Browser Use
Booking slots and looking things up on websites eats time. Browser Use lets AI carry out a task in a real web browser, for programmers who can set it up.
How you use it: A programmer installs the Python package, chooses a cloud or local browser, and writes the task in code.
In plain English
The README says you give it a task and the model does the task in a real browser by writing Python. It runs that Python in a persistent session and talks to Chrome through the Chrome DevTools Protocol. The same README says it runs on any language model, locally or on Browser Use Cloud.
There are three paths. A fully hosted cloud agent, a command-line install aimed at coding agents, and a Python library you call from your own code. The library defaults to a Browser Use model and a cloud browser when you set an API key. You can instead pass a local Chrome and your own model.
The Python library is MIT licensed. The README is explicit that model calls and hosted browsers are billed separately, and it states a cloud browser price of $0.06 per browser-hour. New Google, GitHub, or Microsoft signups get $1 of cloud credit.
Practical ways to use it
Example 1
Book something that lives on a website
The README's own illustration is a task worded as booking the earliest driving test: find a slot, pick a date, and finish the booking.
Example 2
Look up a public fact in the browser
The Python sample asks an agent to find the star count of the browser-use repo and return the result.
Example 3
Keep the browser on your computer
The README shows a local Chrome browser plus your own model client, including a local Ollama-compatible server.
Example 4
Hand the clicking to a hosted browser
With a Browser Use API key, the default agent uses a cloud browser that stops when the run ends. The README says new signups receive $1 of credit.
Who it is for
Developer. The library path is Python. The CLI path is a prompt you paste into a coding agent. This is not a point-and-click product for a first-time computer user.
Difficulty: Developer setup. Python 3.11 or newer, an exact version pin, and either an API key or your own model plus a local browser.
Getting started
- 1
Read the version warning first
The README says to pin an exact version, for example browser-use==1.0.6, and never a range. Releases move quickly and can break your code.
- 2
Create a Python project
The README shows Python 3.11 or newer. With uv, create a project on Python 3.12 and add the pinned package. With pip, use a virtual environment and the same pin.
- 3
Choose cloud or local
For cloud, set BROWSER_USE_API_KEY from the cloud site. For local, follow the sample that sets your model and a local Chrome browser.
- 4
Treat the live view as private
Cloud browsers print a live-view link. The README says anyone with that link can control the browser. Do not forward it.
What you need first
- Accounts
- A Browser Use API key for the cloud default. Without a key, the README says an interactive desktop run opens a login. A local model needs whatever key or server that model uses.
- Hardware
- A local Chrome run uses your computer. The cloud path uses their browsers. The README does not list a minimum laptop spec.
- Cost
- The library is free and MIT licensed. The README says cloud browsers are $0.06 per browser-hour, with $1 of credit for new Google, GitHub, or Microsoft signups.
Cautions
- Anyone with the cloud live-view link can control that browser. The README says so directly.
- Pin the exact version. The README warns that an open version range can pull a breaking release.
- A browser agent can click, type, and submit forms. Give it a narrow task and watch the first runs.
Questions people ask
What does Browser Use do?
It lets a language model carry out a task in a real Chrome browser by writing Python. You can run that browser locally or in Browser Use Cloud.
Who is Browser Use for?
Developers, or people using a coding agent who can paste the CLI install prompt from the README.
What do I need first?
Python 3.11 or newer, and either a Browser Use API key or your own model plus a local browser. Pin the version the README names.
What should I watch out for?
Do not share the live-view link. Keep tasks narrow, because the agent can click and type. Cloud browser time is billed at the rate printed in the README.