I used to spend two hours every morning reading. Credit markets, macro, my portfolio, prediction market odds, research threads, Substack drops from the night before. Good information - but it was manual, repetitive, and happening in my head instead of somewhere persistent.

Two years in restructuring gives you a very specific kind of attention. You read for signal - bankruptcy filings, DIP amendments, credit spread moves, the one paragraph buried in a 10-Q that changes the whole picture. I wanted that same lens running 24/7, not just the two hours I had in the morning before work started.

At some point I realized: this is all just pattern matching and retrieval. Exactly what language models are for. The question wasn't whether to automate it - it was how far to push it, and how to make it actually know me rather than just know the internet.

So I built Kai. A locally-hosted agent running on a Mac Mini under my desk - connected to my life. It holds 13,000+ files of personal context. It knows my portfolio, my thesis areas, my reading list, the Discord channels I care about, the Polymarket positions I'm tracking, the India angle I'm always thinking about. And every morning, before I pick up my phone, it's already done two hours of work and sent the summary to Telegram.