Tripmeter Claude Code & Codex usage limits and API-rate cost in the GNOME panel

A GNOME Shell extension for Claude Code and OpenAI Codex that answers both questions from the top panel: how much of your limits is left, and where the spend actually went — what the same traffic would have cost at API rates, broken down per model, per skill, per subagent and per MCP tool.

Every other panel extension in this space stops at the first question. This one is for when “73% used” is not the thing you wanted to know.

Works with Claude Code alone, Codex alone, or both. Everything is read from the session files already on your machine; nothing is sent anywhere. Tested on GNOME 49 and 50, Wayland and X11, Arch/Manjaro, Fedora and Ubuntu.

What it shows

Limits — every rolling limit with a bar: Claude's 5-hour session, weekly and weekly-Opus quota, Codex's 5-hour and weekly quota. The same figures /usage reports inside Claude Code.
Resets — how long until each limit refills.
Pace — a marker on every bar where an even spread would be by now, so you see whether there is room to use more or when the limit runs out at the current rate.
Several accounts — each Claude Code config directory and Codex home as its own card, with a name, a colour and an on/off switch (or imported from T3 Code).
Cost at API rates — what the last 24 hours / 7 days / 30 days would have cost at API list prices, plus what prompt caching saved. A subscription is flat-rate, so this is not a bill.
Where it went — cost per model, and the share spent under each skill, subagent and MCP tool.
In the panel — Claude's 5-hour session percentage, the highest limit for other providers, the window's cost, or just the icon.

Install

Needs GNOME Shell 49 or 50. Download the zip from the latest release, then go to your Downloads folder:

cd ~/Downloads

Run:

gnome-extensions install --force tripmeter@thesalaty.github.io.shell-extension.zip

Log out and back in — Wayland cannot reload the shell in place — and enable it:

gnome-extensions enable tripmeter@thesalaty.github.io

Compared with the other usage extensions

There are a dozen of these now, so here is an honest table. Feature sets are from each project's own README, checked in August 2026 — a dash means the README does not claim it, not that the project cannot do it, and any of these may have moved on since. Corrections welcome as an issue.

	Claude	Codex	Limits	Cost at API rates	Per model / skill	GNOME
Tripmeter	yes	yes	yes	yes	yes	49–50
Brain Usage	yes	yes	yes	—	—	45–49
AI Usage Bar	yes	yes	yes	—	—	50
Claude + Codex Usage	yes	yes	yes	—	—	48–50
Claude Monitor	yes	yes	yes	—	—	45–49
AI Token Bars	yes	yes	yes	—	—	42
claude-monitor	yes	—	yes	cost + burn rate	—	48

Where the others are ahead, so you can pick properly: AI Usage Bar covers four more vendors (Z.AI, OpenRouter, DeepSeek, Kimi) and shows credit balances for the API-key ones — money, just not the cost-of-your-own-traffic kind this column means. Brain Usage ships a KDE Plasma version and low-quota notifications. Claude Monitor has a macOS menu-bar app. If limits are all you need and you are on GNOME 45–48, those are the better fit — this one starts at 49.

What is here and nowhere else is the second half of the question: cost at API list prices for the last 24h / 7d / 30d, what prompt caching saved, and the split across models, skills, subagents and MCP tools. No helper CLI, no daemon — it reads the session files directly.