AI Meter

AI Meter is a macOS utility that monitors how many tokens your local AI coding agents consume and translates that usage into estimated electricity and water costs. It reads token counts from Codex, Claude Code, Cursor, OpenCode, and Gemini CLI by scanning their local records on your machine, then displays an interactive dashboard showing usage over time alongside environmental impact metrics. The goal is simple: make the invisible cost of AI assisted development visible so you can understand what your workflow actually consumes at the infrastructure level.

The problem it addresses is invisible consumption. Developers using AI coding assistants generate thousands or tens of thousands of tokens daily without any sense of the resources behind those requests. Cloud providers handle the compute in massive data centers, but the electricity and cooling happen somewhere real, and most users have no visibility into what their workflow actually costs in environmental terms. Carbon pledges and sustainability reports are everywhere in corporate communications, yet the tools we use daily provide no feedback on what they consume. AI Meter makes that cost concrete by attaching numbers to your usage patterns, turning abstract compute into tangible resource estimates.

Under the hood the app uses a baseline of 0.39 kilowatt hours per million tokens to estimate facility energy consumption. That figure comes from published estimates of large language model inference costs at scale. You can adjust it anywhere from 0.20 to 0.75 kilowatt hours depending on the efficiency assumptions you want to apply. More efficient data centers using renewable power and advanced cooling might warrant a lower number. Older facilities with less optimized infrastructure might justify a higher estimate. The configurability means you can model different scenarios rather than accepting a single fixed assumption, which matters if you are preparing a report that needs defensible methodology.

Water usage estimation relies on configurable parameters for power usage effectiveness and liters per kilowatt hour of IT load to model direct cooling needs. The defaults reflect industry averages from data center sustainability disclosures, but power users can tune them to match specific data center profiles if they have better information. The point is not to produce an exact audit quality measurement but to provide a reasonable order of magnitude estimate that grounds abstract token counts in physical resources. Better to have an informed approximation than no data at all.

Privacy is absolute. There are no accounts, no usage data leaves your Mac, and the only network activity is a public update check to pull new versions. Everything stays local, which fits the ethos of developers who care about controlling where their data goes. You install the app, it reads local logs, and it shows you a dashboard. That is the entire data flow. No telemetry, no analytics, no cloud sync, no registration form, no email required. The app just does its job and stays out of your way.

The interface is a simple dashboard with an interactive chart. You see token counts over time, cumulative energy estimates, and water usage projections in one view. It runs in the background and updates as your coding agents log activity, so the numbers stay current without manual intervention. You can glance at it periodically to track trends or dig into the data when you need specifics for a sustainability report or internal audit. The learning curve is essentially zero because there is nothing to configure beyond the optional efficiency parameters.

Who needs this? Developers and companies that want to quantify the environmental footprint of their AI assisted workflows. Sustainability teams producing carbon accounting reports need data from every tool the company uses, and coding assistants are a growing slice of that pie. Individual developers curious about their personal impact can see cumulative usage and decide whether to change their habits. Engineering managers evaluating different AI tools might use the data to compare which assistant is more resource efficient for similar tasks. Researchers studying the environmental cost of AI in practice finally have a way to gather real usage data. The audience is niche, but for those who care about this question, there has been no easy way to get an answer until now.

AI Meter is completely free for both individuals and companies. You download the app directly from the website or install via a terminal command, and it works out of the box with no signup, licensing, or payment. The source code is available on GitHub for anyone who wants to inspect the logic, verify privacy claims, or contribute improvements. Open source means you do not have to trust marketing copy; you can read the code yourself and confirm that it does exactly what it says.