i keep getting stuck on this when building AI app flows. the model gets way more useful if it knows the user a little, but passing in too much context feels like a privacy problem waiting to happen. i tried normal onboarding questions. too thin. tried storing app events. useful later, useless on day 1. tried asking users to paste context manually, but nobody wants homework. i’m wondering if the better pattern is a consented user context API where the user approves what gets passed into the...