Homelab

active

AI-focused distributed homelab in a mini server rack — 7+ devices connected via Tailscale mesh.

Ubuntu Server Docker Tailscale systemd Proxmox

A distributed homelab where each device has a dedicated role, connected via Tailscale mesh networking. Housed in a 10-inch mini server rack. Supports personal automation, AI agent infrastructure, and future robotics experimentation.

Devices

Device	Role	Status
MSI Cubi	Production n8n + Obsidian Sync (always-on, headless)	Deployed
X1 Carbon	AI orchestration hub, Zarq Board, agent services	Active
Legion 7i	AI compute (RTX 4090, Ollama), Pong’s home	Active
Synology NAS	Centralized storage and backups	Active
MacBook Neo	Mobile dev laptop	Active
Raspberry Pi 5	Future robotics gateway (1TB NVMe)	Pending
Beelink EQ14	Proxmox virtualization lab	Pending

Network Topology

AT&T Gateway (router, DHCP, Wi-Fi)
    |
    +-- Ubiquiti USW-Lite-8-PoE
    |       +-- Legion 7i (RTX 4090, 32GB DDR5)
    |       +-- Ubiquiti USW-Lite-8-PoE #2
    |               +-- MSI Cubi (32GB, 1TB NVMe)
    |               +-- X1 Carbon
    |               +-- Beelink, Pis, NAS
    |
    +-- Tailscale mesh for remote access everywhere

Services Running

Service	Host
n8n (automation)	MSI Cubi
Obsidian Headless Sync (2 vaults)	MSI Cubi
Zarq Board	X1 Carbon
Ping (Telegram bot)	X1 Carbon
Agent-PKM sync	X1 Carbon
Ollama (local LLM)	Legion
Pong (AI ops agent)	Legion WSL2
Hermes Gateway	Legion WSL2

AI Architecture

The homelab runs a distributed AI agent system:

Pong — Ops engineer on Legion WSL2. Infrastructure management, deployments, documentation.
Ping — Personal AI companion via Telegram. 65+ tools, PKM integration, voice messages.
Zarq — Task board agent on X1 Carbon. Coordinates work across all agents.
Agent-PKM — Knowledge base synced from Obsidian vault. Agents query it for context.

LLM inference uses a fallback chain: Anthropic → OpenRouter → Kimi → Groq → Ollama (local).

Build Journal

Started February 2026 with physical rack assembly and MSI Cubi upgrades (32GB RAM + 1TB NVMe). Got n8n deployed in Docker by March, Obsidian Headless Sync running for two vaults, and the full AI agent stack operational by April. Cable management is still a work in progress.