TrackNotch

Product Introduction

1. Definition: TrackNotch is a native macOS desktop application, specifically categorized as a system monitoring and developer productivity tool. It functions as a real-time dashboard that integrates directly into the macOS menu bar and notch area.
2. Core Value Proposition: It exists to provide developers and power users with a local-first, privacy-centric, and glanceable overview of their usage and spending across multiple Large Language Model (LLM) APIs and applications, eliminating the need to manually check multiple provider dashboards.

Main Features

1. Multi-Provider LLM Usage Monitoring: TrackNotch aggregates data from diverse AI service providers. For Claude Code, it parses local JSONL session files in ~/.claude to calculate real-time context window usage. For OpenAI API and Anthropic API, it uses admin-level API keys (sk-admin-..., sk-ant-admin-...) to fetch organization-wide monthly spending data. For Cursor, it monitors local usage data. This is achieved through a combination of file system observation (FileMonitor) and secure HTTPS API calls.
2. Notch-Native & Menu Bar UI: The application's primary interface is a dynamic "pill" that resides within the physical notch area of modern MacBooks. It uses native AppKit components and Core Animation to provide a springy open animation and smooth ease-out close. On Macs without a notch, it falls back to a standard menu bar extra, ensuring universal macOS compatibility.
3. Local-First Architecture & Privacy Enforcement: All data processing occurs on-device. API keys are stored exclusively in the macOS Keychain Service, never written to disk in plain text. The app performs local file reads (scoped to specific directories like ~/.claude) and makes outbound network requests only to the user-configured provider APIs. It collects zero analytics, telemetry, or diagnostic data, with no data sent to third-party servers.
4. Context Arc Visualization: For active Claude Code sessions, TrackNotch renders a visual arc around the notch pill, dynamically filling to represent the percentage of the context window (e.g., 128K tokens) currently in use. This provides an immediate, non-textual cue for context management.
5. Rate Limit & Budget Tracking: By integrating an Anthropic OAuth token (obtained via claude setup-token), TrackNotch can display real-time 5-hour and 7-day rate limit headers from Anthropic's API responses. For OpenAI and Anthropic, configured admin keys enable live budget tracking, showing cumulative monthly spend directly in the dropdown.

Problems Solved

1. Pain Point: Context Window Management Overhead. Developers using Claude Code or similar LLM-integrated IDEs often lose track of how much context they have consumed in a long-running session, leading to unexpected truncation of earlier conversations.
2. Pain Point: Fragmented Cost Visibility. Developers and teams using multiple LLM APIs (OpenAI, Anthropic, Google Gemini) must log into separate dashboards to monitor usage and control costs, creating administrative friction and potential for budget overruns.
3. Target Audience: The primary user personas are Software Engineers, AI Researchers, and Technical Leads who regularly use LLMs like Claude Code, Cursor AI, and direct API calls from OpenAI and Anthropic in their development workflow. A secondary audience includes FinOps or Engineering Managers who need to oversee team-wide API spending.
4. Use Cases: A developer is engaged in a deep debugging session with Claude Code and uses the context arc to know when to start a new chat to maintain performance. A startup CTO uses the budget tracking feature to get a real-time, aggregated view of their team's monthly LLM API spend across providers without granting full dashboard access to every engineer.

Unique Advantages

1. Differentiation: Unlike cloud-based API dashboards (OpenAI's platform, Anthropic's Console) or generic menu bar monitors, TrackNotch is a unified, local client specifically designed for LLM workflows. It differs from proxy-based solutions by requiring no intermediate server, thus eliminating a potential point of failure, data logging, or latency.
2. Key Innovation: The deep, direct integration with Claude Code's local session files for context tracking is a unique technical approach. It bypasses the need for an API entirely for this feature, enabling functionality that is impossible for web-based dashboards. The notch-native presentation layer is also a distinct UI innovation for macOS utility apps, optimizing for the hardware of modern Apple laptops.

Frequently Asked Questions (FAQ)

1. Is TrackNotch safe to use with my API keys? Yes. TrackNotch employs macOS's built-in Keychain Services for secure credential storage. API keys are never stored in plain text on disk or transmitted to any server other than the respective provider's official API endpoint for data fetching.
2. Why does macOS Gatekeeper block TrackNotch on first launch? The current releases are built with an Apple Developer ID but are not yet notarized by Apple. This is a planned roadmap item. The standard macOS security bypass—right-clicking the app and selecting "Open"—is required only once. You can also run xattr -cr /Applications/TrackNotch.app in Terminal as an alternative.
3. What is the difference between a regular API key and an admin key for cost tracking? For OpenAI, a regular key (sk-...) only allows making API calls. An admin key (sk-admin-...) is required to access the organization's usage and billing data via the API. Similarly, Anthropic requires an organization admin key (sk-ant-admin-...) for spend tracking. These are obtained from your provider's platform settings.
4. Does TrackNotch work on Macs without a notch? Absolutely. On Macs with a traditional menu bar (e.g., iMac, Mac Studio, older MacBooks), TrackNotch automatically defaults to a clean, standard menu bar extra interface, providing all the same functionality.
5. How does TrackNotch track Claude Code usage without an internet connection? The context usage feature for Claude Code works entirely offline. It reads from the local session log files (~/.claude/**/*.jsonl) that the Claude Code CLI tool generates during normal operation. No network call to Anthropic is needed for this specific data point.