Browser Use: a four-day Hacker News prototype to a $17M seed
Two ETH Zurich students built an open-source library that lets AI agents use websites, launched it on Hacker News, and raised a $17 million seed led by Felicis about five months after the repo went up.
- Product
- Browser Use
- Founder
- Magnus Müller and Gregor Žunič
- Launched
- 2024
- Model
- Open-source library plus a paid hosted cloud
Published · 2 min read · 3 sources
Key takeaways
- Browser Use says it built its first prototype in four days and launched it on Hacker News; the GitHub repo was created on 31 October 2024.
- In March 2025 it announced a $17 million seed led by Felicis, saying it had 50,000 GitHub stars and more than 15,000 developers using its tools.
- TechCrunch credited Manus, the viral Chinese agent, with driving awareness; when checked on 12 November 2026 the repo showed about 120,000 stars.
What it is
Browser Use is an open-source Python library for AI agents that need to operate a web browser, plus a paid hosted service, Browser Use Cloud (GitHub). Instead of having a model read screenshots, it converts web pages into structured text that a language model can act on (Browser Use blog).
Timeline
- 2024: Magnus Müller and Gregor Žunič meet while doing master's degrees and found Browser Use through ETH Zurich's Student Project House (TechCrunch).
- 31 October 2024: The browser-use repository is created on GitHub (GitHub). The team writes that it built "an initial prototype" in four days and launched it on Hacker News, and that "the response was immediate and overwhelming" (Browser Use blog).
- Winter 2025: Browser Use joins Y Combinator's W25 batch; Müller says more than 20 companies in that batch used it (TechCrunch).
- March 2025: The viral Manus agent from Chinese startup Butterfly Effect uses Browser Use, which TechCrunch says "drove awareness to new heights" (TechCrunch).
- 22 March 2025: Browser Use announces a $17 million seed led by Felicis, with A Capital, Nexus Ventures, Y Combinator, Paul Graham, SV Angel and others. It reports 50,000 GitHub stars and more than 15,000 developers using and contributing to its tools (Browser Use blog).
- 12 November 2026: The repository shows about 120,000 stars when checked (GitHub).
Why it worked
- Ship first, then talk. "If you have an idea, just build it and launch it," the founders write; the first version took days, not months (Browser Use blog).
- A clear technical bet. Text-based page understanding is pitched as faster, cheaper and more reliable than pixel-based navigation (Browser Use blog). Müller: "A lot of agents rely on vision-based systems and try and navigate websites through screenshots, and in [the] process, things break." (TechCrunch)
- Open source as distribution. Other builders, from YC startups to Manus, could adopt it for free, and Manus's popularity sent attention back to the library (TechCrunch).
What to learn
- Infrastructure for a hot trend can grow faster than the apps on top of it. One viral agent using your library is free marketing (TechCrunch).
- Hacker News is still a good first test for developer tools: a rough prototype was enough to get a strong signal (Browser Use blog).
- Plan the paid layer early. Browser Use now pairs the free library with a hosted cloud product (GitHub). Compare Firecrawl and Manus.
Risks and what didn't work
- Stars are not revenue. No revenue figure appears in the funding announcement; usage is described in stars and developer counts (Browser Use blog).
- The origin story varies. The company's post says the prototype took four days; TechCrunch reports a demo built "in five weeks" (Browser Use blog; TechCrunch).
- Self-reported metrics. The 50,000 stars and 15,000 developers figures are from the company's announcement (Browser Use blog).
FAQ
Short answers drawn from the sections above.
What is Browser Use?
Browser Use is an open-source Python library for AI agents that need to operate a web browser, plus a paid hosted service, Browser Use Cloud (GitHub). Instead of having a model read screenshots, it converts web pages into structured text that a language model can act on (Browser Use blog).
Why did Browser Use work?
- Ship first, then talk. "If you have an idea, just build it and launch it," the founders write; the first version took days, not months (Browser Use blog).
- A clear technical bet. Text-based page understanding is pitched as faster, cheaper and more reliable than pixel-based navigation (Browser Use blog). Müller: "A lot of agents rely on vision-based systems and try and navigate websites through screenshots, and in [the] process, things break." (TechCrunch)
- Open source as distribution. Other builders, from YC startups to Manus, could adopt it for free, and Manus's popularity sent attention back to the library (TechCrunch).
What can indie builders learn from Browser Use?
- Infrastructure for a hot trend can grow faster than the apps on top of it. One viral agent using your library is free marketing (TechCrunch).
- Hacker News is still a good first test for developer tools: a rough prototype was enough to get a strong signal (Browser Use blog).
- Plan the paid layer early. Browser Use now pairs the free library with a hosted cloud product (GitHub). Compare Firecrawl and Manus.
What didn't work for Browser Use, and what are the risks?
- Stars are not revenue. No revenue figure appears in the funding announcement; usage is described in stars and developer counts (Browser Use blog).
- The origin story varies. The company's post says the prototype took four days; TechCrunch reports a demo built "in five weeks" (Browser Use blog; TechCrunch).
- Self-reported metrics. The 50,000 stars and 15,000 developers figures are from the company's announcement (Browser Use blog).
Sources
Every link cited above, in order of first use.
- 1.GitHub (github.com)
- 2.Browser Use blog (browser-use.com)
- 3.TechCrunch (techcrunch.com)
Related case studies
More in AI tools →Neon: serverless Postgres sold to Databricks for about $1B as AI agents took over
Neon built a serverless Postgres that spins up a database in under a second. Four years after its founding, Databricks agreed to buy it for about $1 billion, citing telemetry that more than 80% of new Neon databases were created by AI agents.
Comp AI: an open-source Vanta rival with $6.7M verified revenue
Three founders from a shut-down AI startup built open-source SOC 2 software in 2025. TrustMRR shows $6.7M all-time revenue; a $34M Series A came in 2026.
Bun: 7 million monthly downloads, $0 revenue, then acquired by Anthropic
Jarred Sumner started Bun as a side quest to speed up his own dev loop. It raised $26 million, grew to more than 7 million monthly downloads with no revenue, and joined Anthropic in December 2025 as infrastructure for Claude Code.