Cloudflare open-sources Cloudflare OS, a platform for agents, apps and work

Cloudflare has open-sourced a new version of Cloudflare OS, an AI workspace platform that until now existed only as an internal deployment. The company says any organization can deploy it, connect it to internal systems and adapt it; the code is published on GitHub.

What Cloudflare OS includes

Cloudflare OS combines three parts: an agent workspace grounded in the context and skills a company curates, with an isolated runtime where agents can write and run code; a security and governance framework for access to internal data and services; and a platform for personal, modifiable apps that people can build, share and keep changing.

Workspaces are used in the browser, so staff outside engineering can work with them too. They combine agent sessions, persistent state, outputs and files, resource access and the isolated runtime. Cloudflare says the platform can produce live documents, slides and spreadsheets connected to data sources, build collaborative apps, and turn repetitive jobs into mostly deterministic workflows.

Security: agents start with no access

The platform was rebuilt based on lessons from the first version, which was given to every employee in May. Thousands of people across every function, many of them outside engineering, now use it daily to create documents and slides, automate repeatable tasks and build small apps.

Access is controlled by Cloudflare Access, and every agent and app starts with access to nothing. An agent asks for access to a specific resource, which a user can grant or deny; generated code receives it as a typed binding while credentials stay isolated. Server code runs in a Dynamic Worker with global outbound networking disabled, and client code runs in a sandboxed frame in the browser.

A new component, Gatekeepers, sits between Cloudflare OS and external services, handling OAuth, holding credentials, enforcing policies, recording what was read and mediating actions with externally visible side effects.

Models, cost and customization

Every inference call runs through Cloudflare AI Gateway, so administrators can choose which models are available, route simple jobs to cheaper models and attribute spend to a person, team or workspace. Cloudflare notes that most tasks do not need the most expensive frontier model.

The code is split into two repositories: the core and an example deployment based on Cloudflare's internal setup, which consumes the core without patching it. Cloudflare says the partners Presidio and Happy Cog will help organizations customize the platform. Next steps include bringing Cloudflare OS to the Cloudflare dashboard as a fully managed product, adding containers for development workflows, and bringing workspaces into Slack and other chat tools.