The estate
A Raspberry Pi 5 runs openSUSE MicroOS as the always-on agent host. Home Assistant lives on a separate Orange Pi Zero 3 with a Mosquitto broker and Zigbee devices. A Hetzner arm64 VPS in Helsinki is the public reverse proxy and WireGuard endpoint for the home network. The desktop is Fedora Kinoite.
The point is not the hardware list. The estate is treated like a small production system: an inventory, a documented network, a measurement protocol, and postmortems after failures.
Operating it
- Inventory and naming. Every host has a role and a canonical name; the inventory is the source of truth, not memory.
- Power and availability. A UPS bridge publishes state over MQTT, with a watchdog that reacts to loss of mains and protects the Pi from unclean shutdowns.
- Network. WireGuard connects the VPS and the house; the VPS terminates TLS and fronts Home Assistant.
- Zigbee. A network-attached SLZB-06M coordinator serves five devices across two buildings. I scored configurations by the worst link, not the best, and recorded antenna orientation, door state, and occupancy per reading. Antenna orientation moved links by up to 16 dB; a steel door cost about 6–7 dB. The measured floor, not the average, drives placement.
- Documentation. Incidents get a written timeline, root cause, and follow-up actions. The Zigbee notes include the measurement protocol so tomorrow’s numbers are comparable with yesterday’s.
What this demonstrates
Small systems still have the same failure modes as large ones: silent dependencies, ambiguous ownership, and measurements taken without a control. The habits transfer directly to production: baseline first, change one variable, write down what you did, and design to the worst case.