Cloudflare OS: an open platform for agents, apps, and work
15:00 · August 5, 2026 · Cloudflare AI Blog

Cloudflare OS is an open-source platform that lets everyone in your company build apps, automate work, and safely access internal systems, shaped around what your organization knows and how it operates
Summary
Cloudflare OS is an open-source platform that supplies organizations with agent workspaces grounded in their own terminology, procedures, and internal systems. Each workspace combines persistent agent sessions, files, and an isolated runtime in which the agent can write and execute code. The environment begins with no access rights; agents must request specific resources, which are then supplied as typed capability bindings rather than raw credentials.
Access to external services is mediated by Gatekeepers, service-specific Workers that sit between the platform and systems such as GitHub or data warehouses. A Gatekeeper enforces fine-grained policies, records every resource an agent observes, and prevents the agent or its generated code from reaching the internet except through explicitly granted capabilities. Observation logs travel with any output or shared workspace, allowing subsequent policy checks to block leakage when collaborators or other agents attempt to view results derived from restricted data.
Work that begins as a conversation can be turned into documents, live-connected spreadsheets, or full applications. Each application consists of a Dynamic Worker running server logic and a browser client communicating through an object-capability RPC layer. The same code can be invoked by both human users and agents, and new instances created from shared blueprints receive independent state and credentials. Deterministic workflows can also be extracted from repeated tasks, reserving model calls for steps that require judgment.
Any model may be used; every inference request passes through Cloudflare AI Gateway so that organizations can route work according to cost or capability policies. The entire system is designed for enterprise self-hosting, allowing custom integrations while keeping security and provenance controls inside the platform rather than in individual applications.
Why it matters
Directly addresses AI security and privacy risks through built-in controls, highly actionable for Dutch/EU teams needing GDPR and AI Act compliance when deploying agentic systems.











