I’ve been running five Windows machines for the AI setup. Workstations, edit rigs, an AI agent box. All powerful, none of them always on. That’s a gap.
An Orange Pi 5 fills it.
The hardware
RK3588S SoC, 8GB RAM, Ubuntu 22.04 ARM64. Small board, no fan, negligible power draw. Set up June 16, running continuously since.
What it handles
Wake-on-LAN. Any machine in the fleet can be woken remotely via the Pi. One command from my phone: ssh opi5 wake schnitt-rechts – and the 4080 machine starts up. No need to be home.
Fleet monitoring. Every 10 minutes it pings all hosts and logs the status. The Telegram bot sends alerts when something goes down or comes back, plus a morning summary at 08:00.
Ollama fallback. llama3.2:3b runs on the Pi. ~8 tokens per second – not fast, but available when every other machine is off. Enough for lightweight queries.
Pi-hole. DNS filtering for the whole network. Ads and trackers blocked on every device without touching individual settings.
Remote SSH via Cloudflare tunnel. ssh opi5-cf connects from anywhere without a VPN. Key-only authentication.
What actually changed
The workstations do the heavy lifting. The Pi makes sure I can always reach them, wake them, and have something running when they’re all off.
Small board, big difference in reliability.