Token Maxxer
A private-by-default AI usage tracker and public community leaderboard. Show your usage. Show what you shipped.
Live: https://token-maxxer-ten.vercel.app
User onboarding
- Sign in with GitHub (or email when the operator has configured delivery).
- Open Connect tools, generate the install command, and paste it into a terminal on your coding machine.
- The collector pairs, imports retained Claude Code / Codex / OpenCode usage, and installs background tracking every five minutes. Node.js 20+ is required; no repository clone or open terminal is needed.
- The page confirms Receiving uploads and shows each tool's coverage.
- Review Profile preview and publish your account. All non-hidden project names appear; private projects have no links. Make a project public to enable its link, or hide it to omit it entirely.
macOS uses a launch agent, Linux uses the user's crontab (requires cron), and Windows uses a scheduled task. The collector queues metadata locally while offline and retries. Stop, pause, unpair and revoke controls are included.
Development
npm install
npm run dev
When no database is configured, development starts an isolated local PostgreSQL server automatically and syncs the schema. Development-only email links print in the terminal. This never uploads local AI usage or enables telemetry by itself.
npm test # isolated PostgreSQL, accounting/privacy/retry regression tests
npm run check # lint, collector typecheck, tests, production build
Counting
Headline total = fresh input + output + cache read + cache write + reasoning (only when not already included in output). Codex input/cache and OpenCode reasoning are normalized to exclusive buckets. Daily and weekly views use UTC; weeks start Monday. Codex history is partitioned by day, model and project. Cumulative telemetry contributes differences between checkpoints, not a lifetime total moved to the latest day. Unknown historical timing stays in all-time totals and is excluded from period rankings.
Records deduplicate across collectors within an account. Claude Code transcript collection and telemetry are alternative methods: one canonical method is counted; existing overlapping evidence is retained but excluded from totals. Evidence is locally reported, never provider verified.
Operations and support
Production requires a configured PostgreSQL database, authentication credentials and real secrets. GET /api/health returns 200 only when these are configured and the accounting schema is available. Email sign-in is hidden in production until SMTP delivery is configured.
