ratelworks/token-horse

Token Horse is a small terminal animation that shows a pixel-art horse reacting to how fast an AI model is generating tokens. When the AI is writing quickly the horse gallops, and when work stops the horse stands still and blinks occasionally while waiting for the next task. The project is inspired by the tiny horses that used to appear on old taxi meters, running faster as the fare increased. The tool works with two AI coding environments: Claude Code and the Codex CLI from OpenAI. For Claude Code it plugs into the status line shown at the bottom of the session, configured through a settings.json file with a one-second refresh interval. For Codex CLI, which does not allow external commands in its built-in status area, Token Horse instead reads the session log files that Codex writes to disk and runs separately in a different terminal pane or tmux split. In Claude Code mode, the horse responds to the actual tokens the current session has used. It reads the session transcript file, adds up input and output tokens (excluding cache reads), calculates the rate of change between updates, and maps that number to the horse's speed. The speed is a smooth continuum rather than discrete steps: near 20 tokens per second the horse trots slowly, and above around 900 tokens per second it runs at full gallop. A burst of tokens is reflected immediately in the animation. The horse is drawn using colored terminal blocks (truecolor ANSI) in shades of green, so it looks consistent across different monospace fonts. The default size is 32 characters wide by 8 lines tall. A smaller option at half that size is available with a flag. Installation is a single npm global install. The package can also be tested without installing it using npx. It is licensed under MIT by Ratelworks Inc.