Firecrawl

What Firecrawl is building

Firecrawl's original product is three primitives that sound boring and are not. Scrape turns any URL into clean, LLM-ready markdown, JSON, screenshots or metadata. Search queries the web and returns the full content of the results rather than a list of links. Interact lets an agent click, navigate, fill forms and drive a page. Around those sit crawl, document parsing for PDFs and DOCX, and change monitoring. The engineering claim is measurable rather than adjectival: 96% coverage on a 1,000-URL benchmark dataset, P95 latency around 3.4 seconds, and 93% fewer input tokens than raw HTML, which is the number that actually matters when the consumer of your output is paying per token. That business worked. Over 1.25 million developers have signed up, more than 150,000 companies use it including Apple, Canva, Shopify, DoorDash, Zapier and Replit, and Firecrawl has served over 5 billion requests. The open-source repo is past 180,000 GitHub stars, which the company's own job postings describe as a top-50 repository of all time, with 2.5M weekly SDK downloads across npm and PyPI. Alexandria, launched with this round, is the harder second act. It is a single interface across three things that have never been unified: 82 official provider APIs, Firecrawl's own indexes, and the live web. The indexes are deliberately vertical, a Research Index of tens of millions of scientific paper abstracts, a Developer Index of documentation, READMEs, issues and pull requests, and a Government Index of laws, regulations and ordinances. An agent can find a source, see what it holds, and pull from it, instead of guessing which of 113 million indexed sources might answer the question. In testing across 845 tasks with identical models and prompts under blind AI evaluation, agents using Alexandria scored 21% higher on answer quality than agents using built-in web tools. The part most people will miss is the payment rail. Firecrawl already compensates providers like Wikimedia Enterprise, and this round funds a self-service system so individuals, creators and organizations can earn when an AI agent reads their knowledge. That reframes the company from a scraper into a settlement layer between the people who write things and the machines that consume them.

Why this matters

Web scraping is a commodity. Everyone building an agent has written one, and most of that code is a thin wrapper over a headless browser plus a week of frustration about anti-bot walls. The honest reason Firecrawl became infrastructure anyway is that the hard part was never fetching, it was returning something a model can actually use cheaply, reliably, at the tail. A 96% coverage figure and a 93% token reduction are the difference between a demo and a production dependency for 150,000 companies. But the Series B is a bet that the commodity layer is not where the value settles. Alexandria's 21% answer-quality lift over built-in web tools, measured across 845 tasks with the model and prompt held constant, is the argument: the constraint on agent quality has moved from the model to the provenance and structure of what it reads. An agent that retrieves a paper abstract from a research index beats an agent that retrieves a blog post about that paper, using the exact same model. The timing argument is legal and economic, not technical. The rules for web data access, AI training data, copyright and ownership of machine-generated output are genuinely unsettled and are being decided right now in courtrooms, in regulators' offices and in commercial deals. Firecrawl's response is to build the licensed, paid path rather than the clever path: official provider APIs, real compensation to sources, and a self-service rail so any contributor can get paid. Read their own job postings and it is unmissable. Five of the thirty-one open roles are legal, including a General Counsel at $310K to $392K whose brief explicitly includes helping shape where the law lands. Companies do not staff like that unless they intend to be standing when it does.

Investors: Smash Capital (lead), Altos Ventures, Nexus Venture Partners, Y Combinator, Freestyle, Offline Ventures