GitHub Copilot usage in your menu bar: GitHub's own numbers, not a guess

Copilot's premium-request allowance is one of the least visible limits in any AI coding tool. There's a number somewhere in a GitHub billing settings page, and unless you go check it, the first sign you're close is Copilot quietly falling back to a cheaper model or refusing a premium request mid-session. Token Gauge puts that number where you're already looking, in the menu bar on Mac, the system tray on Windows, or the app window or tray on Linux, so it stops being a surprise.

How Copilot's limits work

GitHub Copilot plans come with a monthly allowance of included premium requests (GitHub calls it "monthly credits" on newer plans, "premium requests" on legacy annual ones). Every completion from a premium model draws down that allowance; once it's spent, requests either cost extra or stop, depending on your plan's overage settings. The allowance resets monthly, and GitHub tracks the balance server-side, which is the same data your Copilot dashboard in VS Code reads from.

What Token Gauge shows for Copilot

Token Gauge reads that same billing endpoint and shows GitHub's own reported percentage, not an estimate, not a token count we've converted into a guess. The headline card reads "Monthly Credits" (or "Premium Requests" on legacy plans), with a status line like "GitHub-reported: 13 of 1500 monthly credits used · Copilot Individual." Underneath, per-model rows show which models you've actually been spending your allowance on. Where GitHub's credits meter isn't exposed for a plan, Token Gauge falls back to real per-model consumption figures instead of inventing a percentage it can't back up. That's the whole "no invented numbers" rule applied to Copilot specifically.

Setup is close to zero-configuration on Mac: Token Gauge looks for a manual token first, then a Copilot editor sign-in already on your machine, then the GitHub CLI if you're signed in with gh. It looks for an editor sign-in on Windows too, but that is not yet verified on a real Windows machine; if it misses, sign in with gh or set a personal access token. If none of those are present it'll ask for a personal access token as a fallback.

Local reads, honest scope

Token Gauge reads Copilot usage from GitHub's authenticated billing API using credentials already on your system, and it never touches your repositories, prompts, or completions. That read doesn't consume any of your allowance. Over the network it makes only three kinds of calls: each provider's own usage and sign-in endpoints, one Lemon Squeezy licence check, and a check for app updates; it isn't fully offline, but your code and prompts never leave the machine because there's nowhere for that data to go.