Browser Use hands your AI agent the keys to a real web browser

Browser Use is an open-source library that lets an AI agent drive a web browser the way a person does: it opens pages, types into forms and presses buttons until the job is done. The project began as a four-day prototype posted on Hacker News, then went through Y Combinator and raised $17 million in seed funding led by Felicis Ventures. Installing the library costs nothing, while a cloud version bills hosted runs by usage. Developers automating scraping, QA or repetitive data entry are the natural audience.

Under the hood: the web page rewritten for the model

Browser Use scans a page for every element an agent can act on, buttons, fields and links, then feeds that inventory to the language model in a structured format. The model picks the next move, runs it through Playwright, checks the outcome and corrects course when a popup or a layout change gets in the way.

The quickest way to grasp it is a scenario. You type 'compare this headset's price across three shops and log everything in a table', then watch the agent open tabs and fill in its grid. Every session gets recorded (handy when the agent wanders off, since the action log tells you why).

• Price watching across competitor sites
• Data extraction and web scraping
• QA runs that mimic a real visitor
• Form filling in bulk

Library, CLI or cloud: three ways to put Browser Use to work

Three routes lead to the same agent, and the split is clean. The Python library suits repeatable automation written into your codebase. The CLI hands browser control to an agent you already run, with Claude Code, Codex and Cursor at the top of the list. The cloud API hosts everything on remote browsers built for stealth and scale.

The numbers back it up: the framework posted an 89.1% success rate across the 586 tasks of the WebVoyager benchmark, and the repository has piled up tens of thousands of stars, rare air among AI projects hosted on GitHub. Founders Magnus Müller and Gregor Zunic also trained a house model, ChatBrowserUse, described as three to five times faster than general models on this kind of work.

The bill, from pip install to cloud plans

Money talk, then. The library is free: pip install browser-use, add a model API key, done. The cloud side runs without a mandatory subscription, starting with 10 free agent tasks per month, then pay-as-you-go pricing: browser sessions at $0.02 per hour, optional proxies, and credit plans from $29 per month. The API moves fast, so the official pricing page is the one reference worth trusting.

For heavy extraction jobs, especially web scraping of dynamic sites, the credit logic makes it easy to estimate cost per task before launching anything big.

Frequently asked questions