The workshop log explains why decisions were made. This series is the other half of the deal: the actual how, distilled from services that run 24/7 on the hardware described in the specs page. Every guide follows three rules — it works on plain Linux, it survives a reboot, and it was extracted from something still running rather than written from memory of something that once worked.
The Self-Hosted Service Skeleton
Dedicated user, directory layout, systemd unit, health endpoint, log rotation — the minimum viable production service, ready in an afternoon.
systemd Discipline for People Who Like Uptime
Restart policies that actually restart, hardening options worth the read, dependency ordering without tears, and why your service dies at boot anyway.
Watchdogs That Don't Cry Wolf
Threshold design, rate-limited alerts, checking reality instead of telemetry (the lesson behind the black-hole bug), and the art of staying quiet most nights.
Snapshot Before You Regret It
A pre-experiment snapshot policy with zero exceptions, backup rotation that fits on one node, and restore drills — because an untested backup is Schrödinger's archive.
Scheduled Jobs Without Cron Hell
Picking between cron, timers, and in-process schedulers; staleness detection via health endpoints first and logs second; making jobs survive deploys.
Multi-Provider API Routing on Consumer Money
The practical companion to Secondary Agent Economics: routing config, fallback order, credit-cap detection, per-leg burn tracking.
// request line open
Want one of these moved to the front, or a guide on something else running here? Ask. The queue reorders by demand.