Cloudflare opens agent workspace platform to developers

Cloudflare has released Cloudflare OS, an open-source artificial intelligence productivity platform aimed at helping businesses create AI agents, develop applications, and automate workplace tasks while preserving data security. Operating as an environment for AI-powered work, Cloudflare OS integrates agent-based interfaces, application development tools, and governance controls within a shared corporate workspace.

Initially developed for internal use, Cloudflare OS became accessible to employees in May, enabling thousands of staff each week to use the platform.

The platform allows users to interact through conversational instructions, creating documents, analysing information, automating repetitive jobs, and writing code for small applications, known as "Gadgets." These Gadgets can possess their own user interfaces, server-side logic, APIs, and persistent data, and can be shared with colleagues, similar to collaborative documents.

Alternatively, users can distribute a "Blueprint," allowing others to create independent versions without transferring the original application's stored data, credentials, or conversation history.

Cloudflare OS emphasizes restricting autonomous agents' access to external resources, with permissions granted through a framework called Gatekeepers. Gatekeepers sit between agents and systems like databases or software repositories, limiting agents to specific resources or operations instead of granting broad account access. Credentials remain separate from agent-generated code, and the system records resources accessed by agents to maintain security policies when work is shared with others.

Built primarily on Cloudflare Workers and related infrastructure, Cloudflare OS can also run on organizations' own servers using workerd, the open-source runtime for Workers. The software is released as early access, and the company advises organizations to review changes and security boundaries before upgrading to production.