Estimates the run's tokens and cost. Each call's prompt is the state and questions the builders
produce for the point (with the point's own agent names, since no engine is loaded offline) at
four characters a token, plus a fixed allowance for the prompt around them. A floor, not a
ceiling.
Estimates the run's tokens and cost. Each call's prompt is the state and questions the builders produce for the point (with the point's own agent names, since no engine is loaded offline) at four characters a token, plus a fixed allowance for the prompt around them. A floor, not a ceiling.