CodexBar: macOS Menu Bar AI Assistant Usage Tracker

CodexBar is a lightweight, free, open-source macOS menu bar application designed to address a critical pain point for developers: knowing when they're approaching rate limits on AI coding assistants. Created by Peter Steinberger, an experienced iOS developer and former PSPDFKit founder, the tool displays real-time session and weekly usage quotas for 20+ AI coding assistants (Claude, Codex, Cursor, Gemini, Copilot, Antigravity, Droid, z.ai, and others) directly in the macOS menu bar. The tagline "May your tokens never run out" captures the core problem it solves—developers using multiple AI assistants with different rate limits need constant visibility into their usage to avoid hitting limits mid-development.

The application displays a minimalist two-bar icon in the menu bar: the top bar represents the 5-hour session window, while the bottom hairline bar shows weekly usage, both with color-coded progress and reset countdowns. Clicking the menu bar icon reveals detailed usage statistics, credits (where available), and reset times for each enabled provider. The tool's architecture supports multiple authentication strategies per provider—CLI-based access, browser cookies (Safari/Chrome/Firefox), OAuth tokens, and API authentication—allowing flexibility for different user setups. Notably, CodexBar prioritizes privacy by reusing existing browser cookies rather than storing passwords, with a transparent audit trail documented in GitHub issue #12.

Recent development has been prolific, with version 0.19.0 adding subscription history charts, Alibaba Coding Plan support, and enhanced Cursor dashboard alignment. The provider architecture is modular and well-documented, making it relatively straightforward to add support for new AI assistants. Peter Steinberger has been actively iterating on the tool based on community feedback, addressing issues like performance optimization (faster JSONL scanning), UX improvements (reducing Keychain prompts), and expanding provider support. The tool remains actively maintained despite Steinberger's recent move to OpenAI in early 2026, suggesting strong commitment to the open-source project.

Privacy and security have been central concerns. When questions arose about disk access (GitHub issue #12), Steinberger explained transparently that CodexBar runs provider CLI binaries to extract usage data—the same binaries users would run in Terminal—and that these CLIs sometimes perform background operations. The tool explicitly does not store credentials and falls back to local CLI output when browser cookies are unavailable. This privacy-first design has been well-received in the developer community, though some competing tools (like ClaudeMeter) have emerged offering similar functionality.

Key Takeaways

About

Sentiment / Tone

Pragmatic and developer-focused with transparent communication about technical tradeoffs. The author presents CodexBar as a practical solution to a real problem, not overselling its capabilities. When privacy concerns arise (disk scanning), Steinberger responds with clear technical explanations rather than defensiveness. The tone across releases and documentation is informative and iterative—the author openly discusses UX decisions (submenu layout for reducing clutter), performance optimizations, and architectural refactoring. The project's messaging ("May your tokens never run out") is lighthearted yet acknowledges genuine developer frustration. Overall, the sentiment is one of honest utility: this tool does one specific job well for a specific audience without unnecessary complexity.

Related Links

Research Notes

Peter Steinberger's background significantly enhances CodexBar's credibility. With 13+ years shipping native iOS (including co-founding PSPDFKit, a major PDF library used by thousands of apps), he brings deep expertise in building reliable, privacy-conscious software for real users. His recent pivot toward AI-powered developer tools (including OpenClaw, Poltergeist, and others) reflects a broader career shift from iOS to polyglot web development while maintaining open-source principles. The fact that he continues maintaining CodexBar after joining OpenAI in February 2026 suggests genuine commitment to the open-source community rather than using it as a short-term marketing vehicle. Competitive context: ClaudeMeter and ClaudeBar are emerging alternatives, but CodexBar's multi-provider support (20+ assistants) and established community give it an advantage. The tool addresses real pain points in the AI-assisted development workflow—hitting rate limits mid-session is genuinely frustrating and breaks concentration. Community discussions on Reddit show users weighing CodexBar against alternatives, with privacy and multi-provider support as deciding factors. Privacy claims are substantiated: The transparent discussion in GitHub issue #12 about disk scanning (explaining that CodexBar runs CLI binaries, just like developers do manually) demonstrates good faith. The architecture document's explicit guardrails against identity leakage between providers shows thoughtful design. However, the tool's reliance on browser cookies does require users to trust that Safari/Chrome/Firefox's cookie storage is secure—a reasonable but not zero-risk assumption. Recent development velocity (versions 0.12→0.19 within 3 months) and feature breadth (historical pace forecasting for Codex, subscription history charts, support for emerging providers) indicate active iteration based on user needs. The detailed provider authoring guide suggests Steinberger wants community contributions, positioning the project for long-term maintenance beyond his personal capacity. One minor caveat: CodexBar is macOS-only (with CLI support for Linux), which limits its addressable market. The mentioned Windows version (Win-CodexBar) exists but isn't directly maintained by Steinberger, so Windows users get second-class support.