Is there not a (local) LLM that just reads your emails and puts them to spam or not? I feel like that wouldn't require an enormous model.