Cloudflare open sources AI workspace platform for enterprise teams

Cloudflare has open sourced Cloudflare OS, a platform that lets organizations deploy AI agent workspaces connected to their own data, tools and internal processes. The company announces in an official Cloudflare Blog post that the software is designed to give employees a browser based workspace where agents can research, create documents, build applications and automate recurring work.

The project addresses a central challenge for workplace AI: useful agents need access to company information and systems, but that access can create substantial security and governance risks. Cloudflare says its platform combines an agent workspace, access controls for internal data and a system for building shareable applications.

Cloudflare first tested an earlier version internally. According to the company, thousands of its employees across different departments now use it each day. They use agents to draft documents and presentations, automate repeatable tasks and create small applications for visualizing data.

From chat to connected work

Cloudflare OS begins with a chat interface, but the company positions it as more than a general purpose chatbot. Each workspace can include organizational context and reusable “skills”. These can contain internal terminology, procedures and instructions for common tasks. The aim is to prevent employees from having to repeatedly explain company specific processes to an AI model.

Agents can also use an isolated environment to write and run code. That allows them to search, filter and analyse information without placing entire datasets into a model’s context window, Cloudflare says. The workspace can turn results into documents, spreadsheets or presentations. These outputs can remain connected to live data and can also be exported to services such as Google Drive.

For more specialized work, users can ask an agent to create an application. Cloudflare says these apps include a browser interface, server logic and their own persistent data. They are private by default but can be shared with colleagues. Teams can either collaborate in the same app or share a blueprint that lets others create an independent copy.

The platform also supports workflows for tasks that do not need an agent to reason through every step. A workflow can use conventional code for predictable actions and call an AI model only where judgment or language generation is needed. Such workflows can run on demand, on a schedule or in response to an event in a connected system.

Security follows the data

Cloudflare’s main technical focus is controlled access. The company argues that issuing broad API keys to employees or AI agents is difficult to limit and audit. Instead, Cloudflare OS uses services called Gatekeepers to sit between an agent and an external system such as GitHub or a data warehouse.

A Gatekeeper can restrict an agent to a particular resource and action. For example, it could allow access to issues in one repository while blocking source code, mask selected data fields or require approval before a pull request is merged. Credentials remain in the Gatekeeper rather than being exposed to the agent or generated application code.

Cloudflare also records which resources an agent has viewed. That record is intended to stop sensitive information from being exposed indirectly through a dashboard, app or shared workspace. When another user opens the output, the system checks whether that person has permission to access the underlying data.

The software works with Cloudflare Access for user authentication and Cloudflare AI Gateway for model selection, spending controls and usage tracking. Organizations can choose which models are available and set budgets or rate limits for teams and workspaces.

Cloudflare has released both the core software and an example deployment based on its own internal setup. The company says organizations can customize the interface, add integrations and build their own Gatekeepers without modifying the core project. It also plans a managed dashboard version, container support for development work and integrations with chat tools including Slack.