Provider Pool

Route requests across a pool of LLM providers the way an answer engine has to: fail over when one errors, never retry a bad request, respect each provider’s rate limit, prefer the cheapest within a budget, and stop calling a provider that keeps failing until it has had time to recover.