How I handle provider fallback + per-model cost attribution behind one OpenAI-compatible endpoint (open-source)

I built an open-source, self-hostable alternative to OpenRouter — one API for 280+ models, with cost tracking