DEV Community

Yanko Alexandrov
Yanko Alexandrov

Posted on

I Built a 15W Always-On AI Assistant Box — Here's What I Learned

Last year I got tired of paying for cloud AI subscriptions while worrying about where my data was going. I wanted an AI assistant that was always on, didn't phone home, and could actually do things.

So I built ClawBox — a dedicated AI assistant box that sits on my desk, runs 24/7, and costs about €0.80/month in electricity.

Why Dedicated Hardware?

I started like most people: throwing money at cloud APIs. The results were great, but every conversation hit someone else's servers, costs compound, and you can't run background automations cheaply 24/7.

The Hardware

After testing several options, I settled on the NVIDIA Jetson Orin Nano Super as the core:

  • 67 TOPS AI performance
  • 8GB LPDDR5 unified memory
  • 512GB NVMe SSD
  • 20W power draw under normal AI workload
Power consumption comparison:
- ClawBox (Jetson Orin): 15-20W
- Mac Mini M4: 65W
- Gaming PC with RTX 4090: 450W
Enter fullscreen mode Exit fullscreen mode

The Software Stack

I open-sourced the software as OpenClaw. It includes Whisper STT (local), Kokoro TTS (local), Llama 3.1 8B at ~15 tok/s, browser automation, and multi-platform messaging.

TCO After 6 Months

Cost Amount
Hardware (one-time) €549
Electricity (6 months) ~€4.80
Total vs cloud VPS Saved ~€200+

A comparable GPU cloud VPS runs €40-80/month. Break-even is around 8-12 months.

Setup: 5 Minutes

  1. Plug in power + Ethernet
  2. Open browser → clawbox.local
  3. Follow configuration wizard
  4. Scan QR code for Telegram

No Docker, no terminal. Full setup guide here.

The full hardware comparison is on the website if you want the deep dive.

Questions? Drop them in the comments!

Top comments (0)