Ultron: Local AI, Connected
The full blueprint — turn a local Mac AI into an always-on assistant with a VPS, nginx, Tailscale, Hermes, and n8n.
Services
Key Results
$0
Monthly Compute Cost
0 KB
Data Leaving Home
The Challenge
A local AI sitting on your desk is a cool experiment until you close the lid. Discord bots go silent at 2 AM, cron pipelines miss their window, and the whole stack goes dark the moment you leave the house. Ultron exists to fix exactly that.
Strategy & Vision
Split brain and body. Keep DeepSeek inference on the Mac's unified memory — fast, silent, and free — then stand an always-on VPS in front as the public door, bridging the two over a private Tailscale link that never touches the public internet.
Design Approach
nginx + a free Let's Encrypt cert in front of everything, Hermes as the agent layer that gives real tool access, and n8n as the visual automation brain that runs the workflows making Ultron useful every single day.
Engineering & Implementation
Headless Ubuntu VPS behind a single nginx reverse proxy; the Mac serves the REST API reachable only over Tailscale. No cloud dependency, no API keys, no rate limits.
The Outcome
An AI that sorts your email, drafts your messages, researches topics, posts news to Discord, and tracks your budget — fully local, zero subscription fees, zero data leaving the house.
Building Your Own Ultron
Here’s the short version of what setting Ultron up actually looks like — and what to expect at each step. No magic, just a handful of moves.
- Pick the brain. Grab an Apple Silicon Mac — the more unified memory, the bigger the model you can run. This is where DeepSeek lives and runs inference locally, silently, for $0/month.
- Rent a front door. Provision a cheap, headless Ubuntu VPS that stays on 24/7. This is what listens for webhooks and keeps your assistant alive while your laptop is closed.
- Wire the private link. Install Tailscale on both machines. The VPS now reaches the Mac over a
100.xprivate IP, never exposed to the public internet. - Stand up nginx. Put nginx in front of everything with a free Let’s Encrypt SSL cert. It becomes the single door everything comes through.
- Give it agency. Install Hermes — the agent layer that lets Ultron read files, search the web, run code, and send Discord messages.
- Add the automation brain. Drop n8n into Docker behind the same nginx proxy and start connecting workflows.
What to expect: the first working loop usually takes an afternoon. The hard part isn’t the tools — each is a one-command install — it’s learning how they talk to each other. The scripts below remove the plumbing so you can focus on the fun part: deciding what your assistant does.
Download the Blueprint
Everything needed to reproduce this setup is packed below. Unzip, follow the README, and you’ll have your own Ultron standing up from scratch.
Download Ultron Setup (ZIP)
From token anxiety to a full AI stack that runs your inbox, your news, and your projects. Anyone can build this.Jeremy Lloyd Creator