Introduction
Codex Desktop users often want a quick glance at their remaining quota or recent token usage without opening a full dashboard. CodexMeter, an independent open-source project maintained by spider-man-peter-park, answers that need with a lightweight macOS menu-bar companion. Instead of burying metrics inside a web interface, the app places quota percentages, a same-day local token estimate, and an estimated response rate directly in the system menu bar. The project is hosted on GitHub under an MIT license and is not an official OpenAI product.
What Happened
The developer identified a core engineering challenge: making usage numbers visible without implying they are more authoritative than their underlying sources. CodexMeter treats three data streams as separate concerns. Quota percentages and reset windows arrive from the Codex account response and are never derived from local token counts. A separate “Today · this Mac” counter estimates input and output tokens from locally available logs for identifiable Desktop sessions and traceable subagents, while excluding known CLI and automatic review sessions. Server-reported account history appears in its own section, retaining original date labels and acknowledging that reporting can lag.
For response speed, the app measures effective tokens per second by dividing output tokens by measurable elapsed time for completed responses. It relies on structural events rather than text length, and it excludes local tool execution from timing windows. Incomplete responses or missing timing boundaries produce no sample, and the recent-response chart is intended as a local observation aid, not a model leaderboard.
Built with Swift, SwiftUI, and AppKit, the companion avoids Electron and third-party runtimes. Each metric gets its own native status item, which users can hide or reposition by holding Command. Clicking an item opens a shared detail panel that refreshes the readings. Installation requires macOS 14 or later and an already signed-in copy of Codex Desktop. Apple Silicon users can use a prebuilt archive or a custom Homebrew tap, while Intel Macs need a source build. The distributed build is ad-hoc signed rather than Apple-notarized, so users should evaluate that trust decision before installing.
Why This Matters
Separating data sources matters because an account limit, a local counter, and a response-timing estimate describe different things. CodexMeter does not convert local tokens into a remaining allowance, nor does it treat a countdown reaching zero as proof that quota has replenished. Cached or stale quota carries a marker, and unavailable data stays unavailable instead of being displayed as zero.
The local token estimate is explicitly not a billing figure or a guaranteed account-wide total across machines and cloud tasks. It excludes unsupported or unidentifiable records rather than guessing them into the total, and it avoids double-counting cached input or reasoning output. Because the app only reads appended log bytes and preserves parsing checkpoints, it does not upload conversations, project paths, or usage counters, and it runs without telemetry or a separate cloud service.
Privacy and system boundaries are also central. For live account information, CodexMeter launches the installed Codex App Server and exchanges JSON-RPC requests over its standard input/output transport. It does not submit prompts, start coding tasks, or consume reset credits. The subprocess uses the existing Codex sign-in and does not read or store authentication tokens. The app requires no Accessibility, Screen Recording, or network-interception permissions, though macOS notification permission is needed for quota and reset-credit reminders.
Key Takeaways
- CodexMeter surfaces quota, same-day local token estimates, and effective response speed through independent macOS menu-bar items.
- Quota data comes from the Codex account response, while local token counts are derived separately from available logs.
- Effective tokens per second measures completed response timing, not streaming decode speed or whole-task throughput.
- The app is built with Swift, SwiftUI, and AppKit, with no Electron dependency or third-party runtime.
- It operates in a read-only capacity: no prompts are submitted, no auth tokens are stored, and no conversation text is uploaded.
- Installation requires macOS 14 or later and a signed-in Codex Desktop; Apple Silicon users have prebuilt options, while Intel Macs require building from source.
- The distributed app is ad-hoc signed and not Apple-notarized, so users should assess trust before installation.
- For broader provider coverage and whole-turn analytics, the related CodexBar project may be worth exploring.
Conclusion
CodexMeter offers a focused, native way for macOS users to keep an eye on Codex Desktop usage without leaving the menu bar. By keeping quota, local estimates, and account history in separate views, the app avoids presenting one measurement as a substitute for another. Its read-only design and minimal permission requests reflect a deliberate effort to monitor usage without overstepping privacy boundaries. The source code, installation instructions, and full measurement limitations are available in the project repository for anyone who wants to inspect, build, or extend the tool.

Discussion
Join the conversation
Thoughtful reactions, questions, and follow-up ideas help shape the next story.