Hey, quick question about structured outputs. When an assistant has multiple structured outputs (e.g. 12), does Vapi make a separate LLM request for each one after the call ends? Or are they batched into one request? Trying to understand if combining multiple structured outputs into one would actually reduce token usage.

From my experience working with Vapi, structured outputs aren’t sent as separate LLM calls one-by-one after the call. They’re typically handled in a single pass (or minimal calls) based on the final transcript + config. So having 12 outputs vs combining them into one doesn’t usually mean 12x token usage but combining them can still be slightly more efficient since you reduce schema overhead and repetition in the prompt. @cem802

I see, thanks for you response. I think then in our case it wont make sense to combine them

yeah so right now each structured output is its own llm request so if you have 12 outputs → that’s 12 separate requests that means your prompt tokens are basically being sent 12x if you combine them into fewer outputs, you’ll reduce token usage a lot since the transcript only gets sent once so it’s definitely more efficient, just keep in mind bigger schemas can sometimes be a bit harder for the model to fill accurately