What each provider exposes, and how Token Gauge reads it
Last updated 14 October 2026, against the adapters in this repository. Providers change these endpoints without notice.
Token Gauge only shows a percentage when the provider publishes one. Token counts from local logs are labelled as logs. It does not raise a limit, and it does not send your code or prompts anywhere.
Claude Code
Anthropic's usage endpoint, read with the local Claude Code login, returns the overall 5-hour window and the weekly window as percentages, plus reset times. Lifetime also shows per-model weekly windows when that payload includes them, such as a separate Fable bar. Local JSONL transcripts supply token totals. The free claude.ai plan has no usage endpoint, so there is nothing to meter. Pro, Max, Team, Enterprise and Console logins are the ones that return limits. The usage read does not consume quota. Overall 5-hour and weekly percentages, and the default 80% and 95% alerts, stay free.
Codex
Codex usage comes from the local rollout logs plus the live rate-limit endpoint the signed-in Codex CLI can reach. Where that endpoint returns a percentage, Token Gauge shows the 5-hour and weekly windows and their reset times. Token counts stay with the logs. This is part of the $11.99 lifetime licence.
Cursor
Cursor's usage API, using the editor session or a token you paste in Settings, returns monthly request usage and spend. Token Gauge shows that percentage and the reset when the response includes them. It does not invent a 5-hour window Cursor does not publish. Lifetime licence.
Copilot
GitHub's billing meter is the same monthly-credits percentage VS Code's Copilot dashboard uses, when that meter is present for the account. Token Gauge also reads real per-model spend from GitHub's report. If the allowance size is not in the response, the app shows consumption and does not invent a percentage. On macOS it can use an existing Copilot editor sign-in or the GitHub CLI. On Windows, editor sign-in discovery is not wired, so the GitHub CLI or a personal access token is the path. Lifetime licence.
Antigravity
Google does not publish a usage percentage for Antigravity. Token Gauge shows token counts from legacy Gemini CLI logs and does not draw a percent bar. The current Antigravity CLI (agy) log path is still in the works, so a machine that only has the new logs may show nothing yet. The agent is labelled Antigravity in the app. Lifetime licence.
What the menu bar adds
On macOS, the primary percentage can sit beside the other window, a minute-precise time left, or the burn pace, and at 100% the menu bar switches to the reset countdown. On Linux the same text shows as the tray label where the desktop supports one. On Windows the tray shows the gauge icon, and the tooltip lists each agent and its reset. Polling moves to every 30 seconds once any included agent reaches 80%, and back to your interval below 75%. While every live window is under 40% it waits up to three times as long, at most 10 minutes. A window already at 100% does not force fast polling; the app checks again at its reset. Lifetime can alert at every 10% instead of a custom list. These readings still come only from the sources above.