MUGGS OF DREAMS

When most people use AI, the chat apps, the AI button inside an app, their request leaves their phone and travels to a data center: a warehouse full of thousands of specialized computers, owned by a large company. The AI runs there, and the answer gets sent back.

Those buildings use enormous amounts of electricity and water for cooling, and more are going up every year. And whatever you type is processed on someone else's machine, under someone else's rules, at a price someone else sets.

AI doesn't have to live in a warehouse. Some AI models are released openly: you can download their files and run them on an ordinary, if powerful, home computer. That's called running a model locally.

MuggsOfDreams runs its AI locally. The pages are built at home, and the art is made at home. By launch, no AI company's servers are involved at all, and nothing about your page depends on an AI company staying in business or keeping its prices the same.

The finished pages are hosted on Cloudflare, ordinary web hosting, not AI. That's what keeps them fast for everyone. The AI work is data-center free.

The main machine is a desktop with an NVIDIA RTX 4090 graphics card. Graphics cards are good at the kind of math AI needs, and the 4090 is one of the strongest you can put in a home computer.

On it runs Qwen, an open model family released for anyone to download and use. The version I use has 27 billion parameters, the learned settings that make up the model's knowledge. Small next to the giant models in data centers, but big enough to write clean, working web pages.

Its job: read your form, start from the shared MuggsOfDreams template, and write the code for your page. By launch, it offers a few options for each part of the page, and I pick the one that fits you best. Then the finished page is compared against an already-approved page, so every page feels like part of the same neighborhood without looking generic.

The second machine, with an NVIDIA RTX 5070 Ti, has one job: making images. I call it the art shop.

It runs ComfyUI, an open-source tool for image-generating models, and FLUX, an open image model. When the builder needs art for a page, it sends a request to the art shop over a cable that connects the two machines directly. The art shop makes the image and sends it back.

Its job: your page doubloon, a Mardi Gras-style coin that represents you, plus any supporting images. There's one firm rule: art that goes on a live page only comes from models whose licenses allow commercial use.

A model on its own just answers questions. To actually build something, it needs a workbench: a place where it can open files, write code, test it, and keep track of what's done. That's OpenCode.

OpenCode is a free, open-source AI coding agent. Unlike most coding tools, it isn't locked to one AI company. You can point it at any model you choose, including one running on your own machine. Here, it connects to Qwen on the 4090.

Its job: organize the build. It breaks each page into a to-do list, works through it one task at a time, and leaves notes for the next step, the way a careful builder works through a punch list.

I'm Sean, a computer science teacher in New Orleans. The machines do the typing, but I'm responsible for what goes live.

Every page is checked against the house rules before it ships. By launch, I'll be reading every form myself, choosing among the options the model offers, and keeping a record of those decisions. When something needs a human touch, it gets one. I'm not a graphic designer, and I'll tell you when it matters.

Every piece of this, from the coding model and the image model to the tool that runs the build, is open. Anyone can download them, look at how they work, and run them without asking permission or paying a monthly fee.

That's the point. MuggsOfDreams is a small proof that a neighborhood can build its own internet presence with tools it can actually own. No warehouse required.