Estimate token count and cost for text across major LLM models without making API calls.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.