Counts tokens and estimates LLM costs for supported models before each API call.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.