The paradox of local AI: everything runs on the PC at home, but you’re not always home. The wrong answer is opening ports on your router (congratulations, your GPU is now the internet’s GPU). The right answer is Tailscale — a private network between your devices, free for personal use, no ports opened, working from any network including hotel Wi-Fi and cellular.
We run our entire production stack this way — image generation checked from a museum, renders queued from a coffee shop.
Setup (genuinely ~15 minutes)
- On the AI PC: install Tailscale, sign in (Google/GitHub/etc). Done — the machine gets a stable private address and name.
- On your phone/laptop: install the app, sign in with the same account. Now they see each other, encrypted end-to-end, wherever both happen to be.
- Open
http://<machine-name>:<port>from the remote device exactly as you would at home — your image generator’s web UI, your LLM chat, anything serving on the box.
That’s the whole core. No subscription, no config files, no dynamic-DNS rituals.
The upgrade that makes it shareable: tailscale serve
Raw ports work, but tailscale serve puts your service behind a clean HTTPS URL on your private network:
tailscale serve --bg 8000
Now https://machine-name.your-tailnet.ts.net reaches the service — valid certificate, no port numbers, works in any browser on your tailnet. (A related funnel command can expose a service to the public internet — think carefully before using that one; the entire point here is not doing that.)
Habits that keep it safe
- Tailnet-only by default. Everything stays private-network unless you have a specific reason.
- Your AI services still have no auth of their own — that’s fine because only your devices can reach them. This model breaks the moment you Funnel something; add real auth first.
- Battery note: the mobile app’s always-on VPN sips battery; toggle it when needed if that bothers you.
What this unlocks
Queue renders from anywhere, chat with your local LLM like it’s a cloud service (it isn’t — it’s yours), check a long job from the road, and let your AI assistant be genuinely useful even when you’re not at the desk it lives under. Local-first doesn’t mean location-locked.
