The homelab
The self-hosted server and GPU machines behind everything else, from file sync and media to local AI inference. Most of this blog documents it.
- Main server
- Linux Mint · i5-9400F · RX 580 · Docker
- Front door
- Nginx Proxy Manager with Let’s Encrypt
- Network
- Tailscale mesh between all machines
- AI nodes
- llama.cpp inference on AMD and NVIDIA GPUs
This is the foundation: one always-on server plus a small fleet of GPU machines, all joined over Tailscale.
What runs on the main server
Everything runs in Docker behind Nginx Proxy Manager.
- Nextcloud for files, with a MariaDB database
- Jellyfin for media, with hardware transcoding on the RX 580
- Pi-hole for network-wide ad blocking and DNS
- A self-hosted mail server for my own domain
- Uptime Kuma, Netdata and Portainer for monitoring and management
- This website, built with Hugo and served by nginx
- The Tensor Games website and a few other small sites
The AI machines
Several desktops and a laptop run local models with llama.cpp across AMD and NVIDIA cards (including Tesla P100 and V100 server GPUs). They feed the voice assistant and the 3D and Blender tools.