Four weeks ago I stood up an always-on Linux box that runs Claude Code agents while my laptop sleeps. It has logged over 800 agent runs since: six scheduled agents, five daily and one weekly, covering personal monitors, data pulls, and draft-for-my-approval writing tasks. This post is the step-by-step of how I built it, what it cost, and what I learned. (I'm keeping the specific workloads and the repo private; the recipe is the transferable part.)
I kept losing work to my own agents. Not in some dramatic way. Just: open a new Claude Code session, ask for a feature, and watch it quietly undo choices the last session made. Different naming. Different file structure. Sometimes it'd refactor something another agent wrote two hours ago because it had no idea that agent existed.
I added a blog to this site. It is markdown files, one build script, and one dependency.