Live usage limits for Claude Code and Codex, in the Windows tray.
See a limit coming before it stops you mid-task. One icon per limit, drawn the way Core Temp shows per-core CPU values — read from the services themselves, so the numbers are not an estimate, and checking them costs no model tokens.
- 345 KB, one file
- MIT, open source
- No telemetry, one endpoint
- Claude Code + Codex
Windows 10 or 11, and the .NET 9 Desktop Runtime — not bundled, so install it first if you do not have it.
Three rules, and nothing else to learn
Watch the second number above climb past 75. Below the warning threshold a value carries its service's colour, shaded apart from the other limits of that service. Above it, severity takes over for everything — so a warning never depends on knowing which colour belongs to which service.
Colour is the second channel, never the only one: the percentage is always on the icon, the hover card names the limit, and both thresholds are yours to move.
Service colour. Terracotta for Claude, green for Codex — that is what tells the icons apart at a glance.
Amber, for every service and every limit. One colour, one meaning.
Crimson. Both thresholds are yours to move, and so are the two service colours.
Read from the services, not guessed
| Claude | The OAuth usage endpoint, using the token Claude Code already keeps on disk. |
|---|---|
| Codex | codex app-server over JSON-RPC, on the machine, no network call of our own. |
Which limits exist depends on your plan, so the icon set is discovered from your account on first run rather than hardcoded. A failed poll keeps the last numbers on screen and backs off instead of blanking the tray.
One file, no installer
Download RateTray.exe
and run it. Every release ships a SHA-256 checksum next to the binary.
Windows 11 hides new tray icons behind the arrow until you drag them onto the taskbar. That catches most people once.
- System
- Windows 10 or 11
- Runtime
- .NET 9 Desktop Runtime
- Signed in to
- Claude Code, Codex, or both
- Licence
- MIT
- Languages
- English, German
Two files read, one address contacted
Credentials are read from the files the official CLIs already maintain, and are never copied, logged, or sent anywhere else. The only destination is Anthropic's usage endpoint; Codex data never leaves your machine. There is no telemetry and no update check.
You should not take that on trust. This reads your credential files, and it was written by an AI coding agent under human direction with no independent human review — a combination worth being sceptical about. Which is the useful thing about 345 KB and 3,600 lines of C#: it is small enough to read yourself before you run it.