Tokenmeter is a Windows desktop app that reads your local Claude Code session files and gives you a live terminal-style dashboard of token usage, costs, and activity patterns — entirely offline, zero telemetry, no API keys.
Usage & Cost
- Total tokens, estimated cost, today's usage, session count — always at a glance
- Accurate cache savings vs uncached input pricing (per-record, per-model)
- 30-day cost projection based on your 7-day rolling average
- Daily cost alert — native notification when you exceed a threshold
Activity
- 14-day daily activity chart with input/output breakdown
- 90-day GitHub-style heatmap of your Claude activity
- Peak hours chart — see which hours you run heaviest
- Per-project sparklines — 14-day mini charts in the breakdown table
- Last active session always pinned at the top
App
- System tray — lives in your taskbar, shows today's token count on hover
- 60-second auto-refresh +
Ctrl+Rto refresh manually - Idle pixel companion with 7 animations: breath, blink, think, dance, wink, surprise, sway
- Dark terminal aesthetic, compact 720×600 window
Download from Releases →
| File | Description |
|---|---|
Tokenmeter Setup 1.0.0.exe |
Installer with wizard — creates Desktop & Start Menu shortcuts |
Tokenmeter 1.0.0.exe |
Portable — drop anywhere and run, no install needed |
Windows SmartScreen: on first launch Windows may say "Unknown publisher" — click More info → Run anyway. This is normal for unsigned apps.
- Windows 10/11 x64
- Claude Code installed and used at least once
Tokenmeter scans %USERPROFILE%\.claude\projects\**\*.jsonl — the same session files Claude Code writes locally. No data ever leaves your machine.
git clone https://github.com/DewashishCodes/tokenmeter
cd tokenmeter
npm install
npm start # dev mode
npm run build # → dist/ (installer + portable)Requires Node.js 18+ and Windows with Developer Mode enabled (Settings → System → For developers).
Cost estimates use pricing.json bundled with the app — hand-editable if Anthropic updates rates. All figures are displayed as ~$ to make clear they are estimates.
MIT © 2026 Dewashish Lambore


