Browser Use
Open source agent that drives a real browser, plus managed browser infrastructure
About Browser Use
Browser Use is a library that gives an LLM control of a real browser, so an agent can click, type, scroll and read pages the way a person does instead of calling an API that may not exist. You describe a task in plain language and the agent works out the steps. That covers the large category of work that is only reachable through a web interface: pulling data out of a portal with no export, completing a booking flow, checking out of a shopping cart, running a repetitive QA path.
Two things are being sold here and they are worth separating. The library is MIT licensed and free, installed with pip install browser-use, and you run it on your own machine against your own model keys. The hosted platform is a paid product that runs the same agent on managed browsers with an API, an SDK, an OpenAPI spec, webhooks and an MCP server, and it handles the parts that are tedious to self-host: concurrency, proxies, session persistence and the fingerprinting work that keeps a headless browser from being blocked. The pricing is published per unit rather than hidden, which is unusual in this category, though it is worth reading closely: agent tokens are billed at 1.2 times the underlying provider rate, so the platform takes a margin on inference unless you bring your own key and pay a smaller orchestration fee instead.
Install the open source library with pip and point it at a model, or sign up for the cloud and call the API. In the library, you give the agent a task string and it launches a browser through Playwright, takes in the page state, decides on an action, executes it and repeats until the task is done or it gives up. On the platform, a POST to the v4 runs endpoint starts the same loop on a managed browser and you receive the completed result rather than a script to maintain. Browser Infrastructure is the lower layer: it hands you a managed browser over SDK, REST or CDP that your own harness can drive, billed per minute of machine time. Sessions support advanced stealth handling, and you can route through a managed proxy, bring your own, or run proxyless and pay for egress.
- •MIT Licensed Library - The core agent is open source and self-hostable with pip install browser-use, with no feature gate between it and the hosted version
- •Managed Browser Infrastructure - Cloud browsers at 0.02 USD per hour, reachable over SDK, REST or Chrome DevTools Protocol, usable with your own agent loop
- •API, SDK, OpenAPI and MCP - Four integration paths including a Model Context Protocol server, so other agents can use it as a tool
- •Bring Your Own Key - Connect an Anthropic, OpenAI or Google key and pay the provider directly plus a 0.2x orchestration fee, instead of the 1.2x passthrough rate
- •Stealth Handling - Fingerprint, canvas, WebGL and audio handling on managed sessions, which is the difference between a working automation and a blocked one
- •Published Unit Rates - Per-hour browser time, per-GB proxy and per-million-token model rates listed on the pricing page rather than quoted on request
Developers automating work that exists only behind a web interface, and teams building agents that need a browser as a tool rather than as the whole product. The free tier is a demo at ten agent tasks a month, so realistically this starts at the 29 USD Dev plan for anyone doing real work, or at zero if you self-host the library and supply your own browsers. It is a good fit for data extraction from portals with no API, for booking and checkout flows, and for QA paths that are too brittle to keep writing as selectors. It is a weaker fit if you already run your own browser fleet, since Browser Infrastructure would be duplicating what you have, and a poor fit if you need a deterministic script: an LLM driving a browser is probabilistic, and for a stable page a plain Playwright script is cheaper and more reliable. Treat the benchmark charts on their site with care, since they are run in Browser Use's own harness against competitors.
Pricing
Usage based, no flat monthly plan
Free tier at 0 USD covers 10 agent tasks a month and 3 concurrent sessions with no card. Paid plans are Dev 29 USD/month, Business 299 USD/month and Scaleup 999 USD/month, each granting that amount as monthly credits. Usage is metered on top: browser sessions at 0.02 USD/hour, managed proxy at 5 USD/GB, and agent tokens at 1.2x provider rates, or bring your own model key for a 0.2x orchestration fee. The library itself is MIT and free to self-host.
From the vendor pricing page, 2026-09-19
