Cloudflare has released Cloudflare OS as open-source software, offering organizations an AI workspace connected to their own systems. The company’s August 5 press release, datelined August 4, says the code is available through its public repository, while a managed deployment option is still planned.
The workspace runs in an organization’s Cloudflare account. Employees use a browser interface for research, documents and automated work, with shared company context intended to reduce the need to explain internal processes again in each session.
Internal systems are connected through controlled access
Cloudflare describes governed connectors called Gatekeepers. The owner of an internal system can determine what an agent may read or change and when a human must approve an action.
The platform uses Cloudflare Access to check users and requests. According to the announcement, agents begin without permissions and receive access for specific tasks. The actual usefulness of that arrangement depends on how each organization connects its tools and defines those permissions.
This is a practical issue for any agent that acts beyond a chat window. An agent may be able to describe a requested change before it has authority to make that change. A connector provides the point where the application can enforce the difference.
Cloudflare also says outputs can become shared applications with their own databases and access controls. That expands the workspace from answering questions into creating tools other employees can use.
Model choice and spending are administrative settings
Through Cloudflare AI Gateway, organizations can use different model providers and view spending by person, team or application. Administrators can apply budgets and rate limits or direct suitable tasks to smaller models.
Those controls make cost visible, but the release does not establish that every organization will spend less after adopting the workspace. The bill depends on the chosen models and the work employees run.
The open-source release is the available starting point. A managed dashboard deployment is described as coming later, and the announcement also identifies partners for tailored implementations. Companies assessing the product should distinguish installing the code from obtaining a fully configured managed service.
