Hello there, my team is working on integrating an assistant into our mobile apps. We’re using the iOS and Android SDKs for this purpose, and we’ve configured a permanent assistant in the Vapi dashboard. We’ve noticed that the assistant call initialization process (POST /call/web + WebRTC with Daily) takes between 4 and 7 seconds (over a Wi-Fi connection) for the assistant to be fully ready to speak when using the Assistant Speaks First setting. We tested an assistant with a bare-minimum configuration (no tool calling; the instructions simply tell it to greet the user), and we still observe this delay. Is there anything that can be done to reduce this initialization latency as much as possible (ideally to around 3 seconds under optimal network conditions)?

This is a real and fairly common issue with Vapi + Daily, so you’re not alone. That 4–7 second wait usually isn’t caused by your assistant setup, but by WebRTC connection setup, media negotiation, and the assistant warming up before it can speak first. Even with a minimal assistant, that overhead is still there. You can shave off a little time by reusing sessions, keeping assistants warm, and simplifying voice/model choices, but getting a consistent ~3 seconds is very hard right now with WebRTC. One thing that helps teams a lot is letting the user speak first or playing a short local greeting while the assistant finishes loading are you starting a brand-new call every time, or keeping a connection alive between interactions? @henryd

@Jarrah Yes, I believe we start a brand-new call every time a web call is initiated. > or keeping a connection alive between interactions? Is that possible? If so, could you explain how that's done?

Hi there, Please call your assistant a few times so that the initial message gets cached on the server which will remove the delay for the first message. Regards Vapi Support.

Could you elaborate on how that should be done?

Hi there, Please call the assistant a few times using the web call functionality so that the initial message gets cached on the server, that should solve the issue of the latency. Regards Vapi Support