Tokens: Scheduled Token Usage Reports for Codex

I built Tokens to turn a raw token counter into useful scheduled reports. The open-source Codex skill reports daily, weekly, and monthly usage at user-selected times, showing a compact period total and a total tracked since setup.

Local collectors read the Codex app-server counter and maintain independent accounting checkpoints. Raw telemetry and accounting state stay on the user’s machine; only the finished report aggregate goes to a persistent ChatGPT relay, which creates or updates the corresponding Scheduled Task for native notifications.

A prepare/commit protocol advances the local checkpoint only after confirmation of the server task update. If a run is missed, the next successful report covers the interval since the previous committed report.

The project uses a Python standard-library reporter and requires no external connector, third-party database, or hourly poller. Reporting is account-level, with no claim of source-level attribution.

Source: https://github.com/maddiedreese/tokens