Cloudflare OS enterprise workspace launches as open-source agent platform

Cloudflare has launched Cloudflare OS, an open-source enterprise workspace for AI agents that runs on its network. The Cloudflare OS enterprise workspace is available now in an open-source repository, while the company says a managed version through its dashboard is coming soon.

The product is Cloudflare’s attempt to turn its agent infrastructure into an employee-facing work surface. Cloudflare says the browser-based software lets non-developers conduct research, create documents tied to live data, run background workflows and turn outputs into shareable applications. Those apps can have isolated databases, real-time features and access controls, according to the company.

Cloudflare said it originally built the system for internal use and claims that thousands of its employees use it across teams. That is a company claim rather than independent evidence of adoption or operating results.

What is Cloudflare OS and how does it differ from Agent Cloud?

Cloudflare OS is presented as a persistent workspace where an employee can work with an agent using company context, files, tools and reusable work practices. Cloudflare’s reference architecture describes outputs such as research, documents, spreadsheets, workflows, source code and applications as durable work products that can be reviewed, shared or extended after a conversation ends.

It is distinct from Cloudflare Agent Cloud, the developer-oriented infrastructure suite Cloudflare expanded in April. Agent Cloud includes components for building, deploying and scaling long-running agents, including Dynamic Workers, Sandboxes, Artifacts and the Think framework. Cloudflare OS sits closer to the end user, packaging related building blocks into a workspace intended for an organization’s workforce.

Cloudflare’s technical documentation describes a reference design using Workers and the Agents SDK for orchestration, Durable Objects for state, AI Gateway for model governance, and MCP server portals to connect enterprise tools. It is architecture guidance, not a statement that every customer deployment will use the same configuration.

How does Cloudflare OS control access and model use?

Cloudflare says Cloudflare Access verifies each user and request, while agents begin with no permissions and receive only task-specific access. The company says its Gatekeepers let internal-system owners set what an agent may view or change, and can require human approval before certain actions proceed.

For model choice, Cloudflare says AI Gateway allows organizations to use different model providers rather than committing the workspace to one provider. It also says administrators can view usage by person, team or app, set budgets and rate limits, and send routine work to lower-cost models. Those security, governance and cost-management outcomes are Cloudflare’s claims.

What is available now?

• Available now: Cloudflare says the open-source Cloudflare OS platform can be obtained from its repository.
• Coming soon: A managed deployment through the Cloudflare dashboard, according to the company.
• Implementation help: Cloudflare named Presidio and Happy Cog among partners that can tailor deployments for customer environments.

The launch extends Cloudflare’s push from agent runtime and model-access infrastructure into enterprise software for internal work. The operational question for prospective users is whether the controls around internal tools, context and approvals can be configured tightly enough for real workflows, rather than just producing another chat interface connected to corporate data.

This story draws on original reporting from SiliconANGLE.