Infrastructure / networking / local AI

Home Lab

I run more hardware at home than I strictly need. This is the current setup and what I'm doing with it.

Compute & services

AI / ML

I use the $20 Ollama Cloud plan and the $100 ChatGPT Pro-Lite plan for cloud models. Which one I use depends on what I'm testing and where the model runs best.

For local inference, I run Qwen3.8 Flash on the Ryzen AI Max node. Its 128GB of unified memory can load large models that won't fit on a normal GPU. The RTX 4070 Ti handles smaller dense models.

Networking

I have a 3 Gbps symmetric fiber connection going into a 10 GbE SFP+ port on the Dream Machine Pro. The main access point also connects over 10 GbE SFP+.

IoT and smart home devices live on their own VLAN and WiFi network. They can't reach the main LAN. The only exception is a firewall rule that lets Home Assistant poll devices; everything else can only reach the internet.

The Synology NAS and Home Assistant both run Tailscale as subnet routers. I can reach internal services and control the house remotely without opening extra ports on the firewall. It works well enough that I mostly forget it's there, which is exactly what I want.

Ports 80 and 443 forward into the k3s cluster. Traefik handles ingress, and cert-manager gets the Let's Encrypt certificates.