[mschild]

> A small local model or batched inference of a small model should do just fine.

Or, you know, Signal/Matrix/WhatsApp/{your_preferred_chat_app}. If you're already texting things, might as well do that.