@antharmaya/tokencost

Price a whole LLM agent run, not one request. Models context regrowth across turns, prompt-cache write premiums and expiry, the minimum cacheable prefix, long-context rates, tool-definition overhead and thinking tokens.