See how many tokens your AI coding agents really burned.
A fast, 100% local command-line tool that reads your Claude Code session history and shows you exactly how many tokens — and how many dollars — you've spent. No account, no sign-up, no upload. Nothing ever leaves your machine.
whoburnedmore.com · the hosted leaderboard
🔥 whoburnedmore — your local AI token burn report
────────────────────────────────────────────────
1.82B tokens burned $3,410.00 est.
12,704 assistant messages · 18 active days · 2026-05-29 → 2026-06-15
By model
████████░░░░░░░░░░ claude-opus-4-8 1.10B $2,512.40
█████░░░░░░░░░░░░░ claude-sonnet-4-6 512.00M $640.10
██░░░░░░░░░░░░░░░░ claude-haiku-4-5 210.00M $36.20
By project
███████░░░░░░░░░░░ api 903.00M $1,640.00
████░░░░░░░░░░░░░░ web-app 540.00M $980.00
██░░░░░░░░░░░░░░░░ infra 377.00M $790.00
Prompt cache 97.4% read-hit rate (1.71B cached reads)
────────────────────────────────────────────────
100% local · nothing left your machine.
Compare on the public board → https://whoburnedmore.com
✨ Features
- 🔒 Completely local. Reads the transcripts already on your disk. Makes zero network requests — no telemetry, no account, no API keys.
- 💸 Real cost estimates. Turns raw token counts into dollar figures using a transparent, editable pricing table.
- 🧩 Breakdowns that matter. See your burn split by model and by project, so you know exactly where the tokens went.
- ⚡ Prompt-cache insight. Surfaces your cache read-hit rate — the single biggest lever on what you actually pay.
- 🖼️ Beautiful HTML dashboard.
--htmlwrites a self-contained, offline dashboard you can open in any browser or share. - 🪶 Tiny & dependency-free at runtime. A ~15 KB bundle with no runtime dependencies. Nothing to trust but the code you can read here.
🚀 Quick start
You'll need Node.js 20+ and some Claude Code usage on this machine.
Run it instantly with npx
npx github:amiinwani/whoburnedmore.com
Or clone and run
git clone https://github.com/amiinwani/whoburnedmore.com.git cd whoburnedmore.com npm install # also builds the CLI npm start # == node dist/cli.js
📖 Usage
whoburnedmore # print your burn report whoburnedmore --html # also write ./whoburnedmore.html whoburnedmore --html out.html # ...to a custom path whoburnedmore --since 30 # only count the last 30 days whoburnedmore --dir <path> # read transcripts from a custom directory whoburnedmore --json # raw aggregated JSON (pipe it anywhere) whoburnedmore --version whoburnedmore --help
| Flag | What it does |
|---|---|
--html [file] |
Write a self-contained HTML dashboard (default ./whoburnedmore.html). |
--since <days> |
Only include usage from the last N days. |
--dir <path> |
Point at a custom transcripts directory instead of ~/.claude/projects. |
--json |
Print the aggregated data as JSON instead of the report. |
--version / --help |
Print the version / help. |
🔍 How it works
Claude Code stores one JSON Lines transcript per session under:
~/.claude/projects/<project>/<session-id>.jsonl
Every assistant turn in those files carries a usage block — input tokens, output tokens, and the prompt-cache read/write counts — plus the model name and a timestamp. whoburnedmore streams through those files line by line, adds the counts up by model, by project, and by day, multiplies by a public pricing table for a dollar estimate, and draws the report. That's the whole trick. The code is small on purpose — start with src/scan.ts.
🛡️ Privacy
This tool is read-only and offline by design.
- It only reads files; it never writes to your transcripts or deletes anything.
- It makes no network requests of any kind — no analytics, no "phone home," no auto-update.
- The optional HTML dashboard is fully self-contained: no external fonts, scripts, or trackers.
- Your prompts and code are never parsed for content — only the numeric
usagecounters and the model name are read.
Want to verify? It's ~400 lines. Read src/, or run with --json and inspect exactly what it computes.
🌐 The hosted version
This CLI is the open-source, local-only sibling of whoburnedmore.com — a public leaderboard where developers compare how much they've burned across Claude Code, Codex, Cursor and more. The website is a separate, hosted product; this repository contains only the local tool and none of its backend.
🤝 Contributing
Issues and PRs welcome — see CONTRIBUTING.md. Good first contributions: add models to the pricing table, support more agents' transcript formats, or improve the dashboard.
npm install # install + build npm test # run the unit tests npm run typecheck # strict type check npm run build # bundle to dist/cli.js
📄 License
MIT © 2026 whoburnedmore