Been building UnoRouter: one API key that reaches 315+ AI models across 47+ providers, with smart routing and automatic failover. It is OpenAI-compatible, so existing tools (Cline, OpenCode, SillyTavern, CLI wrappers) work by swapping the base URL. The part that took longest was normalising provider quirks: every vendor disagrees on streaming semantics, tool-call shapes, and error codes. Happy to answer questions if anyone is wiring up multi-provider routing.