Type-less

Type-less helps users craft shorter, clearer AI prompts that reduces tokens usage and cuts actual CO2 emissions on larger scale, making AI use more sustainable while tracking the environmental impact.

Comment

Have you ever thought that today’s AI use carries a real carbon cost?

Every prompt consumes electricity and if scaled to millions of users, that power turns into significant CO₂ emissions. Global data-centre demand is set to surge this decade largely due to AI, with the IEA projecting that data centres could more than double their electricity consumption by 2030–2035 in its base case. At the single-prompt level, recent measurements from Google estimate a median Gemini text prompt ≈ 0.24 Wh and ≈ 0.03 g CO₂esmall per user, but meaningful at scale. Everyone is excited about GPT-class tools, yet few stop to consider their hidden environmental footprint. We asked: What if we help people communicate more efficiently with AI and cut emissions? That’s how Type-less started. It’s about writing shorter, clearer prompts without losing meaning, so each interaction uses fewer tokens, less compute, less energy, and therefore less CO₂ emission. It’s a small change that can add up when many people do it.

What Type-less does?

Cleans up your prompt: removes polite fluff (“Could you please…”), filler words (“really”, “extremely”), and wordy phrases (“in order to” → “to”).

Keeps your meaning: we protect important parts like negations (“not”, “don’t”) and we don’t touch code/quotes/links.

Shows your impact: a dashboard turns tokens into energy (kWh) and CO₂ (kg) so anyone can understand the footprint of their usage.

Lets you simulate savings: slide a control to see how cutting X% tokens would reduce energy and CO₂ for a week or a month.

How it works?

Tidy the text (spacing, casing, punctuation). Cut politeness and filler (“Could you please explain…” → “Explain…”). Shorten wordy phrases with safe swaps. Passive → active when it’s clearly safe (“It is requested that…” → “Do…”). Split long sentences into short, clear tasks. If still long, keep only the key action and nouns. Meaning stays intact: we guard negations and skip edits inside code/quotes/links. The dashboard shows before/after tokens and converts savings into energy (kWh) and CO₂ (kg) using editable settings.

The idea in plain terms:

Tokens → Energy → CO₂. Fewer tokens = less work for the model = less energy = less CO₂. That’s it. No tricks, just transparent math with adjustable assumptions. (You can literally set your own energy per token and grid CO₂ per kWh to match your region/hardware.)

Design Choice: we did not use big AI models inside Type-less?

To avoid the paradox of “using more AI to reduce AI,” we replaced heavy model calls with lightweight NLP/ML algorithms: NLP (natural language processing): simple, rule-based rewriting and pattern matching for politeness, wordiness, and passive voice. Small ML (machine learning): tiny, offline classifiers and part-of-speech tagging to spot adverbs/intensifiers and passive patterns—no cloud models, no large GPUs.

Why this matters: it’s fast, private (runs locally), and low-energy which supports the sustainability goal rather than working against it.

Tracking progress (MongoDB):

In order to see their daily and monthly progress. We added a very simple tracker: Creates a user entry (name, email). Stores numbers only for each prompt: tokens before and after. Automatically keeps daily and monthly totals of tokens saved, energy, and CO₂. Lets the dashboard show your trend over time (no need to store your actual prompt text).

What inspired us?

We love building with AI, but we also want to use it responsibly. We realized sustainability isn’t only about giant data-center policies—it’s also about everyday choices. If we help people write shorter prompts without losing meaning, we can save time, money, and a little bit of carbon with every interaction.

What we learned?

Small changes add up. Trimming a few tokens per prompt becomes meaningful when repeated millions of times. When people see kWh and CO₂ right next to their usage, they naturally make better choices. Simple beats complicated. A clear, rule-based approach with good safeguards works well and is easy to trust. Language matters. Showing energy and CO₂ (not just “tokens”) helps non-technical teammates understand the impact.

What’s next?

Build Rewards system for CO₂ savings: points, badges, and team shout-outs for hitting monthly goals.

Create MVP browser extension (Chrome/Edge, MV3): make a content script hooks into ChatGPT/Claude/Gemini text boxes and runs on-device Typeless to auto-optimize on send, with a toolbar toggle and strict local processing—delivering instant token/CO₂ savings without changing the user’s workflow.

Smart model choice: Re-route simple tasks to smaller models, keep big ones for hard tasks.

Language packs: Develop more rules for more languages and writing styles.

Overall, Type-less uses lightweight NLP/ML to make prompts shorter and clearer, so your AI use needs fewer tokens, less energy, and produces less CO₂ with a simple dashboard that shows the impact.

Built With

Share this project:

Updates