Browser Use lets an AI agent operate a real browser from a task description. You can run its open-source Python library in your own application, connect an agent or coding tool through the CLI, or send tasks to a fully hosted service that runs both the agent and browser. The right route depends on how much infrastructure, authentication handling, model selection, and cost control you want to own.
This guide shows the current setup pattern, explains what each route actually manages, and covers profiles, cloud pricing, bot challenges, reliability, and common failures. The project README is a rolling document, so verify commands and package requirements at the official README before deploying.
What Browser Use actually provides
Browser Use is an automation layer in which an AI agent interprets a goal, chooses browser actions, observes the page, and continues until it produces a result. It is not one single hosting model. The official project presents three routes:
- Python library: embed the agent in your application, define custom tools and structured output, choose a model, and use either a local or cloud browser.
- CLI: give an existing coding or agent tool access to a browser for interactive work or scripts.
- Fully hosted cloud: submit a task through the hosted API; Browser Use runs the agent and browser infrastructure for you.
A cloud browser means the browser process is hosted remotely. The fully hosted API goes further: it also supplies the agent runtime. Keeping that distinction clear prevents an unexpected split between model bills, browser time, and your own infrastructure.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →#1 Best Overall
Which route should you choose?
| Route | Agent runtime | Browser location | Best fit | Main responsibility |
|---|---|---|---|---|
| Python library | Your application and selected model | Local or cloud | Production code, custom tools, structured results | You manage credentials, orchestration, retries, and deployment |
| CLI | Your coding or agent tool | Local or configured cloud browser | Interactive development and scripts | You manage the invoking tool and environment |
| Fully hosted cloud | Browser Use hosted agent | Browser Use cloud | Fastest path with minimal browser infrastructure | You provide tasks, credentials, policies, and monitoring |
Choose the Python library when your workflow needs application-level control, custom functions, or a predictable output schema. Choose the CLI when you want an existing development assistant to operate a browser without writing an integration first. Choose the hosted API when maintaining browsers, networking, and worker processes is a bigger burden than paying usage charges.
Python setup: your first agent
Prerequisites
- Python 3.11 or later, as shown in the current README example.
- A model provider credential accepted by the model you select.
- A browser target: a local browser for development or a Browser Use cloud browser for hosted execution.
- A clear task with an observable result, such as a page title, extracted table, or saved file.
Install the package
The current quickstart uses uv:
uv add browser-useIf you use another package manager, follow the live installation instructions rather than copying an old snippet. The README changes as the project evolves.
Minimal Python program
The exact model class and credential variable depend on the provider. This pattern shows the control flow; replace the model import and environment setup with the provider documented for your account.
import asyncio
from browser_use import Agent
# Import the model supported by your provider, for example:
# from browser_use import ChatBrowserUse
async def main():
agent = Agent(
task="Open https://example.com, report the page title, and return the first paragraph.",
# llm=ChatBrowserUse(model="your-model"),
)
result = await agent.run()
print(result)
if __name__ == "__main__":
asyncio.run(main())The important pieces are the task, the selected model, and the call to run(). Treat the returned value as agent output, not as proof that a consequential action succeeded. For production, validate the result against the page or an independent API before committing a transaction.
Local versus cloud browser execution
A local browser is useful while developing because you can see the session and reuse a machine’s browser profile. A cloud browser is easier to scale and isolate, but its session state behaves differently. Configure the cloud browser using the options in the current README, then test login-dependent workflows separately from public-page workflows.
CLI route for coding and agent tools
The README’s CLI path directs users to install or upgrade the package and register the Browser Use skill with the coding or agent tool they use. Because command names and registration steps can change, use the CLI section of the official README as the source of truth. In practice, the CLI is useful when you want to say “open this portal, export the results, and save the file” from an existing assistant, while keeping your application code unchanged.
For repeatable scripts, put the task text, expected output location, timeout, and authentication assumptions under version control. Do not put passwords or session cookies in the script; inject them through your secret manager or the tool’s supported credential mechanism.
Fully hosted cloud: what you hand over
With the hosted API, Browser Use operates both the agent and browser. You submit a task and consume the result, while the service handles browser workers and the agent loop. This reduces setup, but you still own task boundaries, authorization, data classification, and verification of side effects.
Model inference and hosted browsers are separate services that can incur usage charges. The project states that “The Python library is free and MIT-licensed.” Free software does not mean free model calls or free browser time.
Pricing and budgeting
The vendor’s product page currently lists browser usage at $0.02 per browser hour, plus internet traffic. It lists Browser Use Agents at model cost + 20%, plus browser time and traffic. These are published rates and may change, so confirm the current figures on the product page before setting a budget.
| Cost component | What it covers | Budgeting implication |
|---|---|---|
| Python package | Open-source library | Free and MIT-licensed according to the project README |
| Model inference | Tokens and provider execution | Varies with model, prompt length, and number of steps |
| Browser time | Hosted browser runtime | Currently listed at $0.02 per browser hour |
| Internet traffic | Network transfer for hosted browsing | Additional usage charge; rate is not stated in the supplied pricing details |
| Hosted agent | Browser Use agent service | Model cost + 20%, plus browser time and traffic |
Reduce waste by ending tasks as soon as the required evidence is collected, avoiding repeated page discovery, and caching stable inputs in your own system. Measure your workflow’s actual browser duration and model usage rather than extrapolating from a toy task.
Authentication, profiles, and session state
Local users can reuse a system Chrome profile. That can simplify development, but it also means the automation inherits the profile’s cookies and permissions; use a dedicated profile rather than a personal one.
Free tools Windows power users keep installed
One-click scans. No signup required.
For cloud browsers, profile sync transfers cookies but not local storage, IndexedDB, or extensions. A site that stores login state outside cookies may therefore show a sign-in screen again. Design the task to detect that state and stop safely instead of repeatedly submitting credentials.
- Use least-privilege accounts and short-lived credentials.
- Confirm the target domain before entering secrets.
- Record whether a run used a fresh session or a reused profile.
- Require a human confirmation before irreversible purchases, deletions, or submissions.
CAPTCHAs and bot detection
The README says Browser Use Cloud provides stealth browsers and proxies intended to reduce bot detection and CAPTCHA challenges. It also cautions that outcomes depend on the site and challenge: no browser configuration guarantees that every CAPTCHA can be avoided or solved. Treat a challenge as a normal failure branch. Stop, surface the page state, and use an approved human or site API path rather than trying to defeat an access control.
Tasks Browser Use can demonstrate
The vendor site demonstrates finding cinema seats, completing a pickup cart, exporting portal results to CSV, and website QA. These examples show the range of interactions—search, form filling, navigation, extraction, and checking—but they do not establish independent success rates or suitability for sensitive transactions. Start with a read-only task and add write actions only after you can validate every intermediate state.
Designing reliable browser tasks
Make the goal testable
Specify the URL or site, the allowed actions, the data to return, and the condition that means “done.” “Check the order” is ambiguous; “open the order page, return the displayed status and order date, and do not change anything” is testable.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsBound time and side effects
Use a task timeout appropriate to the site, cap retries, and make destructive actions explicit. If an element is missing, prefer a structured “not found” result to an invented completion.
Verify outputs
Check URLs, counts, downloaded files, and status text in your own code. For financial, medical, legal, or account-security workflows, require a second signal or human review before treating the run as authoritative.
Troubleshooting common failures
Package or Python-version error
Cause: an older Python interpreter or stale installation command. Fix: use Python 3.11 or later, create a clean environment, upgrade the package, and follow the current README.
The agent loops or stops early
Cause: an underspecified goal, blocked page, or model that cannot interpret the layout. Fix: add the starting URL, expected evidence, and a stop condition; capture the final page state and reduce the task to one navigation objective.
Recommended Free Tools
Cloud login is missing
Cause: profile sync includes cookies but not local storage, IndexedDB, or extensions. Fix: use the supported cloud authentication flow, test a fresh profile, and handle a renewed sign-in explicitly.
CAPTCHA appears
Cause: the site’s risk system challenged the session. Fix: do not assume stealth is a guarantee; stop for an approved human step or use the site’s official API.
Costs exceed expectations
Cause: long browser sessions, repeated retries, traffic, or hosted-agent markup. Fix: measure each run, shorten prompts and navigation, terminate idle sessions, and budget model, browser, traffic, and hosted-agent charges separately.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Or skip the browser setup
If your goal is a clean visual capture rather than clicking through a workflow, ScreenshotNeo is a simpler API route. It accepts consent banners before capture and removes more than 60 known consent platforms, newsletter popups, and chat widgets. Bot checks, blank pages, timeouts, failed loads, and cache hits are not billed, and response headers identify the page verdict and billing status. It also offers an MCP server with take_screenshot, get_page_info, and capture_pdf for Claude, Cursor, and other MCP clients.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →One request returns a PNG, JPEG, WebP, or PDF. See the complete options and parameter reference in the ScreenshotNeo documentation.
cURL
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webpPython
import requests
r = requests.get("https://api.screenshotneo.com/v1/shot", params={"access_key": "YOUR_API_KEY", "url": "https://stripe.com"}, timeout=90)
open("shot.webp", "wb").write(r.content)Node.js
const q = new URLSearchParams({ access_key: 'YOUR_API_KEY', url: 'https://stripe.com' });
const res = await fetch(`https://api.screenshotneo.com/v1/shot?${q}`);Every plan includes the features: free accounts receive 1,000 screenshots a month with no card; paid plans start at $5 for 3,000 shots. Create a free ScreenshotNeo account.
Can I use Browser Use for free?
The Python library itself is free and MIT-licensed, but model inference and hosted browser execution can cost money. The hosted product publishes usage rates, so “free” depends on whether you run a local browser and use a provider with its own free allowance.
Does cloud profile sync copy my whole Chrome profile?
No. The documented sync behavior transfers cookies, not local storage, IndexedDB, or extensions. Plan for an additional sign-in when a site depends on those stores.
Is Browser Use guaranteed to solve CAPTCHAs?
No. The project describes stealth browsers and proxies as ways to reduce challenges, while warning that no configuration guarantees avoidance or solving of every CAPTCHA.
Frequently Asked Questions
What is the difference between a cloud browser and the fully hosted API?
A cloud browser supplies the remote browser process. The fully hosted API supplies that browser plus the Browser Use agent runtime, so your application submits tasks rather than operating the agent loop.
Where should I check for changed installation or pricing details?
Use the rolling official README for package and CLI instructions and the Browser Use product page for current service rates.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
