Personal AI operating system for a homelab
NEXUS OS
Not a chatbot. The brain of the house and the servers.
Overview
NEXUS OS by Nick Gaitan is my flagship personal project: an AI-native control centre for my whole homelab. It runs 24/7 in Docker on a Linux server (Debian 13) and shows, in one place and in real time, every server, container, the network, security, backups, the smart home, email and calendar. Claude is its brain: it answers in AI Chat and Voice using live data from the servers, and works as a sysadmin agent that watches them and reports to me on Telegram.
The challenge
Smart home on one side, servers on another, AI on a third, monitoring on a fourth — dozens of disconnected apps that never talk to each other. Most dashboards show demo or fake numbers, and AI assistants never see your actual infrastructure. I wanted a system that knows exactly what is running, what broke and what needs attention — without exposing anything to the internet.
The solution
The architecture came first — design documents and principles before code: modular, event-driven, API-first, secure by design, privacy-first. Then it was built: a FastAPI backend with PostgreSQL, Redis and MQTT; a Next.js dashboard that also installs as a phone app (PWA); all in Docker. The data is real end to end: Glances on every machine, a secure read-only Docker proxy, and a host collector that every 30 seconds reads SSH logs, disk health (SMART), Tailscale and backup history. AI runs through Claude with no API key in the code, while speech recognition and synthesis run locally (Whisper, Piper). Access is Tailscale-only.
Technologies
From the project
AI Chat
A real question, a real answer: Claude checks the server and reports that all 21 containers are running, which are healthy and which restarted recently.
Voice
Push-to-talk from desktop or iPhone. Speech becomes text locally with Whisper, Claude answers, and the reply is spoken locally with Piper.
Servers
Every machine in the homelab, live: OS, CPU, memory, load, temperatures, uptime and containers per host.
Outcome
- Real-time monitoring of the whole homelab — 3 machines, a NAS and 21 containers on one screen
- Security Center with SSH brute-force and port-scan detection, disk health and instant Telegram alerts
- Claude as an on-server sysadmin agent: morning and evening reports, fixing only low-risk issues on its own
- AI Chat and Voice grounded in live server data — speech-to-text and TTS local, answers from Claude
- Nightly backup monitoring (encrypted disk, OneDrive, Google Drive) with alerts on failure
- Proton Mail and Proton Calendar, Home Assistant and smart devices inside the same system
- Zero fake data: anything not connected is shown plainly as unavailable