Developer Vibe-Codes a macOS App to Track All AI API Usage

A developer juggling subscriptions to multiple AI services has built UsageBoard, a macOS menu bar application that consolidates API usage tracking into a single dashboard — and the entire project was created using the increasingly popular 'vibe coding' approach. The app currently supports 5 major AI platforms and features a plugin-based architecture that makes it easy to add more services over time.

Instead of logging into multiple dashboards throughout the day to monitor token consumption and billing thresholds, UsageBoard sits quietly in the macOS menu bar and delivers real-time usage data with a single click. It is a practical example of how vibe coding — the practice of describing what you want and letting AI generate the code — is producing genuinely useful developer tools.

Key Takeaways

• UsageBoard is a free macOS menu bar app for tracking AI API usage across multiple providers
• Currently supports OpenAI Codex, Zhipu (GLM), DeepSeek, MiniMax, and Tavily
• Built entirely through vibe coding, demonstrating the maturity of AI-assisted development
• Features a plugin-based architecture allowing independent configuration per service
• Includes token usage charts for Codex (local session analysis) and Zhipu (official API)
• Supports auto-refresh, drag-and-drop plugin ordering, disk caching, and auto-updates

The Problem: Dashboard Fatigue in the Multi-Model Era

Developers in 2025 rarely rely on a single AI provider. A typical power user might use OpenAI's Codex for code generation, DeepSeek for cost-effective reasoning tasks, Zhipu's GLM models for Chinese-language processing, and specialized services like Tavily for AI-powered web search. Each of these platforms comes with its own dashboard, its own billing page, and its own usage metrics.

The creator of UsageBoard described exactly this pain point: spending time every day visiting multiple websites just to check how much quota remained. For developers running production workloads or experimenting with multiple models simultaneously, this fragmented monitoring creates real friction. Unlike consolidated cloud dashboards from AWS or Google Cloud, the AI API ecosystem lacks a unified usage layer.

This gap is what motivated the project. Rather than building a complex SaaS product, the developer chose to solve the problem locally with a lightweight macOS application — and used vibe coding to get it done fast.

Inside UsageBoard: Architecture and Features

UsageBoard is not a simple status checker. It is a surprisingly full-featured monitoring tool with thoughtful design decisions that reflect real-world usage patterns.

The app lives in the macOS menu bar, following the convention of popular utilities like iStat Menus or Bartender. Clicking the icon opens a quick preview panel showing all configured services at a glance. Users can choose between grouped display and tabbed display modes depending on how many services they track.

Core features include:

• Manual, scheduled, and per-card refresh — each plugin can have its own refresh interval
• Plugin-based architecture — new services can be added as independent plugins with their own configuration
• Disk caching by state ID — the app displays the last successful data on launch, eliminating blank screens
• Subscription tier badges — visual indicators showing which plan level you are on (displayed as rounded labels with white text on black background)
• Auto-generated settings forms — plugin configuration UI is built automatically from script metadata
• Progress bars with color coding — usage is shown as percentages or raw numbers with visual progress indicators
• Reset time display — shows when quotas reset, critical for managing monthly or daily limits
• Launch at login support — ensures monitoring starts automatically
• Drag-and-drop plugin ordering — customize the display priority of services
• Built-in update checker — supports online updates without manual reinstallation

One particularly clever detail: new plugins are disabled by default and require all mandatory parameters to be filled before activation. This prevents misconfigured plugins from throwing errors on startup. The first launch automatically installs built-in plugins to the user's plugin directory, making initial setup seamless.

Token Usage Charts: Going Beyond Simple Numbers

For 2 of the supported services — Codex and Zhipu — UsageBoard goes beyond simple quota displays and offers full token usage charts. These visualizations help developers understand consumption patterns over time, identify spikes, and plan their usage more effectively.

The implementation differs between the 2 services. Zhipu's charts pull data directly from the official API, which provides historical usage endpoints. Codex's charts, however, are built from local session analysis, parsing conversation logs stored on the user's machine. This distinction matters because not all AI providers expose usage history through their APIs — a limitation that forces creative workarounds.

Compared to tools like LiteLLM's proxy dashboard or Helicone's logging platform, UsageBoard takes a fundamentally different approach. Those tools act as middleware, sitting between your application and the AI provider. UsageBoard queries usage data directly from provider APIs or local data, requiring no changes to existing workflows. This makes it ideal for individual developers who want monitoring without infrastructure overhead.

Vibe Coding Proves Its Worth for Utility Apps

The most noteworthy aspect of this project may not be the app itself but how it was built. Vibe coding — a term popularized by Andrej Karpathy in early 2024 — describes the practice of using AI coding assistants to generate entire applications through natural language descriptions rather than manual programming.

UsageBoard demonstrates that vibe coding has matured beyond simple scripts and prototypes. The app includes sophisticated features like plugin architectures, disk caching, remote image loading with caching, auto-generated UI forms, and system-level integration (menu bar, launch at login). A year ago, building these features through vibe coding alone would have been impractical.

Several factors make macOS utility apps particularly well-suited to the vibe coding approach:

• Well-documented frameworks — SwiftUI and AppKit have extensive documentation that AI models have been trained on
• Constrained scope — menu bar apps have clear boundaries, reducing ambiguity
• Established patterns — macOS utilities follow predictable architectural patterns that AI can replicate
• Visual feedback — developers can immediately see if the generated code works correctly

This project joins a growing list of vibe-coded applications that have gained traction in developer communities, alongside tools like browser extensions, CLI utilities, and automation scripts. The trend suggests that individual developers can now ship polished, functional software at speeds that were previously impossible.

What This Means for the Developer Tool Ecosystem

UsageBoard addresses a genuine gap in the current AI development stack. As the number of AI API providers grows — with new models from Anthropic, Google, Mistral, Cohere, and dozens of Chinese AI labs — the need for unified monitoring will only increase.

For individual developers and small teams, the cost of exceeding API quotas or accidentally running up large bills is a real concern. Services like DeepSeek offer extremely competitive pricing (as low as $0.14 per million input tokens for some models), but even cheap APIs can generate surprising bills at scale. Having a persistent, always-visible usage monitor reduces the risk of billing surprises.

The plugin-based architecture also points to an interesting possibility: community-contributed plugins. If the developer open-sources the plugin format, other developers could contribute connectors for services like Anthropic's Claude API, Google's Gemini API, Replicate, Together AI, or even non-AI services like GitHub Actions minutes or Vercel bandwidth.

Looking Ahead: From Side Project to Essential Tool

UsageBoard currently supports 5 services, but its plugin architecture positions it for rapid expansion. The most obvious next additions would be Anthropic (Claude API usage), OpenAI's broader platform (beyond Codex), and Google's Vertex AI — the 3 largest Western AI API providers by developer adoption.

The broader trend this project represents is equally significant. As AI APIs become as fundamental to software development as cloud infrastructure, the tooling ecosystem around them will mature. We are already seeing this with observability platforms like Langfuse and Braintrust, cost management tools like Keywords AI, and now local utilities like UsageBoard.

For developers who want to try UsageBoard, the app runs natively on macOS and requires API keys for each service you want to monitor. All data queries go through official APIs, meaning no credentials are shared with third-party servers. The disk caching feature ensures the app remains useful even during brief network interruptions.

The project also serves as an inspiring template for other developers considering vibe coding their own utility apps. The combination of a clear problem statement, constrained scope, and well-documented platform APIs creates ideal conditions for AI-assisted development. In 2025, the best developer tools might just be the ones developers build for themselves — in a fraction of the time it used to take.