Browser Use

Market position

Positioned as a leading open-source web-agent project and an infrastructure layer for AI browser automation. Y Combinator describes it as the biggest open-source web agent project; press coverage frames it as an open-source alternative to OpenAI's Operator, priced below it. Its adoption by Butterfly Effect's Manus tool and by more than 20 companies in the same Y Combinator winter batch is cited as evidence of developer traction, and the company says teams at large technology firms use it. Competitive comparisons on its own site are against cloud-browser providers (Notte, Anchor, Kernel, Hyperbrowser, Steel, Browserbase) and against frontier-model computer-use agents from OpenAI, Anthropic, Google and Microsoft.

Differentiators claimed across the sources are: a text/DOM-structured representation of web pages rather than screenshot-based vision, which the founders argue is faster, cheaper and more robust to site changes; model neutrality, since the open-source agent works with any LLM (Gemini, Sonnet, Qwen, DeepSeek-R1 and others) and the cloud API routes to multiple providers with one key; an MIT-licensed open-source core with a large contributor and Discord community that has spawned downstream projects; and self-reported benchmark leadership — first place on the Odysseys leaderboard at 87.4% over 200 long-horizon tasks, and internal benchmarks claiming both higher strict accuracy and lower cost per solved task than frontier models, plus the lowest price per browser hour among named cloud-browser vendors. These benchmark claims originate with the company and are not independently verified in the sources.

Technology

The system converts web page elements — buttons, forms, menus — into a structured, text-like representation that language models can parse, rather than relying on screenshot-based vision, which the founders describe as slow, costly and unreliable. A secondary source describes the platform as combining DOM analysis with visual recognition. The open-source stack is a Python library (Python 3.11+) exposing an Agent class, a Tools decorator for custom actions, system-prompt extension/override, Docker images and an MCP integration. The cloud side offers a v4 REST API for agent runs and a browsers API returning CDP URLs for Playwright/Puppeteer control, with proxy country selection, session persistence, filesystem and memory. Browser Use also provides its own models optimized for browser automation (ChatBrowserUse, bu-2-0-mini-preview, an open-source browser-use/bu-30b-a3b-preview), which it says complete tasks 3-5x faster than general models, while also proxying third-party models such as GPT-5.5, Claude, Gemini and Grok through one API key.

Go-to-market

Developer-led, open-source-first distribution: the MIT-licensed GitHub repository and PyPI package attract developers who can later convert to the paid cloud. Onboarding is self-serve via a free tier, Python and TypeScript SDKs, a REST API and documentation; a CLI "skill" installs Browser Use into third-party coding agents. The launch post invited companies to contact the team directly for API access and custom integrations, indicating a direct sales motion alongside self-serve. Community channels (Discord, Hacker News, Y Combinator) and press coverage have been significant awareness drivers, as has embedding in third-party products such as Manus.

Developers and engineering teams building web automation or embedding browser agents into their own products, plus users of coding agents who want browser tasks executed. Use cases span scraping and monitoring at scale, QA testing, CRM integrations, form filling and authenticated workflows. Reported adopters include Butterfly Effect's Manus and, per a secondary aggregator, engineering teams at large technology companies including Airbnb, Amazon and Anthropic.

Geography

Headquartered in San Francisco, California, with all listed job openings based there. The company originated at ETH Zurich in Switzerland, and its cloud browsers support country-specific proxies (e.g. a US proxy country code).

History

The founders met in 2024 while both were pursuing master's degrees in data science at ETH Zurich, and started the company that year through ETH Zurich's Student Project House accelerator. An initial prototype was built in four days and posted to Hacker News; a fuller demo followed within about five weeks, after which the project was open sourced. The repository grew quickly — roughly 25,000 GitHub stars within three months of the start of the project, cited as 40,000–50,000 stars in Y Combinator materials — and the team joined Y Combinator's Winter 2025 batch. Days after OpenAI launched Operator in early 2025, the team shipped Browser Use Cloud, a hosted version, in response to community demand. In March 2025 the company disclosed a $17 million seed round led by Felicis, and its adoption inside Butterfly Effect's viral Manus product raised its profile further. By the date of the sources the repository stands at about 110,000 stars and the team at seven people.

Risks & controversies

The sources note that websites frequently change and actively detect bots, which breaks automation and requires proxy rotation, CAPTCHA solving and human-like behavior — capabilities Browser Use markets as "stealth" and "cloud browsers that bypass anything," an approach that carries operational and potential legal or platform-policy exposure not discussed in the sources. LinkedIn is cited as an example of a site that changes constantly and causes agents to fail. Performance and pricing comparisons against competitors and frontier models are self-published and run in the company's own harness. The company also competes directly with much larger model providers, including OpenAI's Operator, and depends on third-party LLM providers for parts of its stack.