Every product, pricing, privacy, and setup answer in one place.

Tokens 4 Breakfast (T4B) is a native macOS menu bar app that tracks your AI usage, costs, and rate limits across multiple providers in one unified dashboard. Instead of checking Claude's settings page, then OpenAI's billing page, then Cursor's dashboard separately, T4B aggregates everything into a single live view that's always one click away. It also forecasts when you'll hit rate limits, tracks spending trends over time, and lets you set budget caps with alerts so you never get surprised by a bill.

Anyone who uses more than one AI tool and wants to understand what they're actually spending. If you're a developer using Claude Code and OpenAI together, a power user bouncing between Claude Pro and Cursor, or someone managing a team's AI subscriptions, T4B gives you visibility that no individual AI provider offers on its own. It's particularly valuable if you've ever been rate-limited mid-task and wished you had a warning beforehand.

Yes. T4B is a native macOS app built with Swift and SwiftUI. It runs as a lightweight menu bar utility, uses minimal CPU and memory, and supports both Apple Silicon and Intel Macs running macOS 13 (Ventura) or later. There are no current plans for Windows or Linux, though this may change based on demand.

Tokens 4 Breakfast is built in Germany with a local-first privacy model: no account, no telemetry, and usage data stored on your Mac.

T4B supports 8 providers: Claude Code, Claude Web (claude.ai subscription tracking), OpenAI, GitHub Copilot, Cursor, OpenRouter, DeepSeek, and Mistral. Claude Code connects automatically by reading local session files with no API key needed. GitHub Copilot connects via your local CLI credentials. Claude Web connects via your browser session cookie to track 5-hour and weekly rate limits in real time. All other providers connect via API key. The free plan includes 1 provider; Pro unlocks all 8.

Most Claude usage trackers (ClaudeMeter, ClaudeUsageBar, SessionWatcher, etc.) track a single provider. T4B is the only app that aggregates usage across 8 providers into one dashboard. Beyond aggregation, T4B provides features no single-provider tracker offers: velocity-based forecasting that predicts when you'll hit rate limits before it happens, per-session budget tracking (Focus Mode), cross-provider spending trends over 7 and 30 days, subscription cost aggregation, plan optimization recommendations that tell you if you should upgrade or downgrade, and a built-in privacy audit log. It's the difference between a speedometer and a full vehicle dashboard.

Focus Mode lets you set a dollar budget for a specific work session. Hit Start, pick a budget ($1, $2, $5, or custom), and T4B monitors your real-time spend across all connected providers. You'll get a notification at 80% of your budget and again when you hit 100%. It's useful when you want to timebox AI-assisted tasks without accidentally running up a large bill, especially when using API-billed providers like OpenAI or DeepSeek.

T4B tracks your Claude.ai usage velocity over rolling 40-minute windows and projects whether you'll hit your 5-hour or weekly rate limit before it resets. You'll see one of three states: "Comfortable" (green, well within limits), "Watch It" (amber, approaching the edge), or "Heads Up" (red, on pace to hit the limit with a specific time estimate). This is predictive intelligence that no AI provider surfaces natively. Instead of discovering you're rate-limited in the middle of a critical task, you get advance warning with enough time to adjust your pace.

Yes. T4B includes a subscription tracker where you can add flat-rate subscriptions like Claude Pro, ChatGPT Plus, Cursor Pro, GitHub Copilot, and any other recurring AI cost. The app shows your combined monthly bill (live API costs plus fixed subscriptions) in one place, with multi-currency support. This gives you a true picture of your total AI spend rather than just the variable API portion.

Yes. T4B analyzes your Claude Code usage over the past 30 days, groups events by 5-hour billing blocks, and counts how often you hit or approach your plan's rate limit. Based on this, it recommends whether you should upgrade (you're regularly hitting limits), downgrade (you're using a fraction of your plan), or stay put (your plan matches your usage pattern). This analysis alone can save you $80-$180 per month if you're on the wrong plan tier.

No, and this is not a vague assurance. Here is exactly what happens in the code: when T4B reads Claude Code session files, it extracts only the usage metadata object (token counts, model name, timestamp, cost). The actual message content, your prompts, and AI responses are never read, parsed, stored, or transmitted. The same applies to every other provider: T4B calls their official usage/billing API endpoints, which return aggregate token counts and costs, never conversation content. There is no technical path in the app to access your chats.

No. All usage data is stored locally on your Mac in a SQLite database at ~/.tokensforbreakfast/data.db. There is no cloud backend, no analytics service, no telemetry, and no usage data ever leaves your machine. The only network calls T4B makes are to the AI providers' own APIs (to fetch your usage metrics) and a one-time license validation call when you activate a Pro license. After activation, even the license works offline via a local Keychain cache.

T4B includes a built-in Privacy Audit Log accessible from Settings. Every single outbound network call the app makes is logged with the service name, exact endpoint URL, the reason for the call, and a timestamp. You can review this log at any time to see exactly what was sent and to whom. No other AI tracking app provides this level of transparency. The app is also a native macOS binary with no hidden background processes, sandboxed per Apple's security guidelines.

API keys and session cookies are stored in the macOS Keychain, Apple's encrypted credential store, the same system used by Safari, 1Password, and other trusted apps. They are never written to plain text files, UserDefaults, or logs. Each credential is stored under a unique Keychain key and is only ever sent to its corresponding provider's API (your OpenAI key is only sent to api.openai.com, your Claude session cookie is only sent to claude.ai, etc.).

Yes. The free plan lets you connect 1 provider and includes the full menu bar dashboard, 7-day spending chart, and subscription tracker. It's fully functional with no time limit and no account creation required. If you only use Claude Code and want to track your token usage and costs, the free plan covers that completely.

Pro unlocks all 8 providers, Focus Mode (per-session budget tracking), the Usage Forecast engine, plan optimization recommendations, cross-provider budget alerts, and priority support. It's a one-time purchase with no subscription. Pay once, use forever, including all future updates.

The Pro license is per-device. If you use multiple Macs (e.g., a work laptop and a personal desktop), you'll need a separate license for each. The free plan has no device limit.

Yes. If T4B doesn't meet your expectations, you can request a full refund within 14 days of purchase, no questions asked.

Yes. T4B reads Claude Code's local session files (JSONL format in ~/.claude/projects/) directly on your Mac. No API key, no network call, no configuration beyond enabling the provider toggle. If you use Claude Code, T4B will automatically detect and parse your sessions.

Claude Web tracking monitors your claude.ai subscription rate limits (5-hour window and weekly usage) by calling the same API that claude.ai/settings/usage uses. To set it up, you paste your sessionKey cookie from your browser's developer tools. T4B then polls every 5 minutes to keep the data fresh, auto-refreshes when you open the popover, and recovers automatically after sleep/wake or WiFi changes. Your session key is stored encrypted in the macOS Keychain.

It depends on the provider. Claude Code session files are scanned every 60 seconds. Claude Web usage is polled every 5 minutes (and immediately when you open the popover if data is older than 60 seconds). API-based providers (OpenAI, DeepSeek, Mistral, OpenRouter) are polled on their own intervals based on what their API rate limits allow. You can also hit the refresh button at any time for an instant update.

Yes. T4B detects macOS sleep/wake events and network changes automatically. When you open your laptop or switch WiFi networks, the app re-polls all providers within 30 seconds. For Claude Web specifically, a network reachability monitor detects connectivity changes and triggers an immediate refresh the moment your internet comes back, so you don't have to wait for the next polling cycle.

Yes. In Settings, you can select your exact Claude plan tier: Free, Pro, Team, Max 5x, or Max 20x. The app adjusts its token limit calculations accordingly. For Team plans specifically, T4B searches for team-specific API bucket keys that Anthropic may use for Team accounts, ensuring accurate tracking regardless of plan type. The same applies to GitHub Copilot, which supports Free, Individual, Business, and Enterprise tiers.

No. T4B is a native Swift app compiled for your architecture (Apple Silicon or Intel), not an Electron wrapper. It typically uses less than 30 MB of RAM and negligible CPU. There's no embedded browser, no JavaScript engine, and no background processes beyond lightweight polling timers. You can verify this in Activity Monitor at any time.

ClaudeMeter and ClaudeUsageBar are solid tools for tracking Claude.ai usage specifically. They show your 5-hour window and weekly limits. T4B does that too, but also tracks 7 additional providers, aggregates spend across all of them, forecasts rate limit hits with velocity-based prediction, provides Focus Mode for per-session budgets, tracks historical trends, recommends plan changes, and includes a full privacy audit log. If Claude is your only AI tool, a Claude-specific tracker may be sufficient. If you use two or more tools, T4B is the only app that gives you the full picture.

SessionWatcher focuses on Claude Code and Codex session monitoring, showing token counts and estimated costs per session. T4B covers those same tools plus 6 additional providers, adds cross-provider spend aggregation, 7-day and 30-day trend analysis, budget alerts, subscription tracking, and predictive rate limit forecasting. SessionWatcher is a focused session viewer; T4B is a comprehensive AI spending dashboard.

AgentNotch is a real-time telemetry viewer that shows you what your AI coding assistant is doing right now (file reads, tool calls, shell commands). It's great for watching active sessions unfold. T4B serves a different purpose: tracking costs, usage limits, and spending trends across multiple providers over time. AgentNotch tells you what Claude is doing right now; T4B tells you how much you've spent this week, whether you'll hit your rate limit in 2 hours, and whether you should change your plan. They're complementary tools, not competitors.

This was a known bug in earlier versions caused by a normalization issue in how the API response was parsed. It has been fixed in the latest release. Update to the newest version, and the percentages will match exactly what you see on claude.ai/settings/usage. If the issue persists after updating, toggle Claude Web off and on in Settings to force a fresh connection.

T4B automatically detects sleep/wake and network changes, and re-fetches your Claude Web data within seconds of reconnecting. If you're on an older version, update to the latest release which includes network reachability monitoring and automatic recovery. You should never need to manually toggle the provider anymore.

This happened in earlier versions when your weekly usage reset on Sunday and the Sonnet-specific bucket temporarily returned 0%. The card was hidden by a display threshold that has since been removed. Update to the latest version and the Sonnet card will always be visible alongside the weekly card, even at 0%.