Firecrawl is becoming AI’s default web layer.

Open source is a hard place to fake traction. You either solve a real problem developers keep hitting, or you don’t, and the GitHub graph tells the truth either way.

Firecrawl’s tells a clear story. The project has more than 100,000 GitHub stars, the largest open source repo in its category. More than a million people have signed up for the platform. And the companies relying on it now include Apple, Canva, and Lovable – not as experiments, but as production infrastructure for AI products their users actually depend on.

What started as a developer tool is turning into something larger: the default web layer for AI-native products.

From open-source pull to category position

The pattern matters because it’s the order things happened in. Firecrawl didn’t launch with enterprise contracts and work backward to community. It built in the open, solved a problem engineering teams kept hitting on their own, and let the adoption compound. By the time bigger companies arrived, the trust was already there.

That problem is straightforward to describe and quietly difficult to solve. AI products need current information, and the web is the biggest live source of it – but the web wasn’t built for machines. Pages render dynamically. Content hides behind clicks, scrolls, and overlays. Every team building an AI agent or workflow eventually runs into the same wall, and most of them spend months writing brittle scripts to get past it.

Firecrawl built the layer underneath. The product is organized around three things: search finds the right information on the live web, scrape turns it into clean, structured data, and interact handles the harder cases where a system has to click, navigate, or operate a page to reach what it needs. Together they let an AI system reach the same web a person does – without every team rebuilding the plumbing themselves.

Why this becomes a category

AI agents only work if they can reach the world outside the model, and reaching the world means reaching the live web reliably. That’s now the bottleneck most AI products run into, which is why the web layer is moving from a thing teams build internally to a thing they buy.

The companies that win that shift tend to be the ones developers already trust. Open source momentum here isn’t a marketing channel – it’s the proof that the underlying infrastructure works at scale, across edge cases, with constant community pressure-testing.

Firecrawl is also starting to shape how the layer behaves over time – partnerships including with Wikipedia point at a model where information sources are compensated for the value they provide to AI systems. That’s a signal the company is thinking past extraction to the longer-term economics of an AI-mediated web.

The first wave of AI was about better models. The next is about agents that actually do things, and those agents only work with reliable access to the live web. Firecrawl is becoming the layer that makes that possible.