Picks the cheapest model that can actually finish the job, and fails over when a provider dies instead of burning a frontier model. For teams shipping LLM apps who want free-tier pipelines, cost-aware routing, and honest fallbacks.