Tokens 4 Breakfast

AI spend and limits in the Mac menu bar.

Track token usage, spend, subscriptions, and rate limits across 11 AI providers from one private Mac menu bar app: Claude Code, Claude Web, Cursor, Codex, GitHub Copilot, Anthropic API, OpenAI, Grok, DeepSeek, Mistral, and OpenRouter. Runs locally. No account. Free plan, or Pro for a one-time $14.99.

Download free for macOS · Website · Report a bug / request a feature

  

[!NOTE]
Tokens 4 Breakfast is a commercial, closed-source macOS app. This repository hosts the public docs and issue tracker, not the app's source code. Bug reports and feature requests are welcome in Issues.

Why

If you build with AI, your spend and your limits are spread across a handful of tools, and you usually find out about either one too late:

• The bill. Claude, ChatGPT, Cursor, Copilot, and a few API keys each bill separately, and nothing adds them up until the invoices arrive.
• The limit. You hit a Claude or Codex window halfway through an agent run, with no warning.

Tokens 4 Breakfast keeps both numbers in your menu bar while you work: what your usage costs, what your subscriptions cost, and how much of each limit you have left.

What you get

• Live numbers in the menu bar. Pick what the menu bar shows: a provider's cost, all-provider spend, a subscription total, a Claude 5-hour or weekly cap, or Codex and Grok usage. Up to two extra compact metrics can sit beside it.
• Usage in real money. Claude Code and Codex usage is valued at each model's published API rate, so a flat subscription still shows what the same tokens would have cost. Models without a public rate are marked unpriced, never guessed.
• Limits before you hit them. Claude 5-hour, weekly, and per-model limits, Codex limits, and the Grok weekly limit, each with a reset time and a burn-rate forecast. Claude limits also send alerts as you approach the cap.
• Subscription tracking. Add Claude, ChatGPT, Cursor, Copilot, Grok, and other plans, and see them next to your usage-based spend.
• Focus Mode. Set a dollar budget for a work session and get warned at 80% and 100%.
• Usage Intelligence (Pro). A dashboard with Limit Runway (when you will hit each cap at your current pace), Efficiency (where a cheaper model would do the job), and Plan Value (whether your plan pays for itself), plus breakdowns by provider, model, project, and session.
• Budgets and exports (Pro). Per-provider daily and weekly budgets with alerts, CSV export, and a weekly summary.
• Token Card (Pro). A shareable image of your token count or its API-equivalent value, generated from your own data.
• Your currency. Show spend in EUR, GBP, and more, at a conversion rate you set.
• Privacy Audit Log. Every outbound request the app makes is logged in Settings, with host, reason, and time.

Supported providers

11 providers connect. 8 of them report real spend; Claude Web and Grok report live rate limits only, and DeepSeek reports account balance, so the app shows no dollar figure for those three.

OpenAI, Anthropic API, and Mistral need an organization admin key; a personal key cannot read their usage endpoints. Keys are stored in the macOS Keychain and sent only to that provider's API.

Free vs Pro

Pro includes future updates. It is licensed per Mac and comes with a 14-day money-back guarantee. If you bought Pro at the $7.99 launch price, you keep everything, including future updates.

Privacy and security

• Local storage. Usage lives in a SQLite database on your Mac. There is no Tokens 4 Breakfast account and no cloud backend.
• No prompts, ever. From Claude Code's logs the app keeps usage metadata only (token counts, model, timestamp, cost). Your prompts and responses are not stored or sent anywhere.
• No telemetry. The only network calls go to the providers you connect, a license check when you activate Pro, and the update check. Each one appears in the Privacy Audit Log.
• Credentials in the Keychain. API keys and sessions are stored in the macOS Keychain and sent only to the matching provider.
• Signed and notarized by Apple, with the hardened runtime. Built in Germany.

See SECURITY.md for how to report a vulnerability.

Install

1. Download the latest release.
2. Open the .zip and drag Tokens 4 Breakfast to /Applications.
3. Launch it, click the menu bar icon, and connect a provider in Settings.

Updating: built-in auto-update (Sparkle). The app tells you when a new version is ready to install.
Uninstalling: quit the app and move it to the Trash. Your data is in ~/Library/Application Support/Tokens4Breakfast; saved keys are in your login Keychain under com.onekapisch.tokensforbreakfast.
Requirements: macOS 14.6 Sonoma or later, Apple Silicon or Intel.

Recent releases

The latest release is 2.0.16 (19 September 2026). Highlights from the 2.0 line:

• 2.0.16: GPT-6 and Claude Fable 5.1 priced; every model rate re-checked against the providers' official pricing pages.
• 2.0.15: extra metrics in the menu bar, weekly-limit alerts at 80/95/100%, and a selectable period for the popover's headline figure.
• 2.0.14: Focus sessions back in the popover, a burn-rate forecast per limit, and early-reset tracking.
• 2.0.10: Anthropic API organization spend.
• 2.0.9: spend in your own currency, support for the standalone Copilot CLI, and Codex per-model limits.
• 2.0.6: Grok / SuperGrok weekly limit.

Full notes for each version appear in the app's What's New sheet.

How it compares

ccusage is a command-line tool for local agent usage and costs; Tokens 4 Breakfast puts that kind of visibility in a native menu bar app and adds subscriptions, budgets, and more providers. CodexBar is a free, open-source menu bar tracker for coding-provider limits; Tokens 4 Breakfast also tracks spend, subscriptions, and budgets around those limits. More side-by-side pages: tokens4breakfast.app/alternatives.

Roadmap and feedback

The roadmap follows what users ask for. Want a provider or a feature? Open an issue. Many recent features started as requests here and by email.

FAQ

Is it open source? No, it is a commercial app. This repo holds docs and the issue tracker.

Does it read my prompts? No. It keeps token counts, model names, timestamps, and cost, never conversation content.

Does it send my data anywhere? No telemetry. It only calls the providers you connect, plus the Pro license check and the update check.

Why is there no dollar figure for Claude Web, Grok, or DeepSeek? Those providers report limits or a balance, not spend, and the app does not invent one.

What about Gemini? Google does not currently expose the account usage data the app needs, so Gemini stays disabled until it does.

Can I use Pro on more than one Mac? Pro is licensed per Mac. The free plan has no device limit.

More: full FAQ

More from OneKapisch

Other apps from the same studio (onekapisch.com):

• Mac 4 Breakfast: Your one battery app that does it all.
• LUMEL: Brighter than macOS allows. Darker than its minimum.
• Clip 4 Breakfast: The clipboard manager that never slows your Mac down.
• Chime 4 Breakfast: Run-completion alerts. Open source.
• Easy Write: On-device writing tools. Open source.
• SkyLocation: Know where you are. Without internet or signal.