JP the Cat — The cat that counts your agent tokens.

Real Claude Code dollars, cache-aware and deduplicated; Codex's official 5-hour and weekly limits, automatic; Grok Build as notional list-price dollars from local sessions; Claude's official limits too, once you opt in; and an optional GitHub Copilot CLI row (token estimate, never in the combined total). Plus a per-model provider drawer, a native macOS widget (small, medium, large), light and dark themes (Sunlit & Midnight Ledger), a configurable 7/14/30/90-day window, and an 80% usage alert with JP's meow — all local-first, over a 62-provider catalog with explicit live/next/later labels.

Free CLIs and menu-bar tools mostly show token counts and rough percentages — and several misprice cache tokens. JP starts with accurate Claude Code dollars, Codex's official limits (Claude's too, once you opt in), and honest notional Grok list-price cost, then adds a macOS widget, light and dark themes, and an 80% meow alert. GitHub Copilot stays estimate-only when you enable capture — never smuggled into the headline total.

Different job. Those are agent control panels — approve prompts, jump to terminals. JP the Cat is a cost-and-limits meter: it answers "what am I spending, and how much of the current window is used?" Keep it in your menu bar as a pill, or drop the widget on your desktop in small, medium, or large — in light or dark to match your setup.

macOS 14 (Sonoma) or newer, on an Apple Silicon Mac. Because JP reads the usage logs your tools already keep, you'll want Claude Code, Codex, and/or Grok Build installed — she's happy with any combination. GitHub Copilot CLI is optional: enable capture from Settings → Live data when you want that row. There's no account to create for the local cost engine, and Claude Code / Codex / Grok need nothing extra to start counting.

No. JP the Cat is native Swift — no Electron, no browser engine — so she stays light in the menu bar. She mostly sleeps, waking to re-read your logs only when they change (a file-system watch) or on a gentle timer, then settles back down.

Open Edit Widgets — right-click your desktop, or click the date and time in the menu bar and scroll to the bottom — then search "JP the Cat" and drop the small, medium, or large size wherever you like. JP's Settings has a shortcut that walks you through it. The widget shows your latest totals; the app keeps it fed.

Codex publishes its 5-hour and weekly limits to its own logs, so JP shows them automatically. Claude only reports those official numbers through its statusline, so JP keeps it opt-in: turn on "Claude live limits" in Settings → Live data and she appends one small, clearly-marked block to your Claude statusline that writes the official percentages to a file she reads. Your statusline output is unchanged, nothing else in your Claude config is touched, and you can turn it off anytime. Until you opt in, Claude shows real-dollar cost only — never a guessed limit.

Claude Pro and Max — JP auto-detects your tier (Max 20x and friends), so there's no plan flag to set — Codex on a paid subscription, and Grok Build under SuperGrok or X Premium+ (dollars labelled notional list price, not subscription spend). GitHub Copilot CLI is optional capture only. She reads whatever your tools already log, so you never connect an account or paste an API key for the local cost engine.

Yes, as an optional dashboard row and token-estimate widget view — never as spend. Enable Copilot CLI live data in Settings → Live data (JP only shows the row when the Copilot CLI is actually installed). It needs the real OTel file export path configured; once events land, you get a Copilot sparkline and token estimate with an estimate badge. Copilot stays out of the combined dollar total and usage alerts.

Completely. JP reads the local session logs your tools already write and never uploads your prompts, code, transcripts, or token history. The Mac app's required network calls are limited to the server-anchored trial and salted license checks; website page and performance telemetry is disclosed in the privacy policy.

Drag JP the Cat to the Trash. If you turned on Claude live limits, flip that off first (Settings → Live data) so she cleanly removes the small block she added to your Claude statusline. Her own files live in ~/Library/Application Support/JPtheCat — delete that folder to clear every last trace.

One-time, no subscription, no account. $12.99 covers one Mac with free updates; extra Macs are $9.99 each. Every install starts with a 7-day no-card trial with everything unlocked. After the trial, the live menu-bar dollar total stays free forever, right in your menu bar. A license unlocks everything you open: your 5-hour and weekly limit % right in the menu bar, the full dashboard with its per-model breakdown and detailed gauges, the desktop widget, and JP's 80% usage-limit meow.

Yes. Open Settings in JP, choose Deactivate this Mac, then unlock JP on the new Mac with the same JPC- key from your purchase email. Need more seats instead of moving one? Buy more adds Macs at $9.99 each for returning buyers.

JP updates herself in place with one click, using Sparkle — the same framework countless Mac apps use. A dot in Settings turns amber when a new build is ready; click it (or 'Check for Updates…' in JP's menu) and she installs it while keeping your settings. Updates are free.

Yes. JP reads the on-disk logs your tools already write, so cost and usage math stays local and keeps working offline. Paid features continue from the last signed offline lease during normal offline use; license changes like activation, deactivation, and buying more Macs need the network.

Claude Code is priced from the same per-model rates as your invoice, including the 5-minute vs 1-hour cache-write split and cache-read discount. Codex's 5-hour and weekly numbers come straight from its own rollout log, marked official; opt in and Claude's official limits read official too. Grok uses official xAI API list prices as notional cost on SuperGrok / Premium+ — never claimed as subscription spend. GitHub Copilot is token-estimate only. Anywhere a number is inferred rather than published, JP stamps it est.

A highly opinionated tortoiseshell with impeccable taste and zero patience for wasted tokens. She's lovable, a little bossy, and extremely picky about who she works with — which is exactly why she's so good at keeping your AI spend honest.