AI Gateway

AI Gateway is a proxy for your LLM API calls that gives you logging, analytics, caching, rate limiting, and provider fallbacks — mostly free. Given that every team building with LLMs eventually hand-rolls half of this, getting it by changing a base URL is real value, and the unified analytics view across providers (OpenAI, Anthropic, Workers AI, and friends) is something the providers themselves won’t give you.

The core product is genuinely solid: the proxy behaves, the caching pays for the integration effort on repeated-prompt workloads, and the controls (rate limits, fallbacks) do what they say. The deductions are the extra hop in your most latency-sensitive path, occasional lag supporting new provider API features, and logs that are useful for debugging but not a replacement for real observability tooling.

The one feature to actively avoid is Unified Billing. Routing your provider inference spend through Cloudflare’s billing is a hard sell when Cloudflare has been known for billing issues for years — surprise invoices, opaque line items, slow corrections. That’s not the system to put between you and your inference bill. Happily it’s optional: bring your own provider keys, skip Unified Billing entirely, and the rest of the product is an easy add in front of existing LLM usage.