The latest articles on DEV Community by Beau Sterling (@beausterling). https://dev.to/beausterling https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3314838%2Fd5f13891-e725-4439-83d6-f73f9a24dec7.jpeg https://dev.to/beausterling en Beau Sterling Fri, 02 Oct 2026 15:50:34 +0000 https://dev.to/beausterling/how-much-water-does-claude-code-use-i-built-a-cli-to-measure-it-16in https://dev.to/beausterling/how-much-water-does-claude-code-use-i-built-a-cli-to-measure-it-16in <p>My AI coding agents used about <strong>47.8 litres of water in September</strong>. That's roughly eight toilet flushes, and it's an estimate with a wide range. Here's how I got the number, and the open-source CLI I built to track it live.</p> <div class="ltag-github-readme-tag"> <div class="readme-overview"> <h2> <img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"> <a href="https://github.com/beausterling" rel="noopener noreferrer"> beausterling </a> / <a href="https://github.com/beausterling/drip-ai-water-usage" rel="noopener noreferrer"> drip-ai-water-usage </a> </h2> <h3> How much water is your AI agent drinking? Live water-usage meter for Claude Code &amp; Codex CLI — status line, split-pane meter, research-backed estimates. </h3> </div> <div class="ltag-github-body"> <div id="readme" class="md"><div class="markdown-heading"> <h1 class="heading-element">drip 💧</h1> </div> <p><strong>How much water is your AI agent drinking?</strong> drip estimates the water used by terminal AI agents (Claude Code and Codex CLI) and shows it live in your status line, a split-pane meter, or a full breakdown page, with every number traced back to published research.</p> <div class="snippet-clipboard-content notranslate position-relative overflow-auto"><pre class="notranslate"><code>Opus 5.5 │ my-project │ ████░░░░░░ 48% │ 💧 412 mL (1.7 glasses) · today 3.1 L </code></pre></div> <ul> <li> <strong>Real token counts.</strong> drip reads your agent's local logs, so the token counts aren't guesses.</li> <li> <strong>Research-backed coefficients.</strong> Low, mid and high estimates, each with its source cited.</li> <li> <strong>Private.</strong> Everything is local: no accounts, no telemetry, and nothing leaves your machine.</li> <li> <strong>No dependencies.</strong> Python 3.11+ standard library only.</li> </ul> <div class="markdown-heading"> <h2 class="heading-element">Install</h2> </div> <div class="highlight highlight-source-shell notranslate position-relative overflow-auto js-code-highlight"> <pre>git clone https://github.com/beausterling/drip-ai-water-usage.git <span class="pl-k">~</span>/.drip <span class="pl-k">~</span>/.drip/bin/drip install</pre> </div> <p><code>install</code> plays a short intro, imports your existing Claude Code and Codex history, puts <code>drip</code> on your PATH (via <code>~/.local/bin</code>), and turns on the Claude Code status line…</p></div> </div> <div class="gh-btn-container"><a class="gh-btn" href="https://github.com/beausterling/drip-ai-water-usage" rel="noopener noreferrer">View on GitHub</a></div> </div> <h2> Why water? </h2> <p>Every Claude Code or Codex request runs on GPUs in a data center. Water gets used twice:</p> <ol> <li> <strong>On site:</strong> evaporative cooling. Data centers report this as WUE (litres per kWh).</li> <li> <strong>Off site:</strong> the power plants that generate the electricity also consume water (EWIF, litres per kWh generated).</li> </ol> <p>The second one is usually bigger. In drip's mid estimate it's about 88% of the total.</p> <h2> The formula </h2> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code>water (mL) = tokens × energy per token (Wh) × water per kWh (L/kWh) water per kWh = WUE / PUE + EWIF </code></pre> </div> <ul> <li> <strong>Tokens</strong> are exact. drip reads the logs your agent already writes: <code>~/.claude/projects/**/*.jsonl</code> for Claude Code and <code>~/.codex/sessions/**</code> for Codex CLI. It counts input, output, cache-read and cache-write tokens separately and dedupes by message id.</li> <li> <strong>Energy per token</strong> comes from measured inference energy (ML.ENERGY, Microsoft's study in <em>Joule</em>, Google's disclosure). Output tokens cost far more than input, and cache reads are a small fraction of input. Models are scaled by list price, the only public proxy for how expensive a model is to serve.</li> <li> <strong>Water per kWh</strong> comes from Lawrence Berkeley National Lab's 2024 US data center report and Li et al., <em>Making AI Less Thirsty</em> (CACM 2025). Mid estimate: about 3.6 L/kWh including generation, about 0.4 L/kWh on site only.</li> </ul> <h2> How accurate is it? </h2> <p>Token counts: exact. Water: maybe 3–5× off in either direction, and the low-to-high span is 10–30×, because no provider publishes energy per token. So drip shows a range everywhere instead of a falsely precise number. It's good for trends ("I used 3× more this week") and bad for exact litres.</p> <p>The biggest unknown is cache reads. They're most of an agent's tokens, and nobody knows exactly what they cost.</p> <h2> What it looks like </h2> <p>In Claude Code, a segment in the status line:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight plaintext"><code>Opus 5.5 │ my-project │ ████░░░░░░ 48% │ 💧 412 mL (1.7 glasses) · today 3.1 L </code></pre> </div> <p>For Codex or any other agent, <code>drip run codex</code> opens a split pane with a live meter (a little ASCII bottle that fills up) in Ghostty, Warp, iTerm2, tmux or macOS Terminal.</p> <p><code>drip open</code> shows a breakdown page: per model, per token type, the uncertainty range, 30 days of history, every coefficient linked to its source, and a share card.</p> <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsff5ki1w8b646g1rrd9n.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsff5ki1w8b646g1rrd9n.png" alt="drip breakdown page" width="800" height="479"></a></p> <h2> Install </h2> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>git clone https://github.com/beausterling/drip-ai-water-usage.git ~/.drip ~/.drip/bin/drip <span class="nb">install</span> </code></pre> </div> <p>Python 3.11+, standard library only, nothing leaves your machine. MIT licensed.</p> <h2> What I'd love help with </h2> <p>If you have better data on per-token inference energy, especially for cache reads, open an issue. The coefficients live in one TOML file, and since drip stores tokens rather than water, updating them re-prices your whole history.</p> ai showdev opensource python