The good news: even if you think the AI is good-for-nothing, rewriting HTML into legible plaintext is the one thing it can do pretty well. And probably even a 3B model would do well on this task, so you don't need to send the email body off to the cloud to be data-mined.
That's such a silly overkill use of AI. I'm not having my mail server kick off a 100W GPU every time I receive an email just to extract text from a structured document, nor am I having my computer drain my battery the moment I receive an email. Outsourcing the ridiculous power consumption to the cloud as some kind of dataleak-as-a-service is even worse. I'd rather try to parse the HTML with regex.
You could do that, actually. I brought up AI because it could result in slightly cleaner output than just the naive de-tagging, and because you can use it for general purpose text tasks - not just HTML to plaintext but also semantic message labelling/search, suggestion of task items in a to-do list, maybe some other things too.
