Know what your AI tokens cost
Token Wise is a native macOS app that tracks your AI token usage and costs across Claude, GPT, and more. Completely offline. Privacy-first.
Everything you need to track AI costs
Built for developers who use AI coding assistants daily and want to understand where their tokens go.
Cost Dashboard
See daily, weekly, and monthly spending with rich charts. Break down costs by project, session, and model.
100% Offline & Private
All data stays on your Mac. No accounts, no cloud sync, no telemetry. Your usage data never leaves your machine.
Multi-Model Support
Track tokens across Claude Opus, Sonnet, Haiku, GPT-5, and more — with per-model pricing detected from each session, so mixed-model histories are costed correctly.
AI Auto-Analyze
One click runs your own claude or codex CLI headlessly over your usage summary and returns prioritized, numbers-grounded fixes — no API key, uses your existing subscription. Coming in the next release.
Cache Savings Detector
Pinpoint wasted cache writes, prefix invalidations, and large CLAUDE.md blocks loaded but never referenced. See the dollar amount you could reclaim per session.
Per-Project Breakdown
See exactly how much each project costs. Compare spending across repos and track your top sessions.
Session Deep Dive
Explore individual sessions turn-by-turn. See token usage, context growth, and content analysis for each conversation.
Command Line Interface
Same binary, terminal-friendly. Run `token-wise total`, `today`, or `sessions` for quick stats. Every command supports --json for scripting.
Use it from the terminal
The same binary doubles as a CLI. Pipe it into scripts, drop it in your status bar, or just check your spend without leaving the shell.
$ token-wise total Total: $142.38 (30 days, 247 sessions) $ token-wise today Today: $4.21 (8 sessions) Cache hit rate: 73% $ token-wise sessions --limit 5 PROJECT COST TOKENS CACHE api-gateway $8.42 2.1M 81% web-dashboard $5.17 1.4M 68% ml-pipeline $3.95 980K 72% docs-site $2.10 540K 85% auth-service $1.78 420K 79% # Pipe into anything — every command supports --json $ token-wise today --json | jq .total_cost_usd 4.21
alias token-wise='/Applications/token-wise.app/Contents/MacOS/token-wise'
Let your own AI find the savings
Token Wise discovers the claude or codex CLI already on your machine and runs it headlessly over a summary of your usage — using your existing subscription, no API key. You get prioritized, numbers-grounded fixes and a plain-English read on where the money goes.
$ token-wise analyze Analyzing with codex (this can take a minute)… ## Actionable fixes 1. Move routine work off the premium model — it's 74% of all spend. Even a 10% shift targets ~$2.8K. 2. Trim startup context: ~23K tokens/session paid on every session start. Shrink CLAUDE.md and always-loaded skills. 3. Batch the 86 short sessions (≤4 messages, >$0.50) — each repeats the full cache-write tax. ## Narrative Most spend concentrates in one project and one premium model. Cache hit rate is already strong, so the biggest lever is reducing repeated context volume…
Runs claude -p or codex exec with the login you already have. No extra account, no per-call billing surprise.
The model sees an aggregate summary computed locally — spend, cache rates, overhead, top projects — and must cite it.
Use the Analyze tab in the app, or `token-wise analyze` in scripts. Both pick whichever engine you have installed.
Free to use
No subscriptions. No accounts. No hidden fees.
All features included. No upgrades, no add-ons.
- Full cost dashboard & analytics
- All AI models supported
- Per-project cost breakdown
- Session deep-dive with turn analysis
- AI auto-analyze via your own claude/codex CLI (coming soon)
- Cache hit rate analytics
- Cache savings detector (wasted writes, invalidations, unreferenced context)
- 100% offline — your data stays local
- All future updates included
Supported models
Built-in pricing for all major AI models. Costs calculated per million tokens.
