The platform, released under the Apache 2.0 licence, is not a conventional operating system such as Windows or Linux. Cloudflare describes it instead as an operating environment for AI-powered work, combining agent-based interfaces, application development tools and governance controls within a shared corporate workspace.
Cloudflare OS emerged from technology the company initially developed for its own workforce. The first version was made available across Cloudflare in May, giving employees in engineering, sales and other functions browser-based access to agents connected with company knowledge and internal systems.
Thousands of Cloudflare employees now use the platform each week. During a 30-day period, staff created more than 4,000 applications and tools, while the company estimated that sales teams saved more than 10,000 hours previously spent on manual tasks such as territory planning and preparing proposals.
The open-source version expands on that internal system by allowing organisations to deploy their own instance, customise its interface and connect it with company-specific applications, policies, knowledge and workflows.
Cloudflare OS provides an agent workspace where users can work through conversational instructions rather than traditional programming interfaces. The agent can create documents, analyse information, automate repetitive jobs and write code for small applications that Cloudflare calls “Gadgets”.
Those applications can contain their own user interfaces, server-side logic, APIs and persistent data. They can also be shared with colleagues in a manner resembling collaborative documents. Users may alternatively distribute a “Blueprint”, allowing others to create independent versions without transferring the original application’s stored data, credentials or conversation history.
A significant part of Cloudflare’s approach centres on restricting what autonomous agents can access.
Agents and applications initially receive no access to external resources. Permissions are provided through a framework called Gatekeepers, which sits between an agent and systems such as databases, software repositories or corporate applications.
Gatekeepers can restrict an agent to particular resources or operations rather than granting broad account access. A connection to a code repository, for example, could permit an agent to inspect issues while preventing it from reading source code or merging changes without approval.
Credentials are kept separate from agent-generated code. Cloudflare OS also records resources that an agent has accessed so that security policies can continue to apply when its work is shared with another user.
The mechanism is intended to address a growing enterprise concern around agentic AI. Unlike conventional chatbots, agents can perform actions, use software tools and combine information from multiple systems. Giving such systems long-lived API keys or unrestricted credentials can create risks if permissions are excessive or information moves between applications without adequate controls.
Cloudflare OS also supports Model Context Protocol servers, which have become an increasingly common method of connecting AI systems with external tools and data. Cloudflare argues that MCP access alone does not fully address what happens after an agent reads protected information, prompting it to add controls based on the resources an agent has actually observed.
Generated server code runs within isolated Dynamic Workers with general outbound internet access disabled. Browser-side code operates within a sandboxed frame. External connections must therefore pass through capabilities explicitly provided to the application.
The system is built primarily around Cloudflare Workers and its related infrastructure, although the company says Cloudflare OS can also operate on organisations’ own servers using workerd, the open-source runtime underlying Workers.
Organisations are not tied to a particular large language model. AI requests can be routed through Cloudflare AI Gateway, allowing administrators to select different model providers, apply rate limits, monitor expenditure and assign less expensive models to simpler workloads.
Cloudflare has released both the core Cloudflare OS repository and a separate starter deployment intended to help organisations customise branding, authentication, integrations and deployment processes. The company describes the software as early access and advises organisations to review changes and security boundaries before production upgrades.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
