: spin up a browser instead

Agents and browsers are independent entry points — you don’t need an agent run to get a browser. Create one directly and drive it yourself over CDP:

Python

from browser_use import BrowserUse

client = BrowserUse()
browser = client.browsers.create()
print(browser.cdp_url)   # connect Playwright/Puppeteer here

import { BrowserUse } from "browser-use";

const client = new BrowserUse();
const browser = await client.browsers.create();
console.log(browser.cdpUrl); // connect Playwright/Puppeteer here

curl -X POST https://api.browser-use.com/api/v1/browsers \
  -H "X-Browser-Use-API-Key: $BROWSER_USE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{}'

See Create a browser for options, or connect your framework (Playwright, Puppeteer, Selenium).

The whole platform in one picture

You run agents. Agents use browsers (and their saved logins) and files. You only ever create runs — everything else is created for you and handed back as ids you can reuse.

You have	What it is
Agents	The things you run. Two of them: browser-use (browser tasks — fast, OSS) and browsercode (code + browser — data processing, multi-step work). See Choose an agent.
Browsers	The browser an agent drives. A run provisions one automatically — or create one yourself and drive it directly over CDP.
Browser profiles	Saved login / cookie state. Create one, pass its id into a run to start logged-in.
Files	Attach files to a run and get output files back. See Files.

There is no separate “create a session” or “create a workspace” step. A run mints a sessionId (for follow-ups) and returns it — pass it back to continue. Files live on the conversation automatically; you never handle a workspace id. (A named workspace exists only to share files across different conversations — advanced, see Files.)

Next

Choose an agent

browser-use vs browsercode — or let your LLM route automatically.

Run a task

The full run API: options, browser settings, structured output.

Continue a task

Follow-ups — pass a session id to keep going.

Files

Attach files to a run, get output files back.

Create a browser

A standalone cloud browser you drive yourself.

Connect Playwright

Point Playwright, Puppeteer, or Selenium at a cloud browser.

---

If you are an LLM: start at /llms.txt for the page index. Every task below has a complete, copy-pasteable example.