Live machine grid · mesh proof

Twenty-two machines. One mind.

A self-hosted home cluster in Independence, MO - no cloud bill, no platform team. Everything below runs on owned hardware and reports into one source of truth.

14/20Machines onlineself-hosted, observed, on-call

136k3s podsrunning now

7k3s nodesowned hardware

0cloud billevery watt is mine

The grid

Every node, its role, and whether it's awake.

fleet · independence, mo14 / 20 online

sam-primek3s control plane

sam-worker-01k3s worker

sam-worker-02k3s worker

sam-worker-03k3s worker

sam-worker-04k3s worker

gpu-01k3s · inference · Ollama

gpu-02k3s · local LLMs · MCP

edge-kcingress · Caddy

edge-pdxingress · NGINX

db-primeMySQL · primary

db-replicaMySQL · replica

cache-01Redis · queues

unas-proUNAS Pro · backups

ws-studiodev workstation

obs-01Prometheus · Grafana · Loki

ws-renderrender box

ws-lablab / staging

lap-fieldfield laptop

lap-traveltravel laptop

arch-coldcold archive

14/20online

136k3s pods

7k3s nodes

Workloads

136 pods, scheduled and observed.

k3s · workloads136 pods running

7k3s nodes

31namespaces

100%scheduled

tailscale · tailnetwireguard up

headscale

edge-kc

gpu-01

db-prime

unas-pro

ws-studio

obs-01

Durability

Nothing serious runs without a recovery plan.

UNAS Prowork backup · 3-2-1

synced

6.4 TBof 12 TB usable

Last snapshot14 min ago
Retention90 days
Offsite mirrornightly
RAIDZFS · mirror

Backup doctrine

Work data lands on a UNAS Pro running ZFS mirrors, snapshotted on a schedule and mirrored offsite every night. The rule is simple: three copies, two media, one offsite. If a workstation dies mid-render, the worst case is minutes, not days.

PrimaryUNAS Pro · ZFS mirror
Snapshotshourly, kept 90 days
Offsiteencrypted nightly sync
Restore drillmonthly, automated