Hey HN, we just shipped v0.1 of Hebo Gateway.

There are plenty of gateways already, but we kept running into the same issue: once you need real customization (auth, routing, rate limits, observability, request/response transforms), most “off the shelf” gateways get hard to extend.

Hebo Gateway is for cases where you want the gateway to be part of your app. You can run it standalone, or embed it into an existing backend. It exposes OpenAI-compatible endpoints (/chat/completions, /embeddings, /models), works with any Vercel AI SDK provider, and adds a hook system so you can plug logic into the request lifecycle without forking the core.

Quickstart, examples, and “what’s next” are in the post: https://hebo.ai/blog/260127-hebo-gateway

I would love feedback on OpenAI-compat edge cases you have been bitten by (especially streaming and reasoning-related stuff), and what hooks you wish gateways provided out of the box.