infrastructure · always on

The Rack — the two small computers behind all this.

Most of what I use every day runs on two Raspberry Pis: a Telegram bot that triages my notes with an LLM, an agent orchestrator, nightly encrypted backups to a second machine, and a handful of pipelines.

FIG. 1 — a thought, in production: saved message → LLM triage → inbox card → one-tap approve. backups tick on their own clock.

The point isn't the hardware

The point is the loop. A thought dropped into Telegram at a red light becomes a triaged card in my notes with a proposed action, and my phone asks for one tap of approval. Ideas stop dying in "saved messages", and the whole thing runs on about as much electricity as a lightbulb.

Owned, run, understood

No managed services in the critical path: systemd units, SQLite, restic to a second Pi off-host, a mesh VPN to reach it all from anywhere. When something breaks, the fix is a journalctl away — and every failure has made the system more legible, not more magical.

A full write-up of the setup — hardware list, services, the triage prompt design — is coming as a free guide.