Most workplace AI tools still begin and end with a chat box. You ask a question, receive an answer, and then move the result into another app to finish the job.
Cloudflare wants to go much further. You see, the company has open-sourced Cloudflare OS, an AI-powered workplace platform designed to create documents, presentations, spreadsheets, workflows, and custom applications while connecting to the systems a business already uses.
Despite the name, this is not an operating system like Windows, macOS, or Ubuntu Linux. Cloudflare OS is closer to an AI-native office suite, although even that description does not quite capture what the company is attempting.
Instead of giving employees a fixed collection of applications, Cloudflare OS lets them ask an agent to create tools for specific jobs. A worker could request a project dashboard, reporting system, collaborative whiteboard, customer presentation, or issue tracker without first finding and deploying a separate piece of software.
Cloudflare calls these custom applications Gadgets. Each one can include its own interface, server-side logic, API, and stored data. Users can share a working Gadget with colleagues or distribute a Blueprint that lets other people create their own separate copies.
The entire project is available through the official Cloudflare OS GitHub repository. Organizations can deploy it into a Cloudflare account, connect internal services, select supported AI models, and customize the interface and instructions around how the company actually operates.
Cloudflare has already tested the basic concept inside its own workforce. CIO Sam Rhea explained in a separate post about Cloudflare’s internal AI use that employees wanted agents capable of working with company systems, but providing broad API keys would have created serious security risks. That concern shaped much of Cloudflare OS.
Agents and generated applications begin with access to nothing. A user must grant permission to a particular resource, such as a GitHub repository, document, dataset, or internal service. The credentials remain outside the agent’s reach.
Cloudflare manages these connections through services it calls Gatekeepers. A Gatekeeper can restrict what an agent may see, block certain actions, hide sensitive fields, log activity, apply rate limits, and require human approval before making an external change.
This is more focused than simply connecting an agent to an MCP server and allowing it to call every available tool. Cloudflare argues that businesses also need to know which underlying information an agent has observed and where that information could travel afterward.
For example, an agent could read a restricted database and use it to create a dashboard. Cloudflare OS is designed to prevent that dashboard from becoming a back door through which an unauthorized employee can view the original information.
It is a very ambitious security model, and Cloudflare is clearly making safety a central part of the pitch. Still, companies should test those protections carefully before giving a young platform access to sensitive systems. Cloudflare describes the current release as early access and openly acknowledges that it has rough edges.
The internal usage figures are pretty damn interesting. Cloudflare says thousands of employees use the platform each week and created more than 4,000 applications and tools during one recent 30-day period. The company also estimates its sales teams saved more than 10,000 hours on territory planning, proposal creation, and other manual work.
Keep in mind, those are Cloudflare’s own measurements rather than independently verified results. However, they suggest this is more than a small experiment hidden inside the engineering department.
Rhea described one example involving the company’s IT help desk. A process that previously required downloading ticket data, moving it into Google Sheets, creating charts, and reviewing requests manually became an application inside Cloudflare OS.
The predictable portions can run as ordinary code, while an AI model handles jobs that require judgment, such as interpreting a ticket or drafting a response. That separation matters because not every step needs an expensive model consuming tokens.
Cloudflare OS works with multiple model providers through Cloudflare AI Gateway. Administrators can decide which models employees may use, monitor spending, set budgets, and reserve more expensive models for harder tasks.
The comparison to Microsoft 365 and Google Workspace is not perfect. Cloudflare OS is not currently a polished replacement for every mature application in either suite, and deploying it inside a business will require real technical work. Authentication, integrations, policies, Gatekeepers, and external services all need to be configured correctly.
The more interesting point is that Cloudflare is questioning whether businesses will continue to need the same fixed office applications at all.
Microsoft and Google are adding AI assistants to familiar software. Cloudflare is proposing a workplace where employees describe the tool they need, let an agent build it, and then continue modifying it as the job changes.
That vision may prove too complicated for many organizations, particularly while the software remains in early access. Still, releasing Cloudflare OS as open source gives developers and businesses a chance to evaluate it without relying entirely on Cloudflare’s promises.
This is not yet the end of Microsoft 365 or Google Workspace. It is, however, a credible argument that the next office suite might not look much like an office suite at all.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz