Tokimeter

See what your AI coding tools actually use — and what it's worth.

Local-first usage, cost, and limits tracking for Claude Code and Codex CLI. No account, no telemetry, no prompt upload. Usage metadata only — your prompts and code never leave your machine.

Zero-setup report

npx tokimeter report

Reads the usage metadata your tools already write locally (Claude Code transcripts, Codex session logs) and shows cost and tokens by tool, model, project, and day — including exact cache read/write accounting and what prompt caching saved you. --days=7, --tool claude|codex, --json.

Limits: the 5-hour window

npx tokimeter limits

Claude and ChatGPT subscriptions meter on a 5-hour rolling window plus a weekly cap. limits shows your usage inside those windows, with warnings against budgets you set:

tokimeter config set budget.claude5h 60
tokimeter config set budget.daily 50

Full tracking

npm install -g tokimeter
tokimeter setup --auto     # restart your terminal afterwards

Keep using codex and claude exactly as before. You get live per-call tracking, a Claude status-line HUD (today ~$48.16 · 5h ~$48.16 (80% ⚠)), budgets, and:

tokimeter watch --tool claude    # live feed, per tool
tokimeter latest --tool codex    # most recent calls
tokimeter pricing refresh        # pull community price table (~280 models)
tokimeter doctor                 # verify setup
tokimeter uninstall              # removes everything it installed

Honest numbers

~$X marks API-equivalent estimates computed from exact local token counts — on a subscription that's the value you're extracting, not a bill. Billed API-key usage (via the optional localhost proxy) shows without the ~.

MIT licensed. Source, security notes, and issue tracker: https://github.com/toshipepe/tokimeter