First, it’s not an OS
Cloudflare OS is an open-source AI productivity environment, originally built for Cloudflare’s own staff. Think agent chat, workspaces, and small apps you can build and share with other people. Cloudflare calls those apps “Gadgets”, with “Gatekeepers” controlling what agents and apps can access.
Disclaimer before anyone gets excited: this is not an operating system like Linux or Windows. You are not installing it on your laptop to replace Ubuntu. The OS name is about managing AI workloads and company productivity. Fair enough, but quite a name to live up to.
The idea is that a company can deploy its own version, connect its resources, and give people somewhere to work with AI. It is still early-access software.
The $5 entry ticket
For the hosted deployment on your own Cloudflare account, you need the Workers Paid plan. It starts at $5 a month, which honestly gives you quite a lot beyond just running this thing.
That includes 10 million Worker requests and 30 million CPU milliseconds per month, plus usage allowances for things like KV, D1, and Durable Objects. Useful stuff if you already build on Cloudflare. This is the developer-platform plan, not an upgrade to Cloudflare Pro for your domain.
But $5 is the starting subscription, not a promise that your entire setup will only cost $5. Usage above the included allowances and AI model inference can add to the bill.
I tried it. I still don’t quite get it.
My experience was basically another wrapper for chat. Some spaces, collaborative spaces, models through AI Gateway, and that’s about it for what I got out of it.
Yes, the documentation describes more than that. Building little apps, sharing them, connecting company resources with controlled permissions. I can see why a company might want those things in one place.
But for my own workflow, I couldn’t leverage it into anything close to one-shot orchestration. What I wanted was simple to describe:
Build me a website, deploy it to a Worker, connect my domain, and let me review the result.
Not just generate something inside its own workspace. I want it to work with the infrastructure I already have. Workers, DNS, the actual deployment. From a prompt to a website I can visit.
In the setup I tried, I didn’t have the deployment and DNS access needed to do that. It felt too dumb at this stage for the job I wanted it to handle. Having models available through AI Gateway doesn’t solve that missing access.
One important distinction: there is a Cloudflare Gatekeeper. Its documented capabilities include sign-in, AI Gateway billing, and read-only Workers observability. So saying there is no Cloudflare connection at all would be wrong. But reading Worker logs is a very different thing from deploying a Worker and changing DNS.
So what’s the point?
Maybe I’m expecting the wrong thing from it. Cloudflare seems to be building a customizable AI office suite, while I’m looking for an agent that can actually operate my Cloudflare stack.
The sandboxing and permission controls make sense. I don’t want a chat window casually messing with every domain in my account either. Give it scoped access, show me the proposed changes, and let me approve them.
But without that path from a request to a deployed website, I don’t quite understand what Cloudflare OS can achieve for me that makes it worth another workspace.
For now, that’s my take. Interesting foundation, useful possibilities for company collaboration, but not the prompt-to-website tool I wanted. I’ll look again when it can do more of the work I actually need.