About the Project:

What Inspired Us

When we think about consumerism, we picture physical things like fast fashion. But a new kind of consumption is happening right now: Digital Materialism. Because chat boxes feel infinite, we treat tokens like they are disposable. Research from UC Riverside ("Making AI Less Thirsty") shows that every prompt costs precious, drinking-quality water and grid electricity to cool servers. Like fast fashion, it feels free, but our planet pays the price.

The Problem

A 2026 study by SkillReducer analyzed 55,000 prompts and found that over 60% of the text is useless fluff. Stripping out this noise cuts text in half and actually improves task quality. Microsoft research proved that a lightweight model layer can compress prompts up to 20 times with less than a 2% drop in quality. Inspired by this, we built a two-part framework: a CLI tool for developers and a web extension for everyday web users.

How We Built It

The Architecture: An asynchronous script scrapes the last 10 conversation nodes directly from the chatbot's DOM and saves them to local storage.
The Pipeline: A background service worker detects the data update and automatically launches a sleek single-page dashboard.
The Telemetry: The dashboard calculates text patterns locally, maps prompt history, and uses Chart.js to dynamically chart the user's hidden token and water footprint.

Challenges & What We Learned

Our biggest hurdle was answering tough mentor feedback: Aren’t you wasting more tokens by using AI to audit another AI? We resolved this with two choices:

Model Optimization: Our pipeline routes diagnostics to an ultra-lightweight model tier, which is 94% cheaper and uses a fraction of the computing energy.
Behavioral Training: The audit functions as a teaching tool. By exposing users to their hidden water metrics and habits (like redundant loops), we train them to permanently reduce prompt sizes over time.

Our roadmap takes this further, aiming to run the compression engine entirely client-side via a tiny local LLM using WebGPU for a zero-cloud footprint.