Cloudflare has open-sourced its internal AI agent workspace, Cloudflare OS, allowing employees, including non-coders, to build applications using natural language prompts. The platform features a robust security framework designed to mitigate risks associated with "vibe coding," ensuring AI agents cannot introduce significant security flaws or lead to data breaches.

Developed over several months, Cloudflare OS creates highly secure "isolates"—instances of the V8 JavaScript execution engine—which are significantly faster and more memory-efficient than traditional containers. AI agents start with minimal permissions, requiring explicit requests to access resources, and server code operates with disabled outbound networking. This granular sandboxing and permission model aims to make AI-assisted coding safer for organizations.

The platform supports various AI models, allowing administrators to manage AI spending through budgets and rate limits. Cloudflare learned from early challenges, such as providing overly complex tools to non-technical staff, leading to the development of the Cloudflare Engineering Codex to uphold coding standards. The open-source version requires a Workers Paid plan for deployment.