Cloudflare has open-sourced Cloudflare OS, a platform that gives every person in an organization an agent and workspace grounded in company context, internal systems, and curated skills. The first version ran internally at Cloudflare starting in May 2026, with thousands of employees across functions using it for documents, slides, task automation, and small data apps. The open-source release lets any organization deploy the same foundation, connect it to internal tools, and customize governance, access policies, and workflows.
The platform combines three capabilities: an agent workspace with isolated runtimes, a security and governance layer built around Gatekeepers, and a system for building modifiable apps that can be shared like documents. Inside the workspace, agents write and run code in an isolated runtime, access company resources through typed capabilities, and persist outputs that remain connected to live data. Gatekeepers mediate access to external services, enforce policies at the service level, and log every resource an agent observes so that later access checks can prevent unauthorized sharing. Apps are full-stack applications with client code, server code, APIs, and durable state, deployed on demand as Dynamic Workers backed by Durable Object Facets with SQLite storage. The platform also supports existing Model Context Protocol (MCP) servers through MCP Server Portals, letting organizations connect agents to tools they already use.
Cloudflare designed the system around the observation that most AI workplace tools hand out broad, long-lived API keys, which do not scale and are hard to audit. Instead, every agent and app starts with zero access. Requests for access are granted at the resource level, and credentials never reach the agent or generated code. The same observation log that records what an agent has read also governs downstream sharing: if a sensitive table was used to build a dashboard, Gatekeepers verify that a viewer can access the original table before showing the output. Model inference runs through Cloudflare AI Gateway, giving administrators visibility into spend and the ability to assign cheaper models to routine tasks.
The open-source release includes two repositories: the Cloudflare OS core and an example deployment based on Cloudflare’s internal setup. The example deployment consumes the core without patches and provides configuration, custom UI, internal integrations, analytics, and deployment pipelines. Strategic partners Presidio and Happy Cog will offer customization services for organizations that want to adapt the platform to their own terminology, systems, and workflows. Organizations exploring Cloudflare’s broader platform evolution can also review our coverage of Cloudflare’s Developer Platform migration for cdnjs for additional context.
