Exactly Also I heard they are trying to finetune or host their own speech to speech model instead of using openai realtime api Which is different than vapi that uses other companies models