Sathi is private. I won’t be open-sourcing it because too much of it is wired into my personal accounts and household systems. But it’s the most important project on this list, because it’s the one I bet my own time on. Every concept I teach on The Agentic Engineer was first pressure-tested by handing the keys to Sathi and seeing whether I’d still trust it tomorrow.
Public agent demos are easy. Building an agent you actually trust with your inbox, your calendar, your money, and your family logistics is a different problem entirely. Every failure mode is real, and 'sometimes works' isn't acceptable.
Sathi is built on the same OpenClaw runtime I publish, plus a tightly-curated set of tools (Gmail, Telegram, calendar, WhatsApp, banking integrations). Every action is reviewable, reversible where possible, and logged. Failures are studied, not ignored. Most of what I've learned about harness engineering came from Sathi misbehaving.
Sathi runs daily. It's the proof point behind everything I publish on The Agentic Engineer: I'm not theorising about agents in production, I run one on myself. Cameos throughout the channel as the reason I have time to film at all.