Cloudflare introduces Cloudflare OS, an open-source platform designed for enterprises to securely deploy AI agents, connected applications, and automated workflows across internal systems This article explores cloudflare ai. . The company's earlier version was deployed internally in May, offering staff technical and non-technical teams agent workspaces.
The new release includes improvements based on lessons learned from the initial deployment, focusing on collaboration, data access, and policy enforcement. The platform integrates with existing Model Context Protocol (MCP) servers via MCP Server Portals. Agent Workspace (Source: Cloudflare) Each crafted application incorporates browser client code and server-side logic, utilizing a Dynamic Worker for seamless execution and Durable Object Facet to maintain isolated state while leveraging SQLite storage for efficient data management.
By default, apps are private, but users can share them for real-time collaboration through the creation of independent copies without inheriting the original application's data, conversation history, credentials, or connected resources. Alternative sharing options allow colleagues to independently create identical applications without accessing the shared blueprint’s specific content. All AI inference requests traverse Cloudflare's AI Gateway, enabling administrators to approve models, track spending for users and teams, manage budgets and rate limits, and control access to high-cost frontier models.
Organizations can integrate the platform into their Cloudflare accounts, modify interfaces and integrations, add internal Gatekeepers, and establish their own Access policies, AI Gateway configurations, and enterprise data controls.
