tokenmaxing

tokenmaxing is a local-only, read-only terminal summary for Claude Code and Codex usage — your AI usage, wrapped.

It scans:

• Claude Code: ~/.claude (or $CLAUDE_CONFIG_DIR) for usage stats and transcripts, plus project .claude directories under your home directory.
• Codex: ~/.codex (or $CODEX_HOME).

It renders, one section at a time:

1. Total messages by source
2. Per-month token usage
3. Burn — local spend cache when present, otherwise a clearly marked list-price estimate
4. All-time token usage
5. Token usage summary
6. Input/output token comparison
7. Model token comparison
8. Model standing by input + output tokens
9. Most active time of day (messages per hour)
10. Daily activity matrix (contribution-graph style)

Every count and token figure comes from locally recorded data. The only estimated values are in the burn panel, marked with ≈.

Run

npx tokenmaxing

Or from a clone:

npm start

In a terminal the report reveals one section at a time, with a [NN/total] progress counter. Each section decodes in with a brief scramble-resolve animation.

• enter / space / → — next section
• ← / b — replay the previous section
• a — show all remaining sections
• q / ctrl-c — quit

Provider colors are categorical: Claude is orange, Codex is cyan, the combined total is amber — so any bar, model row, or table line maps to its source at a glance.

Piped or redirected output prints everything at once without colors or animation.

Privacy

The CLI reads local metadata and transcripts to compute usage, but the report does not print prompt text, model responses, tool outputs, credentials, or file contents. Nothing leaves your machine — there is no network call anywhere in the tool.