Glowing abstract browser windows controlled by an AI cursor, navy and orange

Browser Use: Free AI Agent That Clicks Through Sites for You

In short

  • Browser Use is a free, MIT-licensed Python library that lets an AI agent drive a real browser: open pages, click, type, fill in forms.
  • The library costs €0. You pay only for the AI model behind it. A cloud browser, if you want one, starts at $0.02 per hour.
  • Do the math: browser chores handled by a part-time helper cost about €320 a month. An agent doing the same work costs roughly €40.

Every business has browser work nobody wants to do. Copy a price from one site and paste it into another. Download an invoice from a supplier portal. Fill in the same form 50 times. A person costs money for that, and a classic script breaks the moment a button moves.

Browser Use sits in the middle. You describe the task in plain English, and an AI model looks at the page and decides what to click. The library is open source and free. What does it cost to try? About 15 minutes and an API key from any AI provider.

What Browser Use is

Browser Use is an open-source Python library for AI browser agents. It uses the MIT license, so you can build it into commercial work. On GitHub it has around 120,000 stars, which puts it among the most popular agent projects out there. It needs Python 3.11 or newer.

The loop is simple. You give the agent a task. It opens a browser, reads the page, picks an action, runs it, checks the result and repeats until the job is done. You choose the model that does the thinking. There are wrappers for OpenAI, Anthropic and Google models, and local models work through Ollama. The team also offers its own browser-focused model, BU2.

There are three ways in. The Python library, which we use here. A CLI that gives your existing coding agent control of your browser. And a hosted cloud product, if you would rather run nothing yourself.

What it costs

PartPriceNote
Browser Use library€0MIT license, runs on your own machine
Local browser€0Chrome or Chromium on your computer
AI modelPay per useYour provider rates, or free with a local Ollama model
Cloud browser (optional)$0.02 per hour plus trafficAdds stealth mode, proxies, CAPTCHA solving
Cloud signup bonus$1 creditFor new Google, GitHub or Microsoft signups

The only real running cost is the model. Every step the agent takes is a model call. Short, precise tasks stay cheap. Vague tasks burn tokens. The cloud browser is optional, and the agent runs fine in the Chrome on your own computer.

Set it up in 15 minutes

  1. Install uv and run uv init --python 3.12 in a new folder.
  2. Add the library with uv add browser-use.
  3. Put your model API key in a file called .env, for example OPENAI_API_KEY=your-key.
  4. Save the code below as agent.py.
  5. Run it with uv run agent.py.

Here is the code for step 4:

import asyncio

from browser_use import Agent, ChatOpenAI
from dotenv import load_dotenv

load_dotenv()

async def main():
    llm = ChatOpenAI(model='YOUR_MODEL_NAME')
    agent = Agent(
        task='Find the number of stars of the browser-use repo',
        llm=llm,
    )
    history = await agent.run()
    print(history.final_result())

asyncio.run(main())

The agent opens a browser, looks up the repository and prints the answer. Change the task text and you have a new automation. In the llm line, use any model your key has access to. Keep the first tasks small: one site, one clear goal, one result you can check by eye.

Stop paying for the same job twice

Here is the feature that saves real money. Browser Use can save a successful run and replay it later with new values. The replay repeats the clicks and typing without asking the model at every step. Take a job like requesting an invoice copy for a different customer each day. Run it once with the agent, save the history, then rerun it with a new name and email.

The code is short. agent.save_history('invoice_run.json') saves the run, agent.detect_variables() lists the fields you can swap, and load_and_rerun replays it with your values. Two warnings from the docs. If the page changed, the replay stops at the first step it cannot match, and the steps before it already ran, so check whether the form was sent before you retry. And keep the saved file private, because page snapshots can contain sensitive data.

Do the math

Do the math: say a part-time helper does your browser chores for €8 an hour, 10 hours a week. Price checks, form filling, report downloads. That is €80 a week, about €320 a month and €3,840 a year.

Now the agent. Assume 200 tasks a month at an average model cost of €0.20 per task. That number is my assumption, and your real figure depends on the model and how long each task runs. It comes to €40 a month. Add a cloud browser for 40 hours at $0.02 an hour and you pay under €1 more.

The difference is about €280 a month, or roughly €3,300 a year. And the agent works at 3 a.m. without asking for a day off.

Where I would use it

  • A morning price check on competitors across ten listings.
  • Downloading reports and invoices from supplier portals that have no API.
  • Repeating forms: marketplace listings, directory submissions.
  • Testing your own website: send the agent through checkout and see what breaks.
  • Collecting public data for research.

Limits and weak spots

  • Every step is a model call. Long, vague tasks get slow and expensive.
  • Sites with strong bot protection can block it. Cloud browsers with proxies help, but the docs admit that no setup guarantees every CAPTCHA gets solved.
  • Agents make mistakes. Do not point one at your bank account, and do not let it send anything irreversible without a check.
  • If a site terms forbid automated access, respect that.
  • Only the library is free. The model and the hosted browsers are paid.

Alternatives

  • Playwright: free, but you script every click yourself. More reliable on stable sites, more work.
  • Browserbase: hosted browsers with a free plan and paid plans from about $20 a month.
  • Browser Use Cloud: the hosted agent API from the same team, if you do not want to run anything yourself.

Read also: Automate Browser and Desktop Chores for €0 With AutoCua, an Open-Source Agent · Let Your Coding Agent Use Your Real Browser With OpenCode Browse · Set Up Every AI Coding Agent From One File for €0 With Microsoft APM

Ivan’s Take

Worth it. For €0 and fifteen minutes you get a worker that clicks through websites, and that is the cheapest way I know to test whether a browser chore deserves automation. I would start with something boring and harmless, like checking competitor prices for GiantParts listings, and only then move to anything with money attached. Keep the tasks short, save the runs you repeat and watch the model bill for the first week.

Source: Browser Use

II
About the author
Ivan Ivanov
Founder ofAI Rush XGiantPartsSupreme CommerceHawkValueBabyVitaAvtoGear

If it doesn’t save you time or money, you won’t see it here

  • Facts, not hypeEvery price and number is checked against the official source
  • Always the real priceYou see what it costs before you click
  • I say when it’s weakEven when it’s OpenAI or Google
All articles by Ivan →

Similar Posts