I've been running a self-hosted AI assistant 24/7 for the past month. Here's what works, what doesn't, and what surprised me.
The Setup
Hardware: NVIDIA Jetson Orin Nano Super (67 TOPS, 8GB unified memory, 512GB NVMe) Software: OpenClaw (source available) + Ubuntu Power draw: 20W average — my desk lamp uses more. Cost: €549 one-time (I use ClawBox, the pre-built version)
What Runs Locally (No Internet)
Voice Processing
- Whisper — speech-to-text, 90+ languages, runs entirely on-device
- Kokoro — text-to-speech, natural sounding, also fully local
- My voice data never leaves the box. Period.
Local LLMs
- Llama 3.1 8B: ~15 tok/s — good for quick tasks, conversations
- CodeLlama 7B: decent for code snippets
- LLaVA 7B: vision model, can describe images
- Hermes 3 8B: good for structured/agentic tasks
The Reality Check
8GB unified memory = 7-8B parameter models max. For a daily assistant, this covers maybe 60% of what I need. The other 40%? Cloud APIs.
The Hybrid Approach (This Is the Real Insight)
Here's what I actually do:
| Task | Where It Runs | Why |
|---|---|---|
| Voice commands | Local (Whisper + Kokoro) | Privacy — my voice stays home |
| Smart home control | Local (Llama 8B) | Simple commands, no internet needed |
| Quick questions | Local (Llama 8B) | Fast enough, saves API cost |
| Email drafting | Cloud (Claude/GPT) | Needs better reasoning |
| Browser automation | Cloud (Claude/GPT) | Complex multi-step tasks |
| Code review | Cloud (Claude/GPT) | Larger context window needed |
| Document analysis | Cloud (Claude/GPT) | Often exceeds local model capability |
The key insight: privacy where it matters most (voice, personal data processing) + capability where it matters most (complex reasoning).
The ChatGPT Plus Trick (v2.0.0-8)
This was the game-changer for me. OpenClaw now supports ChatGPT Plus/Pro OAuth. Instead of:
- Paying $20/month for ChatGPT Plus (for the web UI)
- Paying separately for API access (for your self-hosted setup)
You just sign in with your existing OpenAI account. Your ClawBox uses your ChatGPT Plus subscription directly. No API keys, no double billing.
Same works for Claude and Gemini with their respective auth methods.
My Daily Workflow
6:00 AM — ClawBox checks email, summarizes what's important, drafts replies Throughout day — Voice commands for quick lookups, timers, smart home Work hours — "Hey ClawBox, review this PR" / "Summarize this PDF" / "Draft a response to..." Evening — "What happened in my inbox today?" / "Set thermostat to 20°" Overnight — Monitors prices on 3 products, checks for important emails
Total hands-on time per day: ~5 minutes of talking to it. It handles the rest.
The Honest Numbers
Monthly Cost
| Item | Cost |
|---|---|
| Electricity (20W × 24/7) | ~€0.80 |
| Cloud API usage (moderate) | ~€8-15 |
| ChatGPT Plus (if using OAuth) | $20 (already paying) |
| Total incremental cost | €0.80-15 |
What I Replaced
| Service | Was Paying | Now |
|---|---|---|
| ChatGPT Plus | $20/month | Same (but now powers ClawBox too) |
| Zapier automations | $20/month | €0 (ClawBox handles it) |
| Todoist + calendar AI | $5/month | €0 |
| Smart speaker (Alexa) | Free but privacy 🙄 | Local voice, no cloud |
Break-Even
Hardware cost (€549) ÷ monthly savings (~€25) = ~22 months. But honestly, I'd pay for the privacy alone.
What I'd Do Differently
- Start with cloud APIs only, then gradually move tasks local as you get comfortable
- Ethernet > WiFi for 24/7 reliability
- Don't try to run everything local — the hybrid approach is genuinely better than either extreme
- Join the community Discord — other users share automation recipes that save hours of figuring things out
Who This Is NOT For
- If you just want ChatGPT in a browser → stick with chatgpt.com
- If you need 70B+ models locally → you need a GPU server, not a Jetson
- If you don't want any cloud AI → it works, but you're limited to 8B models
- If you want gaming/rendering → wrong hardware entirely
Who This IS For
- Privacy-conscious professionals who want AI without sending data to 5 different clouds
- Homelabbers who want a silent, low-power AI addition
- Anyone tired of managing 4 different AI subscriptions
- People who want AI that works FOR them 24/7, not just when they open an app
Links
- ClawBox (pre-built hardware): openclawhardware.dev
- OpenClaw (source-available software): github.com/openclaw/openclaw
- Community Discord: discord.gg/FbKmnxYnpq
- Setup guide: openclawhardware.dev/setup-guide
This isn't sponsored content — I built ClawBox. But the analysis is honest. If you have questions about the setup, drop them in the comments or join our Discord.