On August 5, Cloudflare open-sourced a new version of Cloudflare OS, a platform that gives every person in an organization — not just engineers — an AI agent and a workspace built around how their company actually works. The company has been running it internally since May, with thousands of people across every function using it daily to draft documents, automate repeatable tasks, and build small internal apps. Now anyone can deploy it: the full source is on GitHub at cloudflare/cloudflare-os, ready to connect to your own tools, data, and Cloudflare Access policies.
For agencies and automation teams that have spent the last two years wiring together ChatGPT, Zapier, and custom scripts, Cloudflare OS is worth a close look. It is not another chatbot wrapper — it is an attempt to build the infrastructure layer underneath agentic work: identity, permissions, and app hosting, all in one open-source package.
Cloudflare OS combines three parts. An agent workspace lets anyone start a conversation grounded in context and skills their team has curated — company terminology, standard procedures, the best-known way to do a recurring task — so nobody has to re-explain their business to a model every time. A security and governance framework called Gatekeepers controls what agents can see and do. And a platform for personal, modifiable apps means every "file" an agent produces can be a real full-stack application, not a static export.
You don't need to be a developer to use it. Workspaces run in the browser, and an agent can turn a research question directly into a live document, a shareable dashboard, or a small internal tool with its own interface and database — all without anyone touching a terminal.
The more interesting part for anyone building with AI agents in a business setting is what Cloudflare built to stop agents from leaking things. Every agent inside Cloudflare OS starts with access to nothing. It has to request access to a specific resource — a GitHub repo, a support queue, a data warehouse table — and that resource shows up in generated code as a typed, capability-scoped binding rather than a raw API key. A service-specific "Gatekeeper" worker sits between the agent and each external system, deciding exactly which operations are allowed, masking sensitive fields, and requiring approval before anything with a real-world side effect (like merging a pull request) goes through.
Critically, the platform tracks what an agent has observed, not just what tools it called. If an agent reads a sensitive spreadsheet to build a dashboard, sharing that dashboard doesn't quietly hand the underlying data to someone who was never allowed to see it — the Gatekeepers check the viewer's access against everything the agent touched along the way. That is the exact failure mode most homegrown "AI plus internal data" projects run into, and Cloudflare built it as a platform guarantee instead of leaving it to whoever wires up the integration.
In practice, a workspace can research a topic using a company's own context and resources, write it up as a live document or spreadsheet that stays connected to its source data, or become a full app that a team uses together. For repeatable jobs, Cloudflare OS supports deterministic workflows — a known sequence of steps where a model only gets involved at the one or two points that need judgment, triggerable on a schedule or an event from a connected system. It also plugs into any model through Cloudflare's AI Gateway and works with existing MCP servers an organization already runs, via MCP Server Portals.
Agencies and no-code/automation shops already living in this space — building internal tools, standing up client dashboards, gluing together SaaS tools with agents — now have a self-hostable, open-source option for the layer they've been building by hand: identity, governed data access, and app hosting for agent-produced software. It won't replace a Zapier flow you built in ten minutes, but for anything that touches real company systems and multiple people, the access-control model alone is a meaningful upgrade over "give the agent an API key and hope."
Our cutting-edge features simplify collaboration and creativity, making your workflow intuitive and efficient. Transform your vision into reality effortlessly with Hadidiz Flow.
