n8n Now Runs My ENTIRE Homelab

NetworkChuck Technology Careers Career & Money

About this video

The creator builds ‘Terry,’ an n8n-based AI agent presented as a junior IT employee for a homelab, upgrading an earlier version that could only monitor websites. He hosts a self-hosted n8n instance in the cloud on Hostinger (recommending its KVM 2 plan) so Terry stays up if the homelab fails, and uses Twingate to reach back into the home network securely. Terry is an n8n AI agent node using, in the demo, ChatGPT 4.1 mini (later upgraded to full 4.1 for a harder fault) with simple memory, taught the way a human works: first an HTTP request tool to check a test website in a Docker container on port 8090, then troubleshooting. Because n8n has no native SSH tool, he converts an SSH execute-command node into a sub-workflow with a ‘command’ input, exposes it via the Call n8n Workflow tool, and first locks it to docker ps before letting the model choose commands such as docker inspect and docker logs. A schedule trigger runs Terry every five minutes, using a Set/Edit Fields node to supply the prompt and a fixed chat/session ID. Structured output (a boolean website-up flag plus a message) feeds an IF node so Telegram only alerts when the site is down. Changing the prompt lets Terry fix faults with docker start and re-verify, and for an unseen port conflict caused by a Python server on the same port, he adds human-in-the-loop: Terry must request Telegram approval (send-and-wait) before any system-changing command, with extra structured fields for needs-approval and commands requested, and a loop back into the agent with the session ID so context is retained. Finally he connects Terry via Twingate to real systems — UniFi’s API to list top bandwidth users, Proxmox via API/CLI to count running VMs, and Plex’s API to report active streams — and previews next steps: sub-agents, central documentation, and a help desk.

Key takeaways

Teach the agent like a new hire: give it the same checks a human would do (visit the site, docker ps, docker inspect, docker logs) in its system prompt, and expand permissions gradually
Work around n8n’s lack of an SSH tool by converting an SSH node into a sub-workflow with a command variable and calling it as an agent tool
A schedule trigger plus injected prompt and session ID turns a chat agent into an autonomous monitor — here checking every five minutes
Structured output plus IF/Switch logic prevents notification spam, alerting via Telegram only on failure, a fix, or an approval request
Human-in-the-loop approval via Telegram for system-changing commands, plus cloud hosting and Twingate access, are the guardrails that let the agent safely reach UniFi, Proxmox, Plex and other API/CLI systems

Episode details: published October 3, 2025 · 47:17 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are NetworkChuck’s own claims, shared for education — they are not typical or guaranteed.

← Previous: you need to learn MCP RIGHT NOW!! (Model Context Protocol) · Next: This AI Supercomputer can fit on your desk… →

All NetworkChuck videos · NetworkChuck profile · NetworkChuck on YouTube