Browser Use
Browser UseAgent framework that makes websites usable by AI agents, covering autonomous browsing, form filling, data extraction and multi-step online tasks. Built on Playwright, it gives a model a structured way to observe page state, plan actions and execute browser interactions, and ships both a CLI for one-off tasks and a Python library for repeatable automation with custom tools and structured output.
Stealth, proxy rotation and CAPTCHA handling belong to the separate Browser Use Cloud product rather than the library, which is why the openness axis reads the core as ungated. Verified 2026-08-13 via GitHub and the pypistats API.
Openness
5 high confidence- license
- MIT(OSI, library core fully open)
- source
- public
- managed-tier
- Browser-Use-Cloud(proxies/stealth/captcha/hosted browsers, proprietary)
- core-gated
- ungated(one MIT LICENSE and no ee/ path. The pricing page prices hosted capacity only and states that cloud usage and the open-source browser-use Python library are separate products with different APIs. The four modules reading BROWSER_USE_API_KEY - browser/cloud sandbox skills and llm/browser_use - are published clients to that hosted product and the local agent needs no key)
The core is ungated: the repo root has one MIT LICENSE and no ee/, enterprise/ or commercial/ path, and neither the repo nor the pricing page shows a feature gate. The pricing page prices hosted capacity only - Dev $29/mo for 25 concurrent sessions and 20 browser profiles, Business $299/mo, Scaleup $999/mo, Enterprise custom, plus pay-as-you-go credits - and states outright that "Cloud usage and the open-source browser-use Python library are separate products with different APIs." The one thing that could have gone the other way is that four library modules read BROWSER_USE_API_KEY: browser_use/browser/cloud/cloud.py, browser_use/sandbox/sandbox.py, browser_use/skills/service.py and browser_use/llm/browser_use/chat.py. Each is a published client to the separate hosted product rather than a withheld feature - SkillService fetches skills from the Browser Use API through the browser_use_sdk package, sandbox posts to the hosted sandbox, and chat.py is the ChatBrowserUse endpoint - and the local agent, browser session and tools run with any LLM and no key at all. Selling a separate hosted product is the langchain/LangSmith shape, which the rubric settles as ungated.
- https://github.com/browser-use/browser-use recorded 2026-06-04
MIT (OSI) license, full public source; Browser Use Cloud as optional managed tier
- https://api.github.com/repos/browser-use/browser-use/contents recorded 2026-08-12
Root tree: a single LICENSE (MIT per the repo metadata) beside browser_use/, docker/, examples/, skills/, static/, tests/, bin/ and scripts/. No ee/, enterprise/ or commercial/ directory and no second license, so nothing in the published tree is carved out under different terms.
- https://browser-use.com/pricing recorded 2026-08-12
Prices hosted cloud capacity only: Dev $29/mo (25 concurrent sessions, 20 browser profiles, premium proxy pool, bring-your-own model key, scheduled jobs), Business $299/mo (200 concurrent sessions), Scaleup $999/mo (500 concurrent sessions), Enterprise on negotiated annual credits and SLAs, plus a pay-as-you-go credit option. States "Cloud usage and the open-source browser-use Python library are separate products with different APIs."
- https://raw.githubusercontent.com/browser-use/browser-use/main/CLOUD.md recorded 2026-08-12
Describes Browser Use Cloud as "the fully hosted product made by Browser Use made for users to automate web-based tasks", billed per usage through an API key, with its own v2 REST API at api.browser-use.com, hosted Sessions and Browsers, cloud Browser Profiles and a chromium fork carrying "proprietary optimizations". All of it is hosted infrastructure sold beside the library; none of it is a feature removed from the MIT library, which drives a local browser with any LLM.
Adoption
5 high confidenceBanded on downloads of the declared browser-use package, one of the most-used agent browser-automation libraries; GitHub stars corroborate but are not the basis. The registry reports 53,142,086 downloads in the last month, more than five times the >10M floor for level 5, and a live measured figure like that takes precedence over a hand-set band.
- https://pypistats.org/api/packages/browser-use/recent recorded 2026-08-14
last_month = 53,142,086 downloads for browser-use (last_week 14,575,579, last_day 2,624,999).
- https://pypistats.org/packages/browser-use recorded 2026-08-13
The rendered pypistats page for browser-use; figures read from its API endpoint above.
- https://github.com/browser-use/browser-use recorded 2026-08-13
Repo page, 109,104 stargazers (corroborating).
Capability
4 medium confidenceLeading OSS browser-automation-for-agents library with broad action coverage and a production cloud tier; widely adopted as the default agent browsing layer. Rated 4 on feature_matrix (multi-step web tasks, structured extraction, anti-block infra); no standard benchmark applies to a browser-control library.
- https://github.com/browser-use/browser-use recorded 2026-08-13
README - CLI for one-off tasks, Python library for repeatable automation; "Custom tools, custom system prompts, structured output, fine-grained browser control"; skills/ in the root tree. Stealth, proxy rotation and scaling are attributed to the hosted cloud browsers - "We recommend pairing it with our cloud browsers for leading stealth, proxy rotation, and scaling".
Verified 2026-08-12