GitHub →
—
installations
—
tokens being tracked weekly
—
liters made visible weekly
—
top platform
💧 AI Infrastructure Accountability

Every prompt you send
costs water.
Now you can track it.

AI data centers consumed billions of liters of water last year. Microsoft: +34%. Google: +20%. No one tells you your share. TokenWater generates a memory snippet you paste into your AI tools once — after that, your AI tracks your usage automatically.

✓ You do not need to keep this page open. Paste once. Done.
1
Set your goal

Choose your weekly water quota

Your AI will track toward this limit and alert you automatically at 50%, 75%, 90%, and 95%. Every morning, your first conversation will show you yesterday's usage and your pace for the week.

10L
liters per week
~100,000 tokens
1L casual 3L regular 7L power 15L heavy 30L

≈ roughly one toilet flush per week

Power user range. Long sessions, multiple platforms, research-heavy work. A single long research conversation typically uses 1–2L.

2
Pick your tools

Which AI platforms do you use?

Select every tool you use regularly. You'll get a separate snippet for each. Persistent Memory platforms remember your tracking automatically — set it once and forget it. Session Only platforms require pasting at the start of each conversation. Most people only bother with the persistent ones long-term.

3
Install your tracker

Copy and paste your snippets

Follow the exact steps for each platform. You're pasting into your AI's long-term memory — it only takes a minute and you only do it once. Come back every 4 weeks to grab a fresh snippet with updated benchmarks.

💧
You're set up. Close this tab.

Your AI tools will now track your water usage automatically.
Every morning, your first conversation shows you yesterday's usage and your weekly pace.

The only reason to come back here: refresh your snippet every 4 weeks, or change your quota.
That's it.

Manual Usage Log

Using the API directly? Log exact token counts here for a more precise water estimate. Find them in your API response under usage.input_tokens and usage.output_tokens.

No entries yet.
API Keys (Optional — stored locally only)
The Water Math
The Proxy Metric
~100ml of water consumed per 1,000 tokens processed. Derived from data center PUE data and cooling water intensity ratios. Directional estimate — actual consumption varies by provider, region, and model.
Sources
Li, P. et al. (2023). "Making AI Less Thirsty." UC Riverside. arxiv.org/abs/2304.03271

Luccioni et al. (2023). "Power Hungry Processing." arxiv.org/abs/2311.16863
The Real Numbers
Microsoft: 6.4 billion liters in 2022 (+34% YoY). Google: +20% same period. Neither provides per-user breakdowns. TokenWater is the individual meter that doesn't otherwise exist.
Built By
Ryan Lester — Navy veteran, economics student, foresight researcher at University of Houston. MIT licensed. GitHub →