Privacy

Short version: it runs on your computer, and nothing leaves it unless you share it.

What it reads

The usage records your AI coding agents already keep on your computer: Claude Code in ~/.claude, Codex in ~/.codex, Gemini CLI in ~/.gemini/tmp, OpenCode in ~/.local/share/opencode. From them it counts tokens, model calls, times and the number of words you typed, then prints a receipt.

What it never reads or keeps

It never reads your code. It never keeps your prompts: words are counted in memory and the text is discarded. It never stores project names, folder paths, branch names or file names. It makes no network requests. There is no account and no telemetry.

What it stores on your computer

One file, ~/.token-damage/state.json: a run counter and which adjuster's notes it used recently, so the next receipt doesn't repeat them. Numbers only, never text. Delete it any time.

What leaves your computer

Only a receipt you choose to share: an image, or a link whose numbers live in the link itself, after the #. Browsers never send that part to any server, so this site never sees your numbers.

This website

No cookies, no analytics, no accounts. Fonts are served from this site. It remembers three things in your browser: whether sound is on, which aside it showed last, and which jokes you've already seen, so nothing repeats.

Token Damage is not affiliated with Anthropic, OpenAI, Google or the makers of OpenCode. Product names are used to describe compatibility.

The code is open source, so you can check all of this yourself.