shanggqm/codexU

macOS desktop widget for OpenAI Codex usage, quota tracking, token usage, and today task board

codexU – Local AI‑usage dashboard & leadership score

What it is

A native macOS (menu‑bar + desktop) and Windows x86_64 application that reads the local data files created by OpenAI Codex (and, where available, Claude Code) and shows you real‑time quota, token usage, task status and a proprietary AI leadership score.
All processing happens on the machine; no usage, thread, path or account data is sent to any external service.

Key capabilities

Feature	What you get
Quota overview	Remaining percentage and reset time for Codex’s 5‑hour and 7‑day windows (and, on macOS, Claude Code if present).
Token statistics	Daily, 7‑day and cumulative token counts, broken down by cache‑write, cache‑hit and output tokens.
AI leadership score	A 0‑100 rating (seven badge levels) derived from locally verifiable agent lifecycles, parent‑child relationships, concurrency and autonomous run evidence – not from raw token counts.
Task board	Lists today’s Codex/Claude tasks with status, workspace, timestamps and next automation run time.
Usage trends	Heat‑map of daily token usage, model‑wise area charts (top‑8 models + “other”), and estimated API‑equivalent cost in USD.
Project & tool rankings	Shows which projects, tools and “Skills” consume the most tokens/threads.
Custom UI themes	Six built‑in color palettes (e.g., “青花瓷”, “敦煌飞天”) plus community‑approved palettes, all previewable in a “Liquid Glass” palette gallery.
Menu‑bar widget	Compact runtime view with three transparency modes (minimal, classic, rich) that can display usage, remaining quota, or both.
Cross‑platform	macOS 13+ (Apple Silicon & Intel) and Windows 10/11 x86_64 builds (via Tauri).
Privacy‑first updates	Automatic check for new GitHub releases (optional) without downloading or installing anything silently.

How it works

Reads ~/.codex/ (macOS) or %USERPROFILE%\.codex\ (Windows) for SQLite usage databases, rollout JSONL files, and automation TOML files.
When Claude Code is installed, also reads ~/.claude/ transcript JSONL files and task JSON files.
Aggregates token events, infers model attribution, computes a cost estimate using the current OpenAI pricing page, and feeds the numbers into the UI.
The leadership score is calculated by the bundled ScoreModel v1.3, which only counts verifiable agent lifecycles and concurrency evidence.

Installation

macOS: Download the appropriate DMG (codexU‑<version>-mac-arm64.dmg or ‑mac‑x86_64.dmg) from the latest GitHub release, drag codexU.app to Applications, then allow the app in System Settings → Privacy & Security on first launch.
Windows: Download the MSI or NSIS installer (codexU‑<version>-windows‑x86_64.msi / .exe), run it, and follow the wizard. The Windows build currently supports only Codex data.
From source: make build (requires Xcode command‑line tools on macOS). Use make run, make install, or make release to produce DMG/installer packages.

Usage basics

Press ⌘ + U (or the configured shortcut) to toggle the main window.
The menu‑bar icon opens a Runtime menu where you can switch between Codex and Claude views, open settings, or quit.
Settings let you choose language, appearance (light/dark/auto), which metrics the menu‑bar shows, and whether to auto‑check for updates.
The AI Leadership panel shows a badge, a progress ring, and four core metrics (agents, AI hours, peak concurrency, score) for the past 28 days.

Who benefits

Developers who regularly use Codex (or Codex CLI) and want a quick glance at remaining quota and token consumption.
Teams that also run Claude Code and need a single pane of glass for both services.
Anyone on a ChatGPT Pro/Team plan who wants to monitor daily/weekly usage without opening a browser.

License

All details above are taken directly from the repository’s README; no additional features have been inferred.