Show HN: AI Meter – Local token usage with energy and water estimates
AI Meter is a macOS application that tracks local AI coding agent usage without sending data to external servers, ensuring privacy by default. The tool estimates environmental impact by converting token counts into electricity consumption and direct cooling water usage using configurable factors. It supports major local coding agents like Codex, Claude Code, Cursor, OpenCode, and Gemini CLI by reading local provider files for deduplicated token records. The methodology excludes model training, e
Analysis
TL;DR
- AI Meter is a macOS application that tracks local AI coding agent usage without sending data to external servers, ensuring privacy by default.
- The tool estimates environmental impact by converting token counts into electricity consumption and direct cooling water usage using configurable factors.
- It supports major local coding agents like Codex, Claude Code, Cursor, OpenCode, and Gemini CLI by reading local provider files for deduplicated token records.
- The methodology excludes model training, embodied hardware, grid carbon, and user-device energy, focusing strictly on inference-related facility electricity and water.
Why It Matters
This tool addresses the growing need for transparency regarding the environmental footprint of generative AI, specifically in developer workflows where local inference is becoming more common. By providing localized, private metrics on energy and water usage, it empowers developers to make more sustainable choices without compromising data privacy. It highlights the importance of measuring direct operational impacts (inference) separately from broader lifecycle emissions.
Technical Details
- Data Collection: Reads local provider files from supported coding agents (Codex, Claude Code, Cursor, OpenCode, Gemini CLI), deduplicates token counts, and combines them locally on the Mac.
- Electricity Estimation: Uses a default inference factor of 0.39 facility kWh per million tokens, with an adjustable uncertainty range of 0.20–0.75 kWh.
- Water Usage Calculation: Estimates direct cooling water use based on a Power Usage Effectiveness (PUE) of 1.20 and a Water Usage Effectiveness (WUE) of 0.45 L per IT kWh, both of which are adjustable within the app.
- Privacy Architecture: Operates entirely locally with no accounts required; the only network request is a public update check, ensuring no usage data leaves the device.
- Exclusions: Explicitly excludes model training, embodied hardware, off-site electricity/water, grid carbon, networking, and user-device energy from its calculations.
Industry Insight
- Developers and organizations should consider implementing local monitoring tools to gain visibility into the resource intensity of their AI workflows, particularly as edge and local inference become more prevalent.
- The distinction between inference costs (electricity/water) and broader lifecycle impacts (training/hardware) is critical for accurate sustainability reporting; tools like AI Meter help isolate these variables for better decision-making.
- Privacy-preserving analytics will likely become a standard requirement for enterprise AI adoption, as organizations seek to monitor usage and compliance without exposing sensitive code or data to third-party services.
Disclaimer: The above content is generated by AI and is for reference only.