📖 Engineering & On-Device AI Documentation

MooduTools Engineering & On-Device AI Docs

Technical deep-dives into how MooduTools runs a full generative AI product with zero server infrastructure, absolute on-device privacy, and an in-browser WebGPU LLM (WebLLM) — turning typed emotions into living HTML/CSS art.

💲

Zero-Server Cost

Static hosting on Cloudflare Pages plus a local LLM means the operational cost of one user scales to the entire planet — no inference servers, no GPU fleets, no usage bills.

🔒

100% On-Device Privacy

Mood text is authored, embedded, and answered entirely inside the browser. Zero egress means your emotional data never transits a network, never touches a datacenter.

📱

In-Browser WebGPU

A lightweight LLM (WebLLM) is loaded over a CDN and runs on your device's GPU. No backend, no sign-up, no Chrome flags — and it works in any WebGPU-capable browser.

Published Technical Deep-Dives

In-depth, engineer-authored documentation of the systems behind MooduTools.

How the Magic Happens

Mood inYou type a feeling; it is tokenized on-device and pinned to a strict system prompt.
Local inferenceA lightweight WebGPU LLM runs on your device and returns a raw HTML/CSS fragment only.
SanitizeA pass removes scripts and non-renderable prose before anything touches the DOM.
AnimateThe fragment is injected; CSS keyframes bring the art to life entirely in the browser.

Every document focuses on reproducible engineering: architecture, privacy guarantees, and piping. No ads, no trackers — just the technology.