Cloudflare Launches Open-Source AI Agent Workspace Cloudflare OS With Zero-Trust and Custom Model Support

Cloudflare officially launched Cloudflare OS, its open-source AI agent workspace, on August 6, 2026. The browser-based platform comes with a library of shareable, custom-built micro-applications, offering enterprise employees a smart collaboration environment. Cloudflare developed the platform from the daily needs of its own teams, with internal staff using the system each day to conduct research, write documents, and build applications for their routine work.

From the first day of enterprise adoption, Cloudflare OS captures and accumulates internal knowledge. Employees without professional development backgrounds can draw on the company's internal knowledge base to create bespoke micro-applications, set up workflows, and collaborate and share within their organizations. The framework is fully open source and runs on the enterprise's own Cloudflare account, giving companies complete control over the workflows, contextual data, and system connections they build. The system defaults to a zero-trust architecture, supported by Cloudflare Access technology, which strictly verifies every access and request.

The platform introduces a custom models feature, allowing enterprises to connect any AI model provider through Cloudflare AI Gateway. Managers can precisely control token usage and expenditure, and view cost details by person, team, and application. Administrators can also set budgets and rate limiting rules, or route requests flexibly based on task complexity - for example, assigning routine tasks to small, efficient models and calling on frontier large models when advanced reasoning is required.