Hey guys Im trying to setup voice phone agent When using openai or gemini real time its very fast and good. But when using stt llm tts pipeline using gemini or openai models its very very slow … seconds to get answers and bad answers. Response time in real time is less than 1.5 sec on phone In pipeline is more than 4-5 sec

Hi Jeremy, i noticed that you already have enquired the ask ai service, I mostly agree with their statements. I would like to also add that using a faster LLM model should be the biggest lever with streaming enable. And then using the agent observability feature should allows you to find the biggest bottleneck.

Thks a lot