Footprint Ai

Every AI query uses real energy, real water, and produces real CO₂. Footprint AI makes that invisible cost visible.

Footprint AI estimates the energy consumed by your AI usage across platforms including Claude and ChatGPT. Power consumption, GPU compute, datacenter water consumption for cooling, and carbon dioxide equivalent emissions — then translates them into real-world comparisons you can feel: seconds of shower water, number of cell phone charges, and distance driving.

HOW IT WORKS
Footprint AI detects when you send a query to an AI platform and measures the response duration. Longer, more complex responses use more compute and energy. The extension scales its estimates accordingly, using published research from Epoch AI, the IEA, and peer-reviewed studies on AI energy consumption.

WHAT IT ESTIMATES
• Energy consumption (Wh / kWh)
• Datacenter cooling water (mL / L)
• CO₂ equivalent emissions (mg / g / kg)
• Real-world comparisons: shower time, cell phone charges, driving distance
• GPU compute time (GPU-ms / sec / min / hr)

KEY FEATURES
• Session + lifetime tracking — session resets like a trip odometer, lifetime accumulates forever
• Duration-scaled estimates — longer AI responses register higher energy use
• Dark mode — toggle between light and dark themes
• Compact mode — minimized view for quick glances
• 30+ platforms supported — text, image, video, and audio AI tools
• 100% local — no accounts, no servers, no data leaves your device

PRIVACY
Footprint AI collects zero data. Everything is processed and stored locally on your device using Safari's built-in storage. No account required. No network requests. Uninstalling removes everything. Your AI usage is your business — we just help you see its footprint.

METHODOLOGY
Estimates are based on published research: Epoch AI (2025) for inference energy, Luccioni et al. (2024) for model-specific benchmarks, IEA (2024) for datacenter energy overhead, and EPA eGRID (2023) for grid emissions factors. These are order-of-magnitude estimates — actual usage varies by model, prompt length, and datacenter efficiency. The goal is awareness, not audit-grade precision.