CloudFlare AI Gateway was quite convenient for me - I wanted to give my zeroclaw instance a limited budget to services like Image generation/Replicate/Fal.ai, which would mean for each service I'd have to run my own proxy that stores the keys and cuts off the real calls if we go over the limits. Easy to do but still extra thing to build and maintain.

Instead I put my API keys to cloudflare, set limits, and gave the agent the CloudFlare token, and in minutes it could contact tens of services.

edit: not to mention instead of loading balance to each service I could just keep balance on cloudflare that covers them all

> A little more work perhaps, but not hard at all

Extremely hard to verify it's been done properly across a large organization.