# DontDoThat.click

> Break bad habits using AI that runs entirely in your browser. 100% private — your camera stream never leaves your computer.

## What is DontDoThat?

DontDoThat is a free, privacy-first web tool that helps people overcome bad habits in real-time. It uses your webcam and a CLIP vision model (running locally via WebAssembly) to detect behaviors like nail biting, slouching, skin picking, or any custom habit you want to break. When it catches you — it alerts you instantly with a sound and desktop notification.

The key differentiator: **zero data leaves your device**. The AI model downloads once, gets cached in your browser, and all inference happens on-device. You can disconnect from the internet entirely and it keeps working. There are no servers, no accounts, no tracking.

## Why it works

The hardest part of breaking a bad habit is noticing you're doing it. Most people bite their nails, slouch, or pick at their skin unconsciously — they only realize minutes or hours later. DontDoThat closes that awareness gap by providing instant feedback at the exact moment the behavior occurs.

This approach is grounded in habit reversal training (HRT), a well-established behavioral technique. The first step of HRT is awareness training — learning to notice the habit as it happens. DontDoThat automates that step.

## Who is it for?

- People trying to stop nail biting, skin picking, hair pulling, or other body-focused repetitive behaviors (BFRBs)
- Remote workers who want to improve their posture while working from home
- Anyone practicing mindfulness or self-awareness around unconscious behaviors
- People who have tried habit-tracking apps but found them too manual to maintain

## How it works (technical)

The site is a static HTML/CSS/JS application. It uses Transformers.js to run OpenAI's CLIP (ViT-Base-Patch16) model in the browser via WebAssembly. Every 3 seconds, it captures a frame from the webcam and performs zero-shot image classification against user-defined behavior labels. If any behavior exceeds the confidence threshold, it triggers an alert.

The entire application is three files totaling ~35KB hosted on AWS S3 behind CloudFront. The AI model (~90MB) is served from Hugging Face's CDN and cached by the browser after first load.

## How to verify the privacy claim

1. Open browser DevTools (F12) → Network tab
2. Enable the camera and let detection run
3. Observe: zero outgoing requests after the model loads
4. Alternatively: disconnect from the internet after loading — detection continues

## URL

https://dontdothat.click