Butter is a cache that identifies patterns in LLM responses and saves you money by serving responses directly. It's also deterministic, allowing your AI systems to consistently repeat past behaviors. It's live — try it out here.