Homelab Infrastructure Wiki

Network Overview

This wiki itself is served from Proxmox-1 Docker CT at http://192.168.1.154:8090 (Portainer stack homelab-wiki). Status badges above are checked live from your browser (see note at the bottom of the page about accuracy & assumptions).

Hardware Inventory

Proxmox-1 — Minisforum MS-01 (primary server)

The eGPU is single-tenant. VMs 100 (Windows-11), 102 (SteamOS), 107 (Ollama), and 111 (Mint) all claim the same physical GPU — only one can hold it at a time. Ollama currently has it while running; switching to any of the other three means shutting Ollama down first.

Proxmox-2 — Minisforum MS-01 (secondary server)

Proxmox-3 — Framework board in CoolerMaster case

Zaibatsui Tower — LincPlus LincStation N2 (Unraid NAS)

HP ProLiant MicroServer

Role unconfirmed — not covered by the original architecture document. Likely a potential remote backup target (see Network Overview); currently runs Cockpit and its own FileBrowser Quantum instance. See environment section.

Home Assistant

Pi-hole

Storage expansion (not currently connected)

SABRENT 4-bay hard drive dock (USB3 → Proxmox) — drives: 250 GB WD Black, 320 GB WD Black, 2× Samsung 1 TB. Intended for the OpenMediaVault NAS VM (also inactive).

Virtualization Map

How the Portainer environments map onto physical hosts. Proxmox-1 carries nearly all workloads; Proxmox-2 and Proxmox-3 are earmarked as future cluster nodes.

Proxmox-1 (MS-01) — VMs & CTs

Read live from the Proxmox API on 17 Jul 2026. VMIDs and status are authoritative here; several guests below appear nowhere else in this wiki.

The node itself reports 94 GB RAM. A SaraBackup guest appears in older notes but does not exist here — the Proxmox Backup Server at 192.168.1.219 (pbs.zaibatsui.co.uk) is its likely successor.

Proxmox-2 / Proxmox-3

PVE 8→9 upgrade complete, 23 Aug 2026. All three nodes run pve-manager/9.2.11 on kernel 7.0.14-12-pve.

These two nodes are deliberately cycled on and off, not left running 24/7. They're scale-out capacity — brought online as needed rather than kept up permanently, since there's ample headroom on Proxmox-1 for day-to-day workloads. Don't treat either being offline as a fault; check current status live rather than assuming from this page. As of 24 Aug 2026: Proxmox-2 is up (CPU/RAM verified live, see Hardware Inventory), Proxmox-3 is down.

VMIDs 120 (Docker-PX2) and 116 (Docker-PX3) are the Docker CTs on each node; 122 (CT-Template-2) and 118 (CT-Template-3) are stopped templates. Each node runs Docker with a Portainer agent only when up. Full guest inventory for these two nodes hasn't been re-audited since the upgrade — worth a follow-up pass.

Zaibatsui Tower (Unraid)

NAS shares + Docker (media stack, Immich, agents — see environment section) + one VM: Proxmox Backup Server.

Game Manager Server

IP 192.168.1.114 · Custom board-game collection / session manager, three isolated copies

Each instance is its own stack: static frontend + Python/FastAPI backend (uvicorn) + MongoDB. The primary instance's backend authenticates to the BoardGameGeek (BGG) API, confirming the app pulls game data from BGG for collection/session management.

Plus a Portainer agent for remote management.

GetAMP Game Server

CT 105 on Proxmox-1 · 192.168.2.10 · VLAN 2 (192.168.2.0/24, gateway .1) · panel at zaibatsui.co.uk (CubeCoders AMP) · reviewed 17 Jul 2026

This is the reason the homelab exists in its current form — a Debian 12 box (16 GB RAM allocated, i9-12900H) running eight AMP-managed game server instances. Only one runs at a time in practice.

Port forwards (UniFi, all forwarded to 192.168.2.10)

All 29 UniFi port-forward rules for this VLAN were cross-checked directly against AMP's own instance ports and match exactly.

The main and experimental Satisfactory instances share UniFi's Satisfactory Reliable Messaging forward (8888) — if both were ever started together, whichever isn't bound first would lose that port. In practice this hasn't mattered since only one instance runs at a time. Also note there is no separate SFTP forward for the main Satisfactory instance; only the experimental one (2228) is defined.

GetAMP gained a Portainer agent in July 2026 (environment “GetAMP”) and is now managed alongside the other Docker hosts. It is not covered by the original architecture document beyond its existence.

Home Assistant

Raspberry Pi 4 · 192.168.1.33 (wired, primary) · also on Wi-Fi at 192.168.1.120 · homeassistant.local:8123 via mDNS · Home Assistant OS 18.1 · Core 2026.7.4 · reviewed 9 Aug 2026

Runs as a dedicated HAOS appliance, not a Docker container — separate from the Proxmox/Portainer estate. 1,769 entities across 782 devices, spanning 8 areas: Bedroom, Entrance, Front Door Sensor, Kitchen, Limbo, Living Room, Loft, Office.

Access control

Automations: an Alexa alarm if the front door is left open, with a sensor-button press to silence it.

Flic buttons

A Flic Hub LR bridges six Flic buttons (flicktwist, office green, office red, bedroom, living room, computer) plus a separate Flic IR Blaster, all over the local MQTT broker below. The office pair does the most work — single/double/hold click combinations toggle the office lamp and strip, switch the screens and PC via IR, and control the main lights.

Radios

Integration ecosystem (77 config entries, 50 domains)

Not fully working: Sony Bravia TV and 3 Oral-B toothbrushes show as not_loaded; 2 generic LED-BLE strips are not_loaded; the EPSON XP-8600 printer integration is in setup_retry. None of these are critical, but worth a look if those devices matter day to day.

HP ProLiant MicroServer

LAN 192.168.1.70 · Tailscale 100.86.156.104 / hpproliantmicroserver.tailee040d.ts.net · Ubuntu 26.04 LTS · 2 vCPU, 7.3 GiB RAM · asset tag 5C7344P3MM · back online 17 Jul 2026 after being powered off since 2 Jun

Not covered by the original architecture document, so its intended role is unclear beyond file storage — worth deciding what this box is actually for.

n8n Server

IP 192.168.1.160 · Workflow automation · VM on Proxmox-1

Ollama Server

IP 192.168.1.249 · Local LLM inference · VM on Proxmox-1

Currently holds the passed-through RTX 4080 (Razer Core X eGPU via USB4). It's exclusive to one VM at a time — Windows-11 (100), SteamOS (102) and Mint (111) all claim the same PCI device, so this VM must be shut down before starting any of the others. See Hardware Inventory.

All three run as standalone containers — there are no stacks on this environment. Note there is no ollama container present, so the model runtime itself appears to run natively on the VM rather than in Docker; Open WebUI is only the front-end.

Proxmox Host Server

IP 192.168.1.218 · Docker host "Zaibatsui Server" on the Proxmox-1 hypervisor — dedicated reverse proxy

The ~101 GB RAM Portainer reports here reflects the Proxmox-1 host itself (96 GB MS-01), which also runs all the VMs listed in the Virtualization Map — not just NPM.

Proxmox-1 Docker CT

IP 192.168.1.154 · Main Docker host (LXC on Proxmox-1 / MS-01) — Portainer itself runs here

Stacks

Standalone containers

Penpot was deliberately removed (the architecture doc is stale); its old domain now points at a dead target — see Domains.

Sara-VM

IP 192.168.1.23 · VM on Proxmox-1 · Photo backup & publishing

Tipi Server

IP 192.168.1.66 · VM on Proxmox-1 · Runtipi self-hosted app-store platform

Zaibatsui App Server

IP 192.168.1.56 · Cluster of custom-built full-stack apps

Standard pattern: frontend + FastAPI backend + MongoDB (exceptions noted).

Zaibatsui Tower

IP 192.168.1.69 · Unraid NAS & media server — LincPlus LincStation N2, 10GbE

The *arr stack runs in host-networking mode; all media tools read from and write to Unraid's /mnt/user share.

*arr stack & downloaders

Media & other services

VMs

NAS shares

backups, config, domains, downloads, isos, media, PBS (NFS & SMB), SaraVM (NFS & SMB), scratch, system, Z-Drive (NFS & SMB).

Offline Environments

HP MicroServer, Proxmox-2 Docker CT, and Proxmox-3 Docker CT are currently offline — deliberate, not a fault. The MicroServer is staged as a future remote NAS, waiting on cheap drives, no timeline. The two Proxmox Docker CTs are pre-configured standby capacity held offline until Proxmox-1 runs out of headroom. Observa Server (CT 125) was reviewed and deleted on 22 Jul 2026 — a self-monitoring Prometheus/Grafana stack superseded by Beszel + Uptime Kuma. See Virtualization Map for the full cluster status.

Networking

Key IPs

Tailscale

Tailnet tailee040d.ts.net · account zaibatsui@hotmail.com (Free plan) · 10 machines · reviewed 17 Jul 2026

Remote access runs through a single subnet router — the tailscale VM on Proxmox-1 — advertising 192.168.1.0/24 (approved). No exit node is enabled and no routes are awaiting approval. MagicDNS is on at 100.100.100.100.

Two users share the tailnet: zaibatsui@hotmail.com owns nine machines; sara.gibson.1988@gmail.com owns sara-hp-laptop. Key expiry is disabled on every machine except anthonys-s24-ultra. Both tailscale and moltbot run a Proxmox kernel, so both are guests on Proxmox-1.

getAMP is not reachable over Tailscale. The subnet router advertises only 192.168.1.0/24, but the getAMP VM lives on 192.168.2.10 and has no Tailscale node of its own — so it falls outside the tunnel entirely. Add a second advertised route or install a client there if remote access is wanted.

Domains & External Endpoints

Read directly from Nginx Proxy Manager on 17 Jul 2026 — 41 proxy hosts, all Let's Encrypt and all public. NPM's "Online" means the proxy host is enabled, not that the upstream answers; four of these point at targets that are currently stopped.

Public routing terminates at Nginx Proxy Manager on the Proxmox-1 host (192.168.1.218), which holds the Let's Encrypt certificates.

Storage & Backup

Maintenance approach: regular Proxmox/VM updates; container updates via Watchtower; monitoring via Beszel (CT 110, agents on all 13 systems) and Uptime Kuma (agent-port checks on the same fleet). The Observa CT — a self-contained Prometheus/Grafana/Netdata stack that only monitored itself, and whose cAdvisor had never successfully started (invalid launch flag) — was reviewed and deleted 22 Jul 2026.

Proxmox Backup Server — audit (9 Aug 2026)

VM on Zaibatsui Tower · 192.168.1.219:8007 · pbs.zaibatsui.co.uk · datastore pbs-unraid: 12.7 TB total, 6.5 TB used, 6.3 TB free

Backups run weekly — last run Sun 9 Aug 2026, CTs from 00:00–00:10 and VMs from 05:00–05:04. Sixteen guests are covered:

Hygiene has improved since the last audit: Ollama (107) is fresh again after missing four runs, and FileBrowser-Quantum (117) is finally covered. A daily prune job now runs (keep-last 3, keep-weekly 4, keep-monthly 3, last run OK) and a weekly verify job runs Sundays at 22:00 (30-day outdated threshold, last run OK) — so backups are now both pruned to a sane retention policy and periodically integrity-checked, which they weren't before. No sync job exists, so there is still no second copy of the datastore outside Zaibatsui Tower itself.

External Devices

Services

Day-to-day apps, mirroring Homarr. Addresses verified against Portainer and live-probed; status re-checks every 60s (same caveats as Network Overview).

Apps

Network & infrastructure

Proxmox and n8n are deliberately not live-checked. Proxmox serves a self-signed certificate that fetch() rejects, and n8n's proxied endpoint returns a false-negative 503 to programmatic requests. Both are fine when browsed directly — a red badge here would be a lie.

*arr stack & downloaders

All on Zaibatsui Tower (Unraid) using host networking, so ports are bound directly to 192.168.1.69.

SABnzbd is on 8081, not its usual 8080 — qBittorrent already holds 8080 on this host's network, so they would otherwise collide. Verified by probe.

Household & media

The Ghost blog only answers on its configured domain, so a direct IP probe fails even though the site is up — same class of false negative as n8n.

Utilities (Proxmox-1)

Cross-cutting deployments

Known Issues / To-Do

Check items off as you handle them — state is saved in this browser (via localStorage), so it'll still be checked next time you load the page on this device.

• Proxmox AI Assistant OpenAI key turned out fine — it lives per-user in Mongo (user_api_keys), not the env var. The false "not configured" reading came from /health querying a nonexistent api_keys collection; fixed in server.py on 2026-09-03.
• Proxmox AI Assistant Resolved 2026-09-03: Mongo auth enforced (--auth, verified) and all plaintext SSH credentials removed. ssh_configs uses a private_key (no password stored); the two plaintext location_credentials records that held unencrypted passwords have been removed — collection is now empty. Verified clean.
• NetBox devices missing rack positions: Zaibatsui Tower, HP MicroServer, Proxmox-2/-3, OctoPi, U6 Mesh Secondary and the ASUS AP were added without a rack or location — place them in the elevations

Future plans (from architecture doc)

• Expand the Proxmox cluster with Proxmox-2 (and Proxmox-3)
• Automation and observability improvements (monitoring stack via Docker/Portainer — likely Observa)
• Roll out planned VLAN segmentation: LAN / IoT / Lab

About the Live Status Checks

The status badges and nav dots are checked directly from your browser, not from a background service — this page has no backend beyond the nginx container serving it. On load (and every 60 seconds after, or on demand via "Recheck now"), it attempts a lightweight connection to one known address per environment. A successful connection marks it ONLINE; a timeout or refused connection marks it DOWN. The sidebar's "Environments (online)" / "Environments (offline)" groups are rebuilt after every check, so an environment moves group automatically as its real status changes.

Caveats: checks run from your browser, so they only reflect reachability from this device on this network (VPN/off-LAN will differ, and mixed-content rules apply on HTTPS). Some services cannot be probed reliably — Portainer agent ports expect mutual-TLS, Proxmox serves a self-signed certificate, and n8n and Ghost answer only on their configured domains — so those show as NOT CHECKED rather than guessed.