A tiny KDE Plasma tray utility for Claude Code and OpenAI Codex quota windows. It reads the sessions already maintained by the two CLIs, fetches their usage endpoints every two minutes, and keeps all data in memory. Five-hour, daily, weekly, and model-specific Codex lanes are supported whenever the provider reports them. There is no account setup, database, analytics, helper daemon, or browser-cookie access.
The popup keeps the important details within reach: live quota windows and reset times, plan status, reset credits, local cost and token totals, per-model breakdowns, selectable daily history, and a compact activity grid. Screenshots use redacted local usage values.
Requirements: Linux, Python 3.11+, PyQt6, and KDE Plasma (tested on Plasma 6 / Wayland).
./ai_usage_bar.pyClick the tray meter for the compact details popup. The fixed-size provider switcher opens on Claude and switches to Codex without moving or resizing the popup; right-click the tray icon for refresh and quit. A runtime lock ensures only one copy can run per desktop session.
./install.sh
./install-codexbar-helper.sh # optional exact local cost/model breakdowns
ai-usage-barThe installer copies the one-file application under ~/.local/share, adds a launcher in
~/.local/bin, and enables a standard XDG autostart entry. Run ./uninstall.sh to remove it.
The optional official CodexBar CLI helper is invoked only for local cost scans and exits afterward;
it is not another resident service. Cost results are cached for ten minutes.
The app reads ~/.claude/.credentials.json and ~/.codex/auth.json without modifying them. Tokens are
used only as authorization headers for api.anthropic.com/api/oauth/usage and
chatgpt.com/backend-api/wham/usage. They are never logged or stored by this app. If a session expires,
sign in again with claude or codex login; the next refresh picks up the new credentials.
python3 -m unittest discover -v