What Is a Homelab? (And Why You're About to Fall Down a Wonderful Rabbit Hole)
Your own private cloud — and a wonderful rabbit hole.
The logbook
129 entries on homelab infrastructure, AI, monitoring, and more — recorded from a live cluster, not researched from other people's posts.
LAST ENTRY: OCT 15, 2026 · STATUS: RECORDING
New to homelabbing? Start here — hardware first, then build up the stack.
Your own private cloud — and a wonderful rabbit hole.
Bare metal to backed-up cluster — the whole road, in order.
What actually matters in homelab hardware — and what's just shiny.
Ten services on one used mini PC — with receipts.
1GbE vs 2.5GbE, managed vs not — what your lab really needs.
One USB drive that boots every ISO you drop on it.
USB in, first login out — Proxmox installed without surprises.
Keys in, passwords out, and a root login nobody can brute-force.
New to homelabbing? Start here — hardware first, then build up the stack.
Your first container, and the wizard choices that matter.
First VM, from ISO upload to first login, no step skipped.
Your first stack, and every Compose word finally explained.
Quorum and Corosync demystified, then three nodes join hands.
Network storage in plain English, and which kind fits you.
Samba for the laptops, NFS for the servers, one source of files.
HAOS gets its own VM — the supported way that never fights you.
Most labs have one box quietly carrying everything. Mine did too.
Virtualisation, containers, storage, and cluster management.
local-lvm, ZFS, NFS — what each is actually for.
The morning my root disk hit 99%, and the traps behind it.
Kernel-sharing containers or full VMs — pick with confidence.
The installer says 'no drives found' — here's the driver it's missing.
Clone a ready VM in thirty seconds; never touch the wizard again.
Why the privileged-LXC shortcut bites, and the VM that doesn't.
Virtualisation, containers, storage, and cluster management.
Ads blocked for every device, DNS answered down the hall.
Three layers of firewall, and where each rule belongs.
Backups that dedupe, verify, and run while you sleep.
I timed every step of a real restore so you can plan with numbers.
Success mail is noise. Failure mail is sacred. One command splits them.
Four office castoffs become a production-grade cluster.
Virtualisation, containers, storage, and cluster management.
Your running cluster becomes code without a single restart.
Import until the plan says zero — nothing restarts, nothing drifts.
Ansible adopts the cluster you built by hand — without touching it.
Born from code, configured by code, destroyed on schedule.
Lose every node and walk back in — the restore order that works.
The kill that names itself, the cluster that freezes, the door that won't open.
Virtualisation, containers, storage, and cluster management.
Your passwords, served by Bitwarden apps, hosted by you.
A real wiki with a real database, in half an hour.
Let your nodes route around a dead link — no static routes, no hand-editing.
Proxmox Backup Server 4.2 speaks S3 now — point it at your own MinIO.
Add a different CPU to your cluster? Here's how VMs still live-migrate.
Hypervisor, NAS, or all-in-one — pick by what your lab is for.
Metrics, dashboards, and alerting with Prometheus and Grafana.
The power fails; every machine parks itself before the battery dies.
Monitoring you can set up before lunch and trust by dinner.
24 hours of red monitors while everything was actually up.
The dashboards everyone screenshots, finally explained.
Pull-based metrics and PromQL, minus the jargon.
From no metrics to full-cluster dashboards in one build.
Metrics, dashboards, and alerting with Prometheus and Grafana.
Node metrics plus cluster metrics, meeting in one dashboard.
A simulator that answers: if a node dies tonight, do we fit?
Everything used to email me. Now one thing does — at 8:30, with buttons.
Real screenshots, fake lab — publishing without leaking a thing.
Every token your AI tools burn, graphed next to your CPU temps.
Stop checking that release page. Let a watcher ping you when it changes.
Metrics, dashboards, and alerting with Prometheus and Grafana.
A disk warns you before it dies. Trend SMART and you get weeks of notice.
Three Ollama boxes, one smart endpoint that routes, fails over, and reports.
Remote access, VLANs, switches, and secure tunnels.
Your smart TV has no business seeing your NAS.
The whole lab reachable from anywhere, zero ports forwarded.
Coffee-shop Wi-Fi, but every packet exits through your lab.
Every light is green, every ping times out — meet the four usual suspects.
The router reached the NAS. The container behind it timed out.
Five devices, one router, one port 41641. Only one of them went direct.
Remote access, VLANs, switches, and secure tunnels.
Every obvious suspect was innocent. The culprit: 208 KB.
The reverse proxy explained with doors, not diagrams.
Green padlocks on every service, renewed while you sleep.
Encrypted DNS out of the box, or the biggest blocklist community?
GUI, labels, or one file — pick your reverse proxy.
The fast protocol, the effortless mesh, or the old reliable.
Local LLMs, agents, and self-hosted services.
One VM gets the card; the host gives up control.
Deployment is day one. This is every day after.
Stop inventing a new password for every service you host.
From empty Docker host to single sign-on in about half an hour.
If my laptop gets encrypted, it can't reach back and wipe the backups.
Your NAS protects everything but the apps running on it.
Local LLMs, agents, and self-hosted services.
Google Photos, replaced by a container you own.
Stop bookmarking IPs. One page knows where everything lives.
Three CPU-only nodes serving local LLMs behind one endpoint.
Seven models, no GPU, three runs each, and the newest isn't the fastest.
Local LLMs, agents, and self-hosted services.
A bookmark folder is where links go to die. Give it a local-AI librarian.
Stop hunting through folders. Search the text inside every scan and PDF.
You've fixed this problem before. You just can't find where.
Notes with version history, grep, and no vendor in sight.
Local LLMs, agents, and self-hosted services.
Many readers, one truth — sharing a vault without split-brain.
Auto-sync will eat your notes. Lock, validate, then commit.
The vault stays canonical; Wiki.js is just its pretty face.
Semantic search over your notes, no cloud, no API key.
Local LLMs, agents, and self-hosted services.
Ten ways I broke my second brain so yours stays whole.
Your wiki answers questions now — and cites where it read them.
You told it what you want. It spends all day making that true.
One command gets you a real cluster. Deciding to is the hard part.
Media servers, streaming, and the *arr automation stack.
Pay for polish or own the stack — the honest comparison.
Plex claimed, libraries built, running where the media lives.
Jellyfin running where your files already live.
GPU-accelerated 4K, no Plex Pass, no melting CPU.
Name the files right and Jellyfin matches every poster on the first scan.
Why Plex stutters on weak CPUs — and the setting that saves it.
Media servers, streaming, and the *arr automation stack.
Name the files right and Plex never guesses wrong again.
Full-speed Plex from anywhere, and not one port opened.
Sonarr, Radarr, Prowlarr — who does what, in plain English.
The full *arr stack standing up in one compose file.
Add a title to a list; the lab does everything else.
The one *arr add-on almost everyone ends up wanting.
Media servers, streaming, and the *arr automation stack.
Follow an artist once; new albums just appear.
The family browses and requests; the stack quietly delivers.
Rotated keys, zero errors, days of silence — then the fix.
Four days of zero bytes, and an alert that only ever spoke once.
Full builds and fun experiments — gaming, apps, and tools.
Fabric, mods, and JVM flags in an unprivileged container.
A native app for my cluster, because the browser tab bothered me.