Kudos for saying no. I can't believe we're living in an age where open source developers are being asked openly to leak the private conversations of their users to the public.

Makes sense, thanks for considering (and fwiw I'm also building with a private AI / offline philosophy). In my implementation, the compromise is 1) being super loud with an alert warning the user when they're sharing feedback and 2) providing a setting where the user can turn off feedback buttons. I had gotten user feedback around wanting to tell the AI when it was doing a good/bad job and this seemed like a privacy preserving compromise that could actually improve models. Anyway, good luck and congrats again!