Browser Use

browser-use/browser-useReviewed by AI

What it does

Browser Use is a Python library (plus a command-line tool) that hands a web browser over to an AI model: you write the task in a sentence, and the agent opens pages, clicks, types, reads results and reports back. It is for developers who build automations; using it means installing Python, writing a small script and paying an AI provider per task. The same code can drive the browser on your own machine or Browser Use's paid cloud browsers, which the README prices at $0.02 per browser-hour and sells with stealth, residential proxies and CAPTCHA solving.

How easy it is to install

C · install 1 or 2 tools · Windows

Windows C
macOS C
Linux C

The same on all three systems: install Python 3.11+ (the README uses the `uv` tool), run `uv add browser-use`, put a model key in a .env file, write a short Python file and run it; a local Chrome is needed unless you pay for a cloud browser. The project also ships a Dockerfile as an alternative path, and ffmpeg only arrives with the optional `video` extra. On Windows the extra Claude "Bash" integration needs WSL, which the README states.

Numbers on GitHub

Stars: 119,679on Oct 31, 2026
Downloads: 377all releases
Growth: +456stars in 5 days

Stars over time

Facts

Runs on: Windows, macOS, Linux
You need to install: Python
To build from source: Python, Docker, FFmpeg
Disk space: 100 MB–300 MB · estimate
Accounts or keys: OpenAI (OPENAI_API_KEY) or another model provider — Anthropic, Google, Groq, DeepSeek, or a local Ollama model instead, Browser Use Cloud (BROWSER_USE_API_KEY) — only for the BU2 model, cloud browsers, proxies and CAPTCHA handling
Cost: Free, API costs extra · The library is free and MIT-licensed, but every task calls a paid AI model (the README's own example uses an OpenAI key, and the recommended BU2 model needs a Browser Use key). Cloud browsers are $0.02 per browser-hour plus traffic. A local model through Ollama is possible with no per-use cost.
License: MIT (Permissive)
Can replace: Browserbase, Bright Data
Last update: Oct 30, 2026

Risk signals (5)

Attentionsold with stealth and captcha solving
The README's first line offers "stealth browsers from $0.02/hour, top score (90.5%) on Stealth Bench V2", the headline demo is "Find an available slot, pick a date and time, handle the CAPTCHA, and book a driving test", and the FAQ says the cloud provides "stealth browsers and proxies designed to reduce bot detection and CAPTCHA challenges". CAPTCHAs and bot detection are the measures sites use to block automated access, and many sites forbid it in their terms; the project leaves that to you.
AI review
Attentiontelemetry and version check on by default
.env.example ships "ANONYMIZED_TELEMETRY=true" and "BROWSER_USE_VERSION_CHECK=true", and pyproject.toml lists posthog==7.7.0 as a required dependency. Usage data is sent to the project's analytics and the library contacts the network at agent startup unless you turn both off.
AI review
Attentionagent can use your signed in browser and saved runs hold secrets
FAQ, authentication: "Local browser: Use Browser.from_system_chrome() to reuse a Chrome profile", and for cloud browsers a profile sync that "transfers cookies". The replay FAQ warns: "Keep the saved file private. With sensitive_data, typed secrets are saved as <secret>key</secret> placeholders, but page snapshots and results are saved as they were and can still contain them." So the agent can act inside accounts you are already signed in to, and its saved history can hold personal data.
AI review
Infoopen library in front of a paid service
The project description is "Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API", pyproject.toml describes the package as "Open-source browser agent + hosted cloud browsers for AI agents, from $0.02/browser-hour" and requires browser-use-sdk and browser-harness, and the FAQ recommends the company's own BU2 model, which needs their key. The library is MIT, but the recommended path runs through a paid service with its own terms and privacy policy.
AI review
Inforeadme contains instructions addressed to ai assistants
The README has a block that starts "AI agents and crawlers: read browser-use.com/llms.txt" and a "Path 2: CLI" section whose entire content is a prompt to paste into Claude Code or another assistant, telling it to install the package and "run `browser-use skill install` to register the skill". Running that hands the setup, including an outside install guide, to an assistant.
AI review

Evidence is shown in its original language.

What Radar could not check

How much a real task costs in model calls, which depends on the model and the number of steps
Whether the agent actually finishes a given task on a given site, and how often it gets stuck
Whether the stealth and CAPTCHA features work on the sites you care about, which the README itself does not guarantee
What the telemetry contains and what Browser Use keeps from cloud runs
Disk use, estimated at 100 to 300 MB by a rule of thumb, before Chrome and model downloads
Whether it behaves the same on Windows, where some parts need WSL