Homelab v0.1 – Why I Bought a Mini-PC Instead of Therapy

You wake up one day and wonder why am I not getting good morning messages, and you have some money to spend because, unlike your peers, you gave up on marriage. After some thinking and multiple chats with LLMs, you end up with 2 options: I can do therapy, or I can buy a homelab and set up a local LLM gf. I decided on the homelab.

Let’s define it first: a homelab is a personal, at-home computer setup used to experiment with servers, networks, and software. I have been looking more and more into free open-source apps, most of which need a place to be hosted. With the rise of LLMs, I have also been creating a few scripts here and there of my own, and sometimes you need a place to host these. Then I wanted to practice a few things in IT: get better at DevOps, security tools, architecture, etc. The mix of all of these made homelabbing a good new hobby. I could’ve bought a simple, cheap VPS. However, on these you can’t host your own SIEM, IDS, or IPS, and you can’t be sure it will exist when the AI revolution/takeover comes, so I wanted something that I could take with me to a farm off-grid and be able to have all my apps/tools. A toxic relationship with tech, I guess.

The Wishlist

I started with a simple thing:

• Nextcloud, which is a bunch of services. Think Microsoft 365, but open-source.
• EVE-NG for spawning network labs.
• A Kali Linux machine to sound scary.
• A small local AI model for the “Good Mornings”, “Keep going honey”, and other such messages.
• A blog maybe? Some custom apps.

The Hardware

So, shopping time. I started checking the hardware I would get. I thought of a Raspberry Pi, but I need something that will grow with me. A Raspberry Pi is good for small scripts or apps, but not what I had in mind. So I found out about mini PCs, basically a PC that’s small. After detailed research (read: a few Reddit posts), I decided on the GMKtec K8 Plus: good hardware, nothing fancy, but good enough for most of my needs. And it has Oculink support, which means I can add an eGPU later when my AI model (aka gf) needs an upgrade and I have the money for it.

The Plan

I then started with the plan: Ubuntu Server, then Nextcloud, the local model, etc. But then, I am not sure if I have a thing for complicating my life and taking the hard route, or if it was actually justified, but I moved from a simple server to a virtualization operating system. Let’s say I want to spawn a Windows machine to test something, or I want to test myself and see if I can get past installing Arch. Doing that on an Ubuntu server wouldn’t be that easy. So, a hypervisor for the host operating system. VMware ESXi, I remember that one from one of my internships, but VMware killed their free version and I want to be relying more on FOSS. Proxmox is a good open-source alternative. Now that this is solved, the network diagram.

I will have some hacking tools, and then my personal data. These should be isolated from each other. Then I will have public-facing apps, and those should be in a DMZ. I logged into the router admin page and it doesn’t have VLAN support. I can buy a new one, or I can host my own firewall and do the isolation using that (see, it is the thing I told you about: complicating shit. Maybe it is part of the hobby rules or something.)

Then that night it went something like this:

• I will have multiple public apps, and will probably need a reverse proxy. Enter Caddy.
• I would love to have analytics without giving the data to Google. Enter Matomo.
• Some monitoring and visibility. Enter Prometheus & Grafana.
• Some security, source code scanning. Enter SonarQube.
• …

I ended up with this architecture:

A beauty that lasted less than 24 hours. But that’s a story for v1.0.