The privacy contract

The CLI parses your usage logs on your machine. What gets uploaded is exactly one JSON document, validated against a public schema, containing only:

• token counts (input / output / cache)
• estimated or reported cost totals
• session, project and request counts (numbers, not names)
• active-day counts, streaks, first/last activity dates
• canonical model ids (e.g. claude-opus-4-8)
• an optional display name you type yourself

The schema cannot represent prompts, code, file paths, project names, or timestamps finer than a calendar date. The CLI shows you the full payload and asks for confirmation before uploading; --dry-run never uploads at all. Your IP is not stored — only a salted daily hash used for rate limiting. A Cursor session token, when used, is read locally and sent only to cursor.com — never to our API.

Cards are self-reported statistics, like any year-in-review.

The parser and public schema are open source. View the source on GitHub.