The Short Version
Windows 11 told me my perfectly good hardware was obsolete. So I installed Linux. Then I decided to try running language models locally. Then a job change gave me an excuse to justify new hardware. Now I have dashboards to track all the things I accidentally built.
This is a journal of my crazy journey with Linux and local models.
It Started with a VM
I’d been running Home Assistant in a VM on my Windows 10 laptop for a while. It worked, mostly. But Windows 10 had this habit of requiring updates at the worst times. Slow, manual reboots that would take the VM down with them. Occasionally things wouldn’t come back cleanly. It was the kind of setup that worked until it didn’t, and it didn’t often enough to be annoying.
So I bought a Home Assistant Yellow. Dedicated hardware, purpose-built for Home Assistant, no more VM, no more Windows pulling the rug out from under my automations. It’s been rock solid ever since. That fixed the immediate problem.
It also freed up the laptop. I’d actually tried Linux years ago and bailed. It didn’t solve anything I didn’t already have working in Windows. But with HA off the laptop, I figured I’d put Ubuntu on it just to see how things had changed. Turns out Linux has come a long way. It’s a lot more usable than I expected.
Then Windows 11 Locked the Door
October 2025 hit. Windows 10 stopped getting security updates. My desktop? Not supported for Windows 11. Too old. This machine runs everything I use daily perfectly fine, but Windows can’t go a week without breaking something, and now you’re telling me it’s not good enough? That was it for me.
By late November, early December, every computer in the house was running Linux. The spiral had started.
Things Spiraled from There
I dug an even older Dell laptop out of a closet, installed Ubuntu Server on it, and put it to work. Docker containers, monitoring services, Grafana dashboards. A Raspberry Pi 5 became a NAS. The HA Yellow was already humming along.
Then I decided to try running large language models locally.
Before I go further: these are not intelligent. The industry calls them AI because that sells, but what you’re actually running is a very sophisticated pattern matcher that predicts likely next words based on training data. It doesn’t know things. It doesn’t understand things. It doesn’t think. We are genuinely not there yet, and a lot of the hype assumes we are. I’ll be calling them language models or local models throughout this site, because that’s what they are.
I started with a 7B parameter model on one of the laptops. Slow, rough output, but it worked. No API key, no subscription, no data leaving the house. That was enough to get me curious about what a proper GPU could do.
So I built a dedicated inference server. A mini PC with 128GB of unified memory and an integrated GPU. It’s run models up to 122 billion parameters. I’ll get into the details in a later post, including the honest answer to “was it worth it?”
What’s Running in the Basement
Here’s the current inventory:
- Home Assistant Yellow running Zigbee devices, automations, and energy monitoring
- A dedicated inference server running llama.cpp and language models up to 122B parameters on consumer hardware
- An old Dell laptop pulled out of a closet now running Ubuntu Server, Docker containers, Grafana dashboards, and network monitoring
- A Raspberry Pi 5 as a NAS with file shares and a dev environment
None of this was planned. Each piece solved a specific problem, and each solution created two new projects. Now I have a local LLM helping run the house. That story is still ahead.
What This Site Is For
Everything here comes from things I’ve actually built, broken, and run in production (if you can call a basement “production”). I’ll be writing about the Windows frustration that started it, the Linux migration, running local models on consumer hardware with honest benchmarks, home automation that doesn’t require a cloud subscription, and the tools I’ve built to track the mess.
What’s Next
Next up: the full story of Windows undermining my smart home, and the dedicated hardware that replaced it. Then building a NAS from parts I already had, and how I ended up going full Linux.
Every Thursday I send a dispatch from the basement. Subscribe here. It’s free, it’s short, and it comes straight from the windowless basement.
— Mike