Free · v0.6.2 · signed and notarized
eaSpot
What’s left of your AI limits, right in the menu bar.
eaSpot shows how much of your Claude Code, Codex and ChatGPT limits is left, and colours each one by pace: green with room to spare, yellow when you’re ahead of it, red when it’s nearly out. Under each service it can list the models using the most tokens this week. It never logs in to anything: your tools report to it.
- macOS 14+
- Apple silicon & Intel
- Free
- MIT licence

Set it up
Install, then connect your tools.
Install with Homebrew, or download the zip and move eaSpot to Applications; zip installs run the CLI as /Applications/eaSpot.app/Contents/Helpers/easpot print-config. Open eaSpot, print the hook snippets, add them to ~/.claude/settings.json and ~/.codex/hooks.json, and approve them in Codex with /hooks. Codex numbers appear within seconds, Claude’s within a few minutes.
brew tap x/tap https://x.vibewait.ing/x/homebrew-tap.gitbrew install --cask x/tap/easpoteaspot print-configWhat makes it different
It never calls a provider
Most quota trackers sign in to each service and poll its website or API. eaSpot turns that around: Claude Code, Codex or any other tool writes a small snapshot, and eaSpot shows what’s there. It stores no accounts, cookies or API keys. Two logins of the same app? Give the second one’s hook its own source and name, and each shows as its own line.
One protocol, not endless integrations
Instead of chasing every AI provider, eaSpot defines one small JSON contract. Anything that can print JSON—a script, a CLI, or a model—can report a percentage or a count, name itself and even draw its own icon. eaSpot’s own Claude and Codex writers do exactly that; the app itself knows no provider.
Your pace at a glance
Each limit gets one colour, on the menu bar number and in the menu: green with room to spare, white on pace, yellow when you’re using it faster than it refills, red when it’s nearly out. The bar matches the number, and a mark shows where even use would be by now. If a service has several limits, Settings lets you tick the ones its numbers show, and a limit that’s nearly out joins them on its own.
Small and fast on purpose
The part that stays running is plain AppKit, and the app itself makes no network requests. The menu updates about a tenth of a second after a snapshot changes, and the hooks take a few milliseconds and no model tokens. Counting tokens by model reads only what your logs gained since the last count, and nothing at all while it’s off.
Reading the menu
One colour says where you stand.
Each bar measures the same thing as its number, and the mark on it is where even use would be by now. The line underneath gives the gap in points, like “12% under pace” or “16% over pace · runs out ~Thu”.
| Colour | What it means |
|---|---|
| green | More than 6 points under even pace: room to spare |
| white | On pace, within 6 points either way; the line still gives the small gap, like “On pace · 3% over” |
| yellow ⚠ | More than 6 points ahead of even pace: at this rate it runs out before the reset |
| red ⚠ | 10% or less left |
Tokens by model
See which models use your limits.
Turn on “Tokens by model” in Settings and each service lists its three heaviest models this week, with their tokens and share. eaSpot counts new tokens—what a model read without a cache hit, plus what it wrote—from the Claude Code and Codex logs on this Mac, and reads only what they gained since the last count. While it’s off, nothing is read. Any tool can send its own breakdown through the same protocol.
Inversion of control
The app waits to be told.
eaSpot doesn’t integrate with providers; providers integrate with it, through one contract you can read in a minute. That keeps the app small, fast and free of your logins.
| Many quota trackers | eaSpot | |
|---|---|---|
| Where the numbers come from | The app signs in to each provider and reads its dashboard or API | Your tools write a snapshot; eaSpot reads a folder |
| Logins | Cookies, API keys or OAuth tokens kept by the app | None—eaSpot never holds a login |
| Another AI service | Wait for the app to add an integration | Publish one JSON snapshot yourself |
| Staying current | Polls on a timer | Redraws the moment a tool reports; opening the menu asks Claude and Codex for fresh numbers when they’re old |
| Cost of checking | A request on every refresh | Hooks run outside the model: a few milliseconds, no tokens |
The trade-off
Passive by design, with one catch.
eaSpot waits for your tools to report instead of going out to fetch. That’s what keeps it light and free of your logins. It also means the numbers are only as fresh as the last report, which is usually fine, but not always.
| The good | The catch | |
|---|---|---|
| Passive | It never polls or signs in, so it stays light and holds none of your logins | It only knows what your tools last told it |
| Freshness | The menu changes about a tenth of a second after a tool reports | Use somewhere else, like claude.ai in a browser, the ChatGPT app or another Mac, shows up only after your next tool call or turn reports it |
| Claude | Opening the menu asks Claude for fresh numbers when they’re more than 2 minutes old | That check takes about 3 seconds |
| Codex | Numbers arrive at the end of each tool call or turn, and opening the menu asks Codex for fresh ones when they’re more than 2 minutes old, so an early reset shows before you use Codex again | That check takes about 1.5 seconds, and Codex marks the way it answers as experimental |
| Tokens by model | Turn it on in Settings and each service lists its heaviest models this week, counted from the Claude Code and Codex logs on this Mac | Work outside those logs, like claude.ai or Codex cloud tasks, isn’t counted, and the first count takes a few seconds |
| ChatGPT | ego-chat counts every GPT-6 Pro answer it gets | Messages you send in ChatGPT yourself aren’t counted, so the count is a floor |
Our protocol
Report from anything.
A snapshot is a few lines of JSON: a source, and each limit’s usage and reset time. Pipe it to easpot publish from a script, a CLI or an agent, and it shows up in the menu with the rest. A limit can be a percentage or a count, like 12 of 200 messages, and a source can name itself and bring its own icon. It can also break its usage down by model, and the menu lists each service’s heaviest models.
echo '{"source":"glm","name":"GLM","windows":[ {"id":"five_hour","used_percent":12.5, "resets_at":"2026-10-01T09:00:00Z","window_minutes":300}]}' \ | easpot publishBuilt in
Where the numbers come from.
| Where you work | How eaSpot hears about it | Delay |
|---|---|---|
| Claude Code | A Stop hook starts a quick /usage check when the numbers are older than 3 minutes | Up to about 3 minutes |
| Codex | Stop and PostToolUse hooks read the latest token counts from Codex’s own logs; opening the menu asks Codex for its live limits when they’re old | End of each tool call or turn |
| ChatGPT | ego-chat publishes its GPT-6 Pro count for ChatGPT’s week whenever it changes | Within a minute of an answer |
| Anything else | easpot publish reads a snapshot on standard input | Whenever it runs |
