What is token-stamp?

STAMP is a tool that reads your local AI coding assistant usage logs and turns them into a visual "walking journey." If you have used Claude Code or Codex, those tools write session logs to your computer. STAMP looks at how many tokens you consumed, converts that into steps, distance, and energy, and renders a shareable poster card showing your activity as if you had walked a real route on a map. The tool runs entirely on your machine. It does not upload your data anywhere, and there are no accounts or telemetry. You start it with a single command, npx token-stamp, which launches a local server and opens a browser window. You can also install it globally or run it directly from the cloned repository. It works on macOS, Windows, and Linux, and requires no external dependencies to install. STAMP reads from two specific locations on your disk. For Claude Code, it parses JSONL files in ~/.claude/projects/ that contain token usage data for each assistant response. For Codex, it reads session files in ~/.codex/sessions/ and takes the final cumulative token count per session. The only outbound network requests are for map tiles, an optional geocoding lookup, and road snapping for the decorative walking route. The energy calculation weights tokens by role. Generated output tokens count for about 3 joules each, fresh input tokens for 0.3 joules, and cache reads for 0.05 joules. The tool then divides total energy by 167 joules per step to get a step count, and uses a 0.762 meter stride for distance. The README notes that counting cache reads at full price would overstate energy by roughly 30 times, so the weighting matters. A methodology page inside the app explains the full derivation. The map route is decorative. It uses real map tiles and real streets, and starts at your actual location, but the loop shape is a deterministic emblem derived from your token data, not an actual GPS trace. The README describes the route size as roughly 2 to 3 kilometers and purely decorative.