I’m experimenting with the latest ChatGPT 4o model in Vapi and I’m running into some hallucination issues. I’m wondering if my system prompt might be too large and causing the model to drift. I’m considering restructuring my setup using a squad‑style approach, for example: one main routing agent one agent dedicated to booking one for cancellations one for rescheduling, etc. Would splitting tasks across multiple specialized agents actually help reduce hallucinations, or is this unnecessary and unlikely to improve reliability? Also, what are the best practices you recommend for minimizing hallucinations when using ChatGPT models in Vapi? Thanks in advance!

Hi, Splitting into Squads can help reduce hallucinations if your current assistant handles too many responsibilities in one large prompt. When logic is separated (routing → booking → cancellations → etc.), each assistant has clearer goals and fewer tools, which improves reliability. That said, hallucinations are usually caused by unclear tool rules or ambiguous instructions rather than prompt size alone. To improve stability: - Make tool calls explicit (“MUST call X before responding”) - Lower temperature for production - Keep each assistant focused on one task Squads can help structurally, but tightening prompt logic is just as important.