It started with one NAS. It did not stay that way.

In 2020 I bought a Synology NAS to stop losing photos. Six years later it’s a rack of machines, a structured network and about thirty containers — and I’m still not entirely sure how that happened.

How it’s designed

Three principles, learned by breaking them first.

• One job per machine. Consolidating everything onto one box is how a hobby becomes an outage. Separate hardware for separate purposes means I can take something down without taking everything down.
• Everything in containers. Docker means any service I run is described in a file, rebuildable, and disposable. I can destroy a container with no ceremony — which is the only way I’ve found to actually experiment.
• The network is segmented. Home automation, personal services and my development work sit on separate logical networks. Not because I expect an attack, but because I want to know which device is talking to what.

What it’s for

Self-hosted tools. The things most people rent — file sync, media, photos, notes, documents, reading, passwords. My data lives on my own hardware, on my own disks, backed up on a schedule I set.

A development and staging ground. Nearly everything on this site runs somewhere in that rack before it goes anywhere else. It’s where I break things safely, on purpose, with room to get it wrong.

Home automation. Home Assistant has been running this house since 2020 and has never once been switched off. Lighting, climate, energy monitoring, presence, and the small automations that quietly remove friction from a family’s day. It is the least glamorous and most-used thing I run.

And increasingly, an agent. I run a self-hosted AI assistant that helps me manage and monitor the lab — reading metrics, tracking what’s running, chasing problems, and doing the tedious parts of keeping thirty containers alive. Running the infrastructure and building the intelligence that watches it turned out to be the same hobby.

What I’ve learned

The hardware is the easy part. The hard parts are backups you’ve actually tested, monitoring that tells you something is wrong before someone else does, and the discipline to document a change when you’re in a hurry. I’ve got all three wrong at least once.