This past year, my job as Cloudflare's CTO included a problem you probably recognize: get AI into the hands of every member of our team, without losing control of it.
It started to get messy, fast. Models from half a dozen providers. API keys pasted into scripts and forgotten. Token bills nobody could explain. Agents reaching for tools and data with no one watching. Every team moving quickly, and no single place to see any of it.
So we built the controls we needed. Cloudflare’s expertise lies in proxying traffic and applying controls. This is exactly what every team needs to manage their AI usage. Today, every AI call and every agent at Cloudflare runs through two products already in your dashboard: AI Gateway and MCP portals.
These are the same tools that I use. Here are the five ways Cloudflare uses them every day:
1. Every model, one place. I point every team at a single endpoint that fronts every provider — OpenAI, Anthropic, Google, Kimi, GLM, our own Workers AI. Pay through Cloudflare on one bill or bring your own keys.
2. Spend that I can see and cap. Every request, token, and dollar shows up in one view. We set limits by model, provider, team, user, or app, protecting against runaway agents and surprise invoices. We're constantly innovating in the space, and wanted to give you a sneak peek of a cost control dashboard, coming soon. |