homelab · rack elevation, roughly

the lab

It started as "I wanna run Plex for cheap." Now it's a UDM Pro, a Beelink mini PC running Proxmox, and an Unraid box I call the big rig, with around 50 containers between them.

It teaches me something new every week, usually at 2am when something decides to break. That's the point. It's a safe place to be wrong before production teaches me the same lesson.

PP-01 CAT6

UDM Pro

Gateway and DNS for the house.

PVE · Proxmox

A Beelink mini PC doing a lot of heavy lifting for its size.

lxc: RustDesk server, VPN router, Nginx Proxy Manager
vm: Home Assistant, two Hermes agents (the house + my fiancée), Docker with ~20 containers

PROXMOX VE

BEELINK MINI PC

VLANS: REBUILT 4X

> snapshot first_

Unraid · big rig

Runs the whole media stack, about 30 containers, and a VM for my work Hermes agent.

cpu: i7-11700K
gpu: RTX 3070 8GB
ram: 64GB
disk: ~10 drives, ~12TB usable after parity
case: Jonsbo N5

Media

Plex
Jellyfin
Sonarr
Radarr
SABnzbd
Overseerr
Dispatcharr
Kavita

The reason any of this exists. Lives on Unraid, where my pile of mismatched old drives can all pitch in.

AI + agents

Hermes Agent ×3
Honcho
Ollama
Qdrant
n8n
MCP servers

Hermes agents for the house, my fiancée, and work, with Honcho for their memory. Plus local models, a vector store, and n8n.

Home

Home Assistant
Zigbee2MQTT
Mosquitto
Frigate

Home automation, the Zigbee mesh, and the security cameras.

Network

Nginx Proxy Manager
wg-easy
VPN router
DDNS
RustDesk

The reverse proxy, remote access, and the DDNS that keeps the domain pointed home.

Game servers

Minecraft
Palworld

Spun up whenever the group gets back into a game.

Apps

Vaultwarden
Firefly III
Netdata
nalyd.dev

Passwords, finances, monitoring, and the website you're reading.

+ the long tail · ~50 containers total

one light per container. 29 are listed above; the rest are supporting cast.

network, sketched

               internet
                  │  ddns keeps the domain pointed home
                  ▼
       ┌─────────────────────┐
       │ udm pro             │◄── me, away from home
       │ gateway + dns       │    (wireguard via wg-easy)
       └──────────┬──────────┘
                  │
══════════════════╪═══════════ lan · vlans ════════════
                  │
       ┌──────────┴───────────────┐
       │                          │
┌──────┴──────────────┐  ┌────────┴────────────┐
│ proxmox · beelink   │  │ unraid · big rig    │
│ nginx proxy manager │  │ media stack         │
│ home assistant      │  │ ollama, frigate     │
│ hermes agents       │  │ game servers        │
│ ~20 containers      │  │ ~30 containers      │
└─────────────────────┘  └─────────────────────┘

Simplified, and not to scale. VLANs got rebuilt four times before they were right.

things i broke

Running your own infrastructure means things break at 2am. This is the running list.

when	what broke	how it got fixed
spring 2026	Updated a container that shipped a breaking change.	Rolled back to the snapshot I took that morning, fixed the update path, and was back online in about 10 minutes.
the proxmox era	The entire media stack, back when it all ran on Proxmox.	Restored it from old backups. Then moved media to Unraid, so my mismatched old drives could pool together without fighting mergerfs and FUSE.
ongoing	Local LLMs on consumer hardware.	Learned what Ollama and Qdrant actually run well on, and what needs real GPU compute.
a while back	Container orchestration.	Tried three different approaches before settling on the one that works.
a while back	The VLANs.	Rebuilt them four times before I got it right.

what it taught me

know the revert path

I ask "what's the revert path?" before touching anything prod-adjacent. Snapshot before you update. You'll thank yourself at 2am.

monitor before you need it

Netdata tells me when CPU spikes or a container restarts. Same rule at work: observability goes in before the incident, not after.

automate the toil

Manual processes fail at the worst time. Automated backups and scheduled jobs handle the recurring stuff.

know how traffic flows

VLANs, reverse proxies, DNS, DDNS. Understanding how traffic gets in and out of a system transfers everywhere.

Want to compare setups? Email me, or sign the guestbook with what yours runs.

The rack