usage-limit-tracker

A pi / oh-my-pi extension that shows a usage-limit bar for every subscription you are logged into — Claude (5h / 7d), ChatGPT Codex (5h / 7d), and any other OAuth provider oh-my-pi reports on.

One line per subscription, below the editor. Colors reflect risk, not raw percent:

• │ marks how much of the window has elapsed. Fill left of the tick = under pace.
• Color = risk of hitting the limit before reset. It combines how full the bucket is, your projected finish at the current rate, and how far ahead of linear pace you are. Early-window bursts are damped.
• ● chill / on track, ▲ warning / hot.
• Palette is muted by default (sage → amber → dusty red). vivid and pastel are available.

Install

oh-my-pi

omp install npm:pi-usage-limit-tracker

omp install needs bun on your PATH for npm sources. Without it, clone this repo and run omp plugin link ./usage-limit-tracker.

Or install through the omp marketplace (no bun needed):

/marketplace add mattstrayer/usage-limit-tracker
/marketplace install usage-limit-tracker@usage-limit-tracker

Pick one install path. Both register /limits, so installing both loads the extension twice.

pi

pi install npm:pi-usage-limit-tracker      # or: pi install git:github.com/mattstrayer/usage-limit-tracker@v0.1.0

Or copy index.ts + src/ into ~/.pi/agent/extensions/usage-limit-tracker/ (pi) or ~/.omp/agent/extensions/usage-limit-tracker/ (omp).

No runtime dependencies. Node 22+ (the host runs TypeScript directly).

Where the numbers come from

Host	Primary	Fill-in
oh-my-pi	Rate-limit response headers on every reply	authStorage.fetchUsageReports() — omp's own usage subsystem, covers every logged-in provider
pi	Rate-limit response headers on every reply	GET api.anthropic.com/api/oauth/usage and GET chatgpt.com/backend-api/wham/usage with the OAuth tokens pi already holds

Headers: anthropic-ratelimit-unified-{5h,7d}-{utilization,reset} and x-codex-{primary,secondary}-{used-percent,window-minutes,reset-at}.

When a provider reports several buckets, each bar gets a suffix: 7d·opus (model tier), 7d·google (family from the provider's label), or an id segment as a last resort. Buckets without a reset time show a plain gauge with no countdown; the extension never invents one.

Polling runs at start and every 5 minutes; the display re-renders every minute so countdowns move. Bars for a provider appear as soon as any data arrives. API-key logins do not expose quota, so they show nothing.

Commands

The command is /limits, not /usage — oh-my-pi already ships a built-in /usage (show / reset).

Command	Effect
/limits	Refresh now and print a plain-text breakdown with the data source
/limits toggle	Hide / show the widget
/limits palette muted|vivid|pastel	Switch colors

Env: USAGE_LIMITS_HIDDEN=1 starts hidden. USAGE_LIMITS_PALETTE=vivid sets the palette. USAGE_LIMITS_DEBUG=1 logs provider/limit ids to stderr (never tokens).

Development

npm test        # node --test, zero deps
npm run demo    # print sample lines
npm run svg     # regenerate docs/demo.svg (README image)
omp -e ./index.ts   # or: pi -e ./index.ts

Layout:

• src/smart-color.ts — risk model, palettes, HSB ramp (pure)
• src/render.ts — bar/line rendering (pure)
• src/sources/ — header parsers, Anthropic + Codex usage APIs, omp report adapter
• src/store.ts — per-provider merge
• src/extension.ts — host wiring (events, widget, timers, /limits)

Adding a provider: write a parser that returns a Subscription (src/types.ts) and call store.upsert from a header hook or the poll.

Related

• The risk-based coloring is inspired by TokenEater, a macOS menu-bar usage monitor.

MIT