Cloudflare launches open-source OS for AI agents, enabling autonomous apps and workflows on its edge infrastructure.
AI & Agents ·
Cloudflare has open-sourced a rebuilt version of Cloudflare OS, a platform that bundles an agent workspace, security governance layer, and tools for deploying personal apps as Cloudflare Workers. The company initially built the system for internal use, giving its employees access in May, but is now releasing it so other organizations can deploy and connect it to their own systems. The architecture emphasizes keeping agents isolated from broad API access: agents request specific resources, and a "Gatekeeper"—a service-specific Worker—mediates each request, ensuring credentials never reach the agent itself.
The platform addresses a core security tension in autonomous systems: handing agents or users wide-access API keys is difficult to audit and constrain. Cloudflare OS instead logs every observation and enforces access checks before an agent can view or act on data. Each app an agent creates becomes a real Cloudflare Worker using Dynamic Workers and Durable Object Facets, communicating through Cap'n Web, Cloudflare's open-source object-capability RPC system. The security framework sits between agents and systems of record, making governance a platform feature rather than something each builder must implement independently.
A structural trade-off remains unresolved: while the code is open source, the runtime lives on Cloudflare's edge infrastructure under the company's control. This means agents, apps, and governance all execute within Cloudflare's centralized environment, even when deployed by external organizations on their own data.