I setup a home lab in my garage this year. It hosts a bunch of fun services like video streaming, local LLM inference, and web hosting. In fact, the website you are reading right now is being served from a web server in garage. I really enjoy setting these up since I like knowing how computer software works and setting up Linux systems is not something I get to do as a software engineer.

Here is the current state of things. There are two general-purpose machines running proxmox and a beefier one with a GPU for LLM inference.

Hardware

Here are the basic specs for each of the machines. For non-LLM use buying used Dell Optiplexes off Ebay is an amazing value. For the price of the inference server (which is quite modest by inference machine standards) I could have purchased 6 Dell Optiplexes!

Proxmox1:

• Operating System: Proxmox
• Machine: Dell Optiplex 7070 Micro
• Specs: 16GB RAM, 512GB NVMe SSD, Intel i7
• Price: $230

Proxmox2:

• Operating System: Proxmox
• Machine: Dell Optiplex 7080 SFF
• Specs: 32GB RAM, 256 NVMe SSD, Intel i5
• Price: $450 ($250 for machine, $200 for 32GB RAM upgrade)

Inference:

• Operating System: Ubuntu Server
• Specs: 32GB RAM, 1 TB M.2 NVMe SSD, Ryzen 5
• Graphics: NVIDIA RTX 5060 Ti 16GB
• Price: $1500

All the machines are connected to power via a 1500VA/900W UPS ($190). Networking is an unmanaged Gigabit switch ($20) and Google Nest WiFi router. Running servers off WiFi isn’t ideal but I’ve found it more than stable enough for my use.

Services

For Proxmox machines I try to keep one-service-per-container to keep things simple. Take a look at this screenshot of my Proxmox2 machine; each container is just named by the service it runs and you can easily monitor hardware usage.

Here’s a quick run-down of the services I run sorted by most-used to least-used with a brief description of why I use each:

• Pi-hole: DNS resolver. I use it to block Reddit and Hacker News mainly.
• Caddy: Web server. I use it to host this website!
• Anki: Flash card server.
• Pi-agent: Agentic coding harness.
• Hermes: AI assistant.
• Open WebUI: Chat harness.
• Jellyfin: Video streaming server.
• Radarr / Sonarr / Prowlarr: Automation for downloading movies & shows.

For the inference machine I run just one service:

• Ollama: Runs local LLM inference.

For accessing these services outside my local network I use Tailscale. It’s free and easy to setup.

Future

I’m reasonably happy with the current state of things but here’s what I’d like to add in the future:

1. NAS server: For video storage, backups.
2. Beefier inference machine: High VRAM GPUs are VERY expensive right now. I’d love to pickup an RTX 5090 32GB but $5K is pricey for hobby use.