I was just thinking that doing non stop local speech to text is feasible and it could then infer a journal or something like that with a small local model. You would then put a mike on you and just say something like "remember this".

I think you can vibe it in a day.