Cloudflare has released the code behind an internal tool it calls Cloudflare OS, making it free for any business to download and run on its own systems. The company built it so its own staff, including people with no coding background, could describe a task in plain English and get a working app, document, or automated workflow back. Cloudflare said thousands of its employees already use the platform across engineering, sales, and other business functions, turning a described task into a working application.
The person behind the project has been building toward this for a decade. The tool is a remake of Sandstorm.io, a startup Cloudflare principal engineer Kenton Varda ran ten years ago, rebuilt this time on Cloudflare's own cloud computing platform and combined with AI. The pitch is that Cloudflare finally has both pieces in place, its own servers and modern AI, to make that old idea work at real scale.
The real selling point is not the app builder itself. Plenty of tools already let non coders describe software and get working code back. It is the safety claim wrapped around it: every app an employee builds gets locked inside its own sealed compartment, with its own private database and no outbound internet access unless specifically switched on. Varda put it plainly on social media, saying "the sandbox is so secure that you can pretty much go wild, the AI cannot introduce a significant security bug."
That claim matters because the problem it targets is real. Close to half of workers already use AI tools their employer has not approved, and most of those workers rely on free versions with no built-in data protection. Deloitte's 2026 research found that access to AI at work rose sharply, yet only one in five companies has a mature system for overseeing how it gets used. Independent testing backs this up: Veracode's 2025 review of AI-written code found close to half of it carried security weaknesses tied to well known vulnerability categories.
Giving employees a sanctioned way to build things, instead of counting on people not doing it anyway, is a sensible bet. But the safety claim is Cloudflare's own, and outside commentators have flagged a tension inside it. A sandbox can only guarantee safety if the app inside truly cannot touch anything outside its walls, and that same restriction limits how useful the app can be. The tighter the box, the less an app can do with real business systems, so somewhere a line gets drawn between safe and useful, and nobody outside Cloudflare has tested exactly where that line sits.
There is a business angle worth watching too. Cloudflare is mid pivot toward selling itself as the infrastructure for AI agents rather than just network security, having grown revenue by 34 percent in the first quarter of 2026 while cutting a fifth of its workforce toward what it calls an AI-first way of operating. Every app built on Cloudflare OS runs on Cloudflare's own paid computing platform, so a free tool that draws more companies onto it also feeds Cloudflare's core business.
For a company without a large software budget, this free and open route is notably cheaper than the alternatives on the market. Comparable enterprise AI agent tools carry real costs, with Glean's enterprise contracts starting above fifty thousand dollars a year and Microsoft's Copilot Studio priced at roughly two hundred dollars per user. If the sandbox claims hold up under outside scrutiny, a free version of that setup changes the math for smaller operators who never had budget for tools like this before.