Cloudflare OS – an open-source AI productivity environment

Cloudflare has open-sourced Cloudflare OS, an AI productivity environment first built for internal use at the company. A large portion of Cloudflare’s workforce—from engineering to sales and roles in between—already uses it every day. It is not a traditional computer operating system. The name is used in two deliberate senses: a company-level system for getting work done with AI under security constraints the security team can accept, and a runtime layer for AI workloads in the same way a conventional OS schedules and isolates compute.

Technically, Cloudflare OS centers on three product surfaces. First is an agent chat UI where people assign tasks to agents that are preloaded with knowledge of how the company operates, so prompts are grounded in real process and policy rather than a generic assistant. Second is sandboxed application development: users can ask agents to build gadgets—small personal apps—and share those gadgets with colleagues without dropping the isolation boundary. Third is an explicit security model that treats AI productivity as something that must be governed, not bolted on after the fact; the “operating system for AI workloads” framing is about managing those agents and apps the way an OS manages processes and resources.

For engineers and builders, the interesting piece is the closed loop from chat to code to share. Instead of a free-floating chatbot and a separate prototype repo, the same environment is meant to hold company context, generate small tools, and keep execution sandboxed so sharing is possible without expanding trust by default. That pattern maps cleanly onto internal tooling: ticket helpers, ops gadgets, sales workflows, and one-off utilities that usually die as private scripts because review and distribution are expensive. Sandboxing and preloaded company knowledge are the mechanisms that make daily use across non-engineering roles plausible.

Market context is a company dogfooding a full AI productivity stack and then releasing it as open source, rather than only publishing model APIs or a thin chat wrapper. Cloudflare’s pitch is organizational: safe AI productivity at workforce scale, not a new consumer desktop. Competing efforts often stop at the agent interface or at enterprise policy layers that never ship the build-and-share path. Cloudflare OS puts chat, gadget development, and security in one package and backs that with internal adoption from engineering through sales.

What to watch next is how outsiders adopt the same three pillars without Cloudflare’s internal knowledge base. Agent quality will depend on how well company knowledge can be loaded and kept current; gadget usefulness will depend on how strict the sandbox is relative to real app needs; and security teams will judge whether the isolation model is something they can operationalize, not just demo. The open-source release turns an internal platform into a reference architecture: if the chat-plus-sandbox-plus-policy split holds up outside Cloudflare, it is a concrete blueprint for enterprise AI productivity rather than another general-purpose assistant.