H2AI estimates the water, energy, tokens, and cost behind every ChatGPT, Claude, Gemini, and DeepSeek reply — right where you're already chatting.
No setup beyond installing it once.
Add the extension once. It quietly watches for finished responses on the sites you already use.
H2AI never touches your prompt data or API keys — it reads the visible response text already on your screen.
A card appears under the response with tokens, cost, and water — and adds to your running session total.
H2AI reads the length of the response you just got and estimates what it took to make.
Scaled from published research on AI inference water use — methodology one tap away.
Input and output tokens priced against current API rates for the model you're talking to.
See input vs. output tokens separately — the number that actually drives cost and footprint.
A running tally for the conversation you're in, plus a lifetime total across every site.
Haiku and Opus don't cost the same — H2AI detects the model in use and adjusts accordingly.
Every calculation happens in your browser. No prompt or response text is ever sent anywhere.
No AI provider publishes real per-query water or energy data. H2AI estimates from response length using published research (including Google's 2025 Gemini inference study) and current public API pricing. Built for awareness, not as an audited metering tool — every card links to the methodology behind its number.
Tokenistt is an AI infrastructure startup building the tooling layer for how people and teams work with AI models. H2AI started as an internal curiosity about what a single prompt actually costs — we're releasing it free for anyone to use.
This extension is free, always will be, and isn't a product pitch — it's a small, honest tool from a team that spends most of its day thinking about AI infrastructure.
Get H2AI free → tokenistt.com ↗Yes — free forever, no account, no paid tier. It's a small tool Tokenistt built and is giving away.
No — they're estimates based on response length, published research, and current public API pricing. No AI provider discloses real per-query figures, so treat this as an awareness tool, not a metering device. Every card links to the methodology behind its number.
No. All calculation happens locally in your browser using chrome.storage. No prompt or response text is ever transmitted.
ChatGPT, Claude, Gemini, and DeepSeek today. More may be added later.
It started as an internal curiosity about what a single AI prompt actually costs — in tokens, dollars, and water. We thought it was worth sharing.
Not on the Chrome Web Store yet — until then, it installs from GitHub as an unpacked extension. Same code, one extra click.
In Chrome (or Edge), go to chrome://extensions and toggle Developer mode on, top right.
Click Load unpacked and select the ai-water-meter folder from what you downloaded.
Visit ChatGPT, Claude, Gemini, or DeepSeek and send a message — the card appears under the reply.
The README covers the same steps in more detail, plus what to do if the card doesn't show up.
View on GitHub →