claude-usage

Reports Claude Code token usage grouped by model for a date range, read straight from the local transcript files in ~/.claude/projects/. On the subscription plan this is the only place the numbers live — the API console only covers pay-as-you-go API keys, not subscription sessions.

How it works

Every assistant turn is logged to a *.jsonl transcript with a message.usage block (input_tokens, output_tokens, cache_creation_input_tokens, cache_read_input_tokens) and an ISO timestamp. The bundled script walks all transcripts — main sessions, subagents, and workflow runs — sums usage for messages whose day falls in the requested range, and groups by model. Days are bucketed in IST (+05:30) by default.

Your job: turn the user's phrase into --start / --end, then run the script