How It Works

Data Collection

When you run the setup command, a small hook script is installed in your Claude Code configuration. This hook fires automatically after each Claude Code session ends.

The hook reads your local session file, extracts token counts and model names only, then sends them to our server. No prompts, responses, or code are ever transmitted.

Claude Code session ends
→ Hook parses local JSONL
→ Extracts: model, input/output/cache tokens
→ POST to /api/usage/report
→ Server calculates cost & stores

Cost Calculation

Costs are calculated server-side using Anthropic's published pricing (per 1M tokens):

ModelInputOutputCache ReadCache Write
Opus 4.6$15$75$1.50$18.75
Sonnet 4.6$3$15$0.30$3.75
Haiku 4.5$0.80$4$0.08$1

Formula: (input tokens / 1M × rate) + (output tokens / 1M × rate) + (cache read / 1M × rate) + (cache write / 1M × rate)

Leaderboard Ranking

Rankings can be sorted by:

  • Tokens— total input + output tokens across all sessions
  • Cost— estimated USD cost based on model pricing

Filter by period (today, week, month, all time) and role (developer, non-developer).

Streaks & Gamification

Day Streak counts consecutive days with at least one Claude Code session. Streaks stay active if you used Claude yesterday (grace period until end of today).

Weekly MVP is the user with the highest token usage in the past 7 days, shown on the team stats page.

Privacy

We collect only:

  • Model name (e.g. claude-sonnet-4-6)
  • Token counts (input, output, cache read, cache write)
  • Session ID and date

Prompts, responses, and code are never collected or transmitted.