From Inbox to Character: Building a Private, Local AI Email Agent

There are 18k+ emails in my personal inbox, and it's only one of the accounts I have. I wanted to search through them semantically, get AI summaries, draft replies, and run email campaigns - all from my phone. I didn't want OpenAI reading my emails or Google's AI. Or anyone's. For me, local AI is the only real answer for private data processing because no one can read your data or train their models on your data while you're paying for the service. So I built my own - a private, local email agent project called Llamail. Its default synthetic persona is Sable. It helps me search, summarize, and manage my emails, and can also chat in a casual, roleplay-like manner to make the whole thing a bit more fun. ~3700 lines of Python, a Llama model running on my average, consumer laptop GPU, and a Telegram bot as the interface - and everything is handcrafted, not vibecoded. Here's how. Meet Sable - a private, local email agent with a configurable synthetic persona. Two common workflows: RAG-based