Agreed, lack of advanced voice mode (conversational UX) is definitely something missing on Kagi Assistant, and as such, has me occasionally hopping onto ChatGPT. However, I do think the voice assistant should be an agnostic, open source tool, that serves as an abstraction layer above the models. This would allow the user to continue to choose which model they prefer to have the conversation with (I'm not sure any other companies providing access to multiple LLMs have this feature either, so could be a big win for Kagi!).
I'd go as far as saying this is an expected feature within a dedicated AI app these days, so relates to: https://kagifeedback.org/d/4752-kagi-assistant-app