On March 3, 2025, the Browser Use Python package was downloaded roughly 5,000 times. One week later, on March 10, that number was 28,000 [TechCrunch, March 2025]. The trigger was Manus, the Chinese general-purpose agent that briefly broke AI Twitter, and which uses Browser Use as one of its underlying tools. For a small open-source project from a pair of founders nobody had heard of in January, that is the kind of week most companies wait years for.
Browser Use, founded in 2024 by Magnus Müller and Gregor Zunic and based in San Francisco, builds the unglamorous plumbing that lets large language models actually do things on websites: fill out forms, run searches, pull research, click buttons, navigate flows that were designed for humans with eyeballs and mice [CO/AI]. The technical wedge is straightforward to describe and harder to execute. Instead of asking a model to interpret pixels, Browser Use converts each web page into structured text that an LLM can process deterministically [SiliconANGLE, March 2025]. The page becomes something closer to a spreadsheet than a screenshot. That choice trades visual fidelity for reliability, and reliability is what agents have been short on.
The bet is that the web, as it exists today, is hostile terrain for software that is supposed to behave like a person. Every modern site is a thicket of JavaScript, pop-ups, dynamic IDs, and anti-bot defenses. Anthropic's Computer Use, OpenAI's Operator, and a handful of startups including Skyvern, MultiOn, and Browserbase are all attacking the same problem from different angles, some with vision-first models, some with managed browser infrastructure. Browser Use sits at the open-source layer underneath much of this activity. The company shipped a free Python library, watched developers wire it into their own agents, and is now building a hosted cloud product on top with stealth browsers, CAPTCHA solving, and residential proxies [Browser Use].
Why the upside is real
The distribution numbers are unusual for infrastructure this young. Browser Use crossed 50,000 GitHub stars within three months of launch [Y Combinator] and is now reported at 78,312 [openalternative.co]. GitHub stars are a vanity metric in isolation, but combined with a 5.6x jump in daily downloads and adoption inside a viral product like Manus, they suggest the project found genuine product-market fit with the constituency that matters most for picks-and-shovels infrastructure: the developers building the next layer up.
The investor list reflects that read. The $17 million seed announced in March 2025 includes Felicis, 468 Capital, A.Capital, Nexus Venture Partners, SV Angel, Liquid2, Pioneer Fund, and Paul Graham personally [SiliconANGLE, March 2025]. Y Combinator is in as accelerator. Felicis, in its investment note, framed the thesis as enabling reliable web interaction for agents [Felicis], which is the polite way of saying the current state of the art is not reliable enough for production. If even a fraction of the agent demos shipping today need to graduate to actual paid workflows, somebody has to make the browser layer not break.
Daily downloads, March 3 2025 | 5000 | downloads
Daily downloads, March 10 2025 | 28000 | downloads
GitHub stars at 3 months | 50000 | stars
GitHub stars, latest | 78312 | stars
Müller and Zunic are running a team of seven [Y Combinator], which is small for the scope of what they have taken on but appropriate for an open-source project where the community is doing a meaningful share of the integration work. The seed round gives them roughly two to three years of runway at conservative burn, which is the window they need to figure out whether the commercial product, the hosted cloud, can capture enough of the value the open-source library is generating elsewhere.
What bears say, and what bulls answer
The sharpest concern is the one that haunts every open-source infrastructure company: monetization. Browser Use is free, the code is on GitHub, and a competent engineering team can self-host it indefinitely. Anthropic's Computer Use ships inside Claude at no marginal cost to existing customers, and Browserbase has been building a managed browser business with its own fundraising momentum. The bear case is that Browser Use ends up as the Linux of agent browsing: enormously important, widely deployed, and commercially captured by someone else. The bull answer, visible in the cloud product roadmap, is that running browsers reliably at scale (with stealth, CAPTCHA handling, proxies, and observability) is hard enough that most teams will pay rather than build [Browser Use]. The same pattern produced Vercel from Next.js and HashiCorp from Terraform. It does not always work, but when it does the outcomes are large.
Back of the envelope
A rough sketch of the economics: if Browser Use converts even 1% of its current download base into paying cloud customers at $200 per month average contract value, that is 280 customers (using the March 10 download number as a floor) generating roughly $670,000 in ARR. Push conversion to 3% and ACV to $500 (plausible for teams running production agent workloads) and the same funnel produces roughly $5 million ARR. Neither number justifies the seed valuation on its own, but both are achievable inside 18 months from a base that is still growing weekly. The interesting question is not whether the cloud product can reach single-digit-millions ARR. It is whether agent traffic on the open web grows fast enough over the next three years that Browser Use becomes the default substrate, in which case the relevant comparison is not a SaaS tool but a usage-metered infrastructure layer with Stripe-shaped economics.
What to watch
The next twelve months will turn on three things: whether the hosted cloud converts a meaningful slice of the open-source base into paying customers, whether a major agent platform standardizes on Browser Use as its default browser layer, and whether Anthropic or OpenAI ship something that makes the open-source library redundant. A Series A in late 2025 or early 2026 is the likely tell. If Felicis or a new lead writes a $40-60 million check at a valuation step-up, the cloud metrics worked. If the round is quieter, the monetization question is still open.
The company Browser Use most has to beat is Browserbase, the well-funded managed-browser startup that is building the same hosted infrastructure from the proprietary side rather than the open-source side. One of them will own how AI agents touch the web. The other will be a footnote in someone else's S-1.