# TokenRank

> Compare locally collected aggregate AI token activity across Codex, Claude Code, Gemini, Qwen, and more. Prompts, code, and chats stay local.

TokenRank is a public leaderboard for locally collected aggregate token activity across supported AI agents and tools. It provides an activity signal, not a productivity score, without exposing prompts, source code, chats, filenames, or file contents.

## Core Pages

- [Leaderboard](https://tokenrank.org/) - Public ranking by total tokens, estimated cost, and individual AI agents and tools.
- [How TokenRank works](https://tokenrank.org/#how-it-works) - Three-step explanation of identity, local aggregation, and ranking.
- [FAQ](https://tokenrank.org/#faq) - Direct answers about measurement, privacy, supported tools, deduplication, and joining.
- [Rules](https://tokenrank.org/rules) - Privacy, fair-play, and scoring rules.
- [Onboarding](https://tokenrank.org/onboard) - Local preview before login, identity claim, collector setup, first upload, and dashboard handoff flow.

## Machine-Readable Endpoints

- [Boards API](https://tokenrank.org/api/boards) - Supported leaderboard boards and tool keys.
- [Leaderboard API](https://tokenrank.org/api/leaderboard) - Public leaderboard data with board and range query parameters.

## Privacy Boundary

The collector sends only aggregate rows: date, tool, model, input tokens, output tokens, cache read tokens, cache write tokens, and totals. Raw prompts, code, conversations, filenames, and file contents are intentionally excluded.
Current public rows are collector-reported local aggregates. The server validates structure, account ownership, totals, and duplicate keys, but does not verify the data against provider bills. Estimated cost is not a bill.