Hand-Me-Down Homelab

At 6:32 every morning, an AI reads the alert emails my house sent overnight and decides whether any of them need me. Most mornings, none do. The lab runs in my basement, on a 2012 Mac mini, a tower built from a friend’s spare parts, a small mini PC and a few Raspberry Pis. The one purchase the book can’t avoid is a graphics card.

This book shows you how to build your own version from whatever is in your drawer. Each chapter says what you’ll build, what you need and how hard it is:

• Hardware and network: Proxmox on old machines, backups that fail loudly, and a separate network for smart-home gadgets
• An AI sysadmin: a wiki the AI keeps up to date for you, house rules it follows, and a record of its own mistakes
• Keeping it alive: monitoring, alerts you’ve proved actually arrive, and an AI that reads them every morning
• Home Assistant: Zigbee sensors, a spoken door chime, radar stair lights, dashboards and a nightly report card
• Names, certificates and remote access: your own DNS, your own certificate authority, and remote access with no open ports
• Local AI on one graphics card: a private chatbot, images, video and music, and seven AI applications sharing one card
• Agents and apps: a sandboxed AI agent, a searchable family photo library, and your first iPhone app
• Files, movies and a radio: private file sync, home movies on the living-room TV, and a radio anyone in the house can tune
• Eyes on the street: cameras that tell a person from a car, and the speed of every car that passes
• Watching the wire: an intrusion-detection sensor, a hundred thousand alerts a day tuned into a readable digest, and passive Wi-Fi monitoring

Real failures, with dates. Every chapter has war stories from the lab, told as symptom, wrong guess, cause, fix and lesson. Most guides show you the path that works. The lessons are in the paths that didn’t.

Learn to work with an AI on something real. Chatting with an AI teaches you little about trusting it. This book gives one a job: reading logs, running commands and fixing things on your own network. It shows you how to set that up so you can trust the result: the notes it keeps, the house rules it follows, asking before anything irreversible, and checks that prove the work was really done. The assistant can’t touch money or passwords, and it’s wrong often enough that everything is built around catching it. Those habits carry over to any AI tool you use, at work or at home. An appendix gives you the starter files to copy.

Installs are playbooks. Where a section installs something, a companion playbook is written for your AI assistant to carry out on your network: it finds your addresses, asks before anything risky, and checks each step by its effect. Each was tested on a clean machine, except the few that need hardware a test machine can’t have, and those say so. The playbooks and scripts are free under the MIT license at handmedownhomelab.com/code, so you can read them before you buy.

Also inside: what the lab really cost and what I’d do differently, the minimum lab for one old PC and one Raspberry Pi, and how to keep your family, street and accounts out of it when you publish your own.

For tinkerers who know some Linux, have an old machine doing nothing, and want to learn to work with an AI on something that matters.

Updates are free. Errata and new editions reach every buyer at no charge.

How this book was made: Luis Albisu built and runs the lab in this book. Claude, an AI, drafted and revised the text from the lab’s own records, and every claim was checked against the running lab.