Google meters Gemini CLI per model family: Pro requests and Flash requests draw from separate daily buckets, each with its own reset time. When a bucket runs dry the CLI falls back or stops, usually mid-task. Three ways to check before that happens:
1. Inside the CLI: /stats
Type /stats in a Gemini CLI session to print the current session token count and the daily request totals the CLI has recorded. It is exact but only covers the machine and account you are on, and it costs you a terminal round-trip every time you check.
2. Your Google plan page
The Google AI plan page shows what your subscription includes. It is authoritative for entitlements but does not show live per-bucket consumption for the CLI's free tier.
3. A menu-bar meter that watches it for you
Agent Island (free, open source, MIT) reads the Gemini CLI's own local credentials and shows both buckets — Pro and Flash percentages with reset countdowns — in the Mac menu bar, next to Claude Code, Codex, Grok, and Cursor if you run those too. Session activity from ~/.gemini drives a live working/idle state on the island. Nothing is uploaded; every number is computed on your machine.
Download for macOS — sign in to Gemini CLI once (gemini in a terminal), and the row appears automatically.
FAQ
Does Gemini CLI have separate limits for Pro and Flash?
Yes. Pro and Flash requests draw from separate buckets with independent reset times, which is why a single percentage cannot describe your Gemini quota.
Can Agent Island show Gemini usage without an API key?
Yes. It reads the oauth-personal credentials the Gemini CLI already stores locally after you sign in with Google. API-key and Vertex setups are explicitly unsupported rather than shown as fake data.
Does checking usage consume quota?
The /stats command does not consume model requests, and Agent Island polls Google's quota endpoint, not the model API.
Agent Island is open source (MIT). See the changelog, the comparison hub, or the product FAQ.