Every developer today relies on LLMs. Whether it's drafting boilerplate, refactoring complex logic, or debugging edge cases on ChatGPT, Google Gemini, or Claude, queries are sent dozens of times a day.
Yet the thermodynamic and compute realities of these queries remain invisible. Matrix multiplications run on energy-dense hardware, requiring massive power draw and water-cooling towers at scale.
I wanted to make this resource footprint visible without adding friction or guilt-tripping. So I built EcoPrompt — an ultra-lightweight, 100% client-side Chrome extension (Manifest V3) that calculates the environmental and financial cost of your prompts in real time.
🌿 What EcoPrompt Measures
Whenever you submit a prompt on ChatGPT, Claude, or Gemini, an unobtrusive toast notification computes:
- 🍃 Carbon Footprint (CO₂e): In grams and milligrams, with real-world equivalents (e.g. smartphone recharges).
- 💧 Water Consumption (mL): Factoring on-site data center cooling evaporation and grid-level water draw.
- ⚡ Active Energy (Wh): Real-time compute draw based on model architectures and Power Usage Effectiveness (PUE).
- 💶 API Cost: Public market token cost benchmarks.
🛠️ Architecture & Strict Constraints
Building for production while respecting developer workflows led to three zero-compromise rules:
Zero Telemetry & 100% Client-Side Privacy:
No prompt text is ever captured, saved, or routed through an external server. The extension strictly measures character length in-memory to calibrate KV-cache compute multipliers, then purges the string immediately. Everything runs insidechrome.storage.local.Vanilla JS & Native SVG (Bundle < 150 KB):
No bloated charting libraries (no Chart.js, no D3, no npm baggage). The integrated analytics dashboard dynamically renders timelines (Day, Week, Month, Year) using a custom lightweight SVG generator.Custom Daily Budgets & Batching Prompts:
Users can configure soft daily thresholds (prompts, CO₂, water, or energy). A subtle counter notifies you upon reaching your budget, promoting mindful prompt engineering and query batching.
🚀 Try It on the Chrome Web Store
The extension is free and available directly on the Chrome Web Store:
👉 Install EcoPrompt on the Chrome Web Store
I’d love to hear your thoughts, UI feedback, or suggestions for additional AI platforms to support!
Top comments (2)
Thanks for reading! 👋
I'm curious to hear your perspective: how mindful are you currently of the compute, water, or energy cost of your day-to-day AI queries?
Also, are there specific platforms or local models (like Mistral, DeepSeek, or Ollama) you'd like to see supported next?
Drop your thoughts below, and if you test out the extension, I’d love to hear your feedback on the toast UI!
👉 Install EcoPrompt on the Chrome Web Store
The main structural challenge with modeling inference costs at the client boundary is the asymmetry between prefill throughput and decode latency. In standard chat interfaces, the user sees a single round-trip, but the physical hardware experiences two completely different operating modes. The prefill phase saturates tensor cores processing input tokens in parallel, whereas autoregressive decoding is bound by memory bandwidth and KV-cache residency per active batch slot.
When developers talk about model efficiency, they often look at aggregate token pricing from API rate cards. That pricing smooths out the reality that maintaining high-concurrency batch slots on water-cooled clusters carries fixed thermal and electrical overhead whether the generation is dense reasoning or short boilerplate. Making those physical constraints visible in the local workspace creates a much clearer mental model for why prompt batching and context pruning actually matter.