Heads-up display · refreshed daily

Claude Code
receipts.

Every claim on the home page about Claude-in-the-loop is downstream of one number: how much I actually drive the model. This page reads my local Claude Code stats cache — the same data behind /usage — and lays it out: tokens, sessions, tool calls, model mix. No edits, no hand-tuning. If activity drops, that shows up here too.

Updated …

Total tokens · 7d — input + output

Total tokens · 30d — input + output

Tool calls · 30d — 7d: —

Cache read · lifetime — — messages

01 Daily token volume

Last 30 days of total tokens per day (input + cached + output). Bars are CSS-only — no chart library, just the numbers as they came back from the API.

No daily data yet — first run will populate this.

02 Model mix · 30d

Which Claude models I actually use. Tokens are the sum of input (uncached + cache reads + cache writes) and output. Sorted by total.

No model breakdown yet — first run will populate this.

03 How this page works

Source Local ~/.claude/stats-cache.json on my laptop — the same data Claude Code's /usage renders. No Anthropic API call.
Cadence A small Python script reads the cache, writes claude-stats.json, and (via a launchd job) commits + pushes if anything changed.
Deploy Cloudflare Pages picks up the new commit and ships the static site within a minute.
Window Trailing 30 days for window totals + daily chart; lifetime numbers for cache reads and messages.

Source: github.com/sepulworld/resume-ai-mix-master · see scripts/fetch_claude_stats.py.

Claude Codereceipts.

01 Daily token volume

02 Model mix · 30d

03 How this page works

Claude Code
receipts.