The usage view reads the transcripts Claude Code already writes to disk and shows what the work would have cost on the API, by day, by model and by project, plus where you stand in the current five-hour window.
- 5 h
- window tracked against your heaviest on record
- 4
- ranges: 7 days, 30 days, 90 days, everything
- 0
- network calls: it reads files already on your disk
1/4 · What it would have cost
Every token Claude Code used, read from its own transcripts and priced at API rates. Not a bill: a sense of scale.
Recorded in Station, in a sample project · waits sped up
Where the tokens went
Tokens a day stacked by model, which models did the work, which projects took it, what the agents did with their tool calls, and when in the day the work happens. Lines written and removed, and how much was served from cache.
Before the first prompt
How many tokens a new session in this project has already read before you type anything, split into CLAUDE.md files, memory, Station’s briefing and Claude Code itself. The number that tells you a CLAUDE.md has grown too long.
Two days, every feature, no card.