Tunneling Local LLMs to the Cloud with Tailscale

5/24/2025•Qorvinus
LLMsVPSDev ToolsAutomationn8n

Why I Did This

I’ve been running lightweight LLMs locally using Ollama — it's fast, simple, and a great fit for small workflows or testing. But local-only meant they were cut off from the rest of my stack, especially my VPS-hosted services. I didn’t want to expose ports or mess with reverse proxies just to bridge that gap.

So I used Tailscale to create a private mesh network between my local machine and my VPS.

Now, apps on my VPS — including my self-hosted n8n instance — can make secure HTTPS calls directly to the LLMs running on my dev machine. Handy for experimenting, automating, or even just skipping cloud costs when I don’t need GPT-4-level firepower.

How It Works

  • Ollama runs various LLMs on my local dev machine.
  • Tailscale connects that machine to my VPS via a secure private network.
  • n8n, hosted on my VPS, uses HTTP Request nodes to talk to the local LLMs.
  • Other web apps can hit the same endpoints, as long as they're inside the Tailscale network.

I didn't need to do any SSL configuration or domain setup — Tailscale handles secure transport with minimal fuss.

Tailscale made this whole setup super easy! I was connected and using my local LLMs in minutes. Tailscale offers a very generous free tier, which is what I'm using here.

Stack Rundown

  • Ollama – local LLM serving
  • Tailscale – mesh VPN for private networking
  • n8n – workflow automation on the VPS
  • HTTPS nodes – n8n’s built-in HTTP request functionality
  • Ubuntu VPS – hosting the cloud side of things

What I'd Improve or Tweak

  • Add a local reverse proxy (like Caddy) to standardize routes and allow friendly naming (e.g., http://llm.local.tailnet).
  • Maybe use Tailscale's Funnel feature to expose selected endpoints to the internet if I ever need that.
  • Look into monitoring usage and response times for better observability — right now it’s all vibes and cURL.