drip

I made drip because no AI provider publishes water usage per token. It estimates the water your AI coding agent uses and shows it live while you work, built from published research (ML.ENERGY, Microsoft, Google, Lawrence Berkeley National Lab, CACM) with every coefficient linked to its source. It includes a water meter in the Claude Code status line, a live meter beside Codex CLI via 'drip run codex', and a browser breakdown by model with low, mid, and high range plus 30-day history. Token counts come from your agent's logs; water is estimated, so it shows a range. It runs locally, requires no account, sends no telemetry, and is free under the MIT license.