See the water footprint of your AI coding, live.

Water is a VS Code extension that reads your Claude Code and Codex logs on disk and shows today's estimated water and energy use right in the status bar. Nothing leaves your machine.

Get it for VS Code View source

2.7 L–2.5 kL

What it does

Everything runs in the extension host. No API key. No account. Works offline.

Reads your logs locally

Parses ~/.claude/projects and ~/.codex/sessions directly, deduping messages and counting per model.

Live in the status bar

Today's water estimate updates as you work and deepens from light to deep blue as it climbs. Click for the full panel.

Per-model breakdown

See which models drove your footprint. Heavyweight reasoning models get a 10x weight, because model choice is the biggest lever.

An honest range, never one fake number

Published estimates for the water cost of one AI prompt disagree by more than 1000x. So Water shows a range, with every coefficient labeled and sourced.

0.3 mL
vendor-optimistic

519 mL
worst-case ceiling

Optimistic — Google ~0.26 mL/prompt, Sam Altman ~0.32 mL (~0.34 Wh) per prompt.
Pessimistic — UC Riverside ~10–25 mL/prompt; 519 mL worst-case ceiling.
Reasoning multiplier — 10x for heavyweight models (Opus, o1/o3, GPT-5, etc.).

Logged vs actual — a feature, not a bug

Raw JSONL logs undercount real token usage by 10–100x: they use streaming placeholders and omit hidden thinking tokens. Water always shows the logged sum, and when you paste your true total from the statusline it shows the corrected figure and the undercount factor, instead of hiding the discrepancy.

Install

Free on the VS Code Marketplace. Or build the VSIX from source.

VS Code Marketplace GitHub & VSIX

Or from the command line, once published:

code --install-extension royalpinto007.water

Water · measured locally, shared only if you choose to.
Built by royalpinto007 · VS Code and the VS Code logo are trademarks of Microsoft.