Product·Widgets & gauges

Limits without
opening the app.

Mac menu-bar gauges sit next to the clock. iOS Home and Lock Screen widgets keep rails on the glance surface. Watch complications and swipeable limit pages live on the wrist. Completion chimes respect quiet hours. Every number comes from a real poll on a real machine, or it says so.

Download for Mac Read the docs

63%

41% Wed 21:36

Lock Screen

63%

resets 2h 14m

Home

63%

41%

28%

01 · Mac menu bar

Two rails. Always in the corner.

Continuum paints compact gauges in the menu bar: session window and weekly cap for the providers you care about. Click for the full panel with every connected plan, both rails, and a reset countdown.

The rail is the signature instrument, not decoration. A horizontal track, a provider-tinted fill, a tick at the 80% mark, and a warn colour at the cap. It replaced every ring and arc in the app precisely so the same shape reads identically at 16pt in the menu bar, at widget size on a Lock Screen, and full width on the Usage tab.

Every provider you have connected gets its own gauge on the same scale, so a Claude window and a Codex window are comparable at a glance instead of two dashboards with different units. Providers you turn off disappear everywhere at once: menu bar, widgets, complications, and the Usage tab. Their history stays on disk and returns when you re-enable them. If you have been reading quota out of a terminal, the mechanics behind these numbers are the same ones covered in building a Claude Code usage monitor. Deeper spend history lives in analytics.

live poll hover detail multi-provider

UsageCodeSettings

menu bar

Claude2h 14m

63%

week

41%

Codex4h 02m

41%

week

29%

live · polledhover for detail

02 · What you can add

Four widgets, not one with settings.

They answer different questions, so they are different widgets. Pick the one that matches the glance you actually take, and put it where you already look.

01 · one provider

Provider usage

Session window and weekly cap for a single provider. Ships in Lock Screen circular, rectangular and inline families plus Home Screen small and medium, so the same data fits a corner of the Lock Screen or a Home Screen tile.

02 · the lineup

All providers

Every enabled provider stacked in one tile, in the same order the Usage tab reads. Medium shows the top four; large shows them all plus your secondary meters if you run more than one account per vendor.

03 · the money

Spend

Cost today, this week, and this month over a seven-day stacked bar. Large adds the legend and side-by-side repo and model leaderboards, so you can see which project ate the budget without opening anything.

The fourth is the wrist. Complications carry the same rail onto circular, corner, rectangular and inline watch-face slots, and a separate plan-waiting complication surfaces a session that needs a decision. If the goal is simply to stop hitting the wall mid-refactor, how the 5-hour and weekly limits actually work is worth reading once, and the five ways to check usage covers what you can do without any of this installed.

9:41𝗅𝗍𝖾 ▮▮▮

Home Screen

03 · iPhone widgets

Home Screen and Lock Screen.

Add the rails you care about. Medium Home Screen stacks for multi-provider, compact Lock Screen families for a single number, StandBy-friendly layouts on the same tokens as the app.

A widget is not its own poller. The iPhone app writes each snapshot into a shared App Group container on every poll, and every widget reads those exact bytes. That is why the Lock Screen and the app never disagree by a few percent: there is one number on the device, not three that drifted apart.

When the data behind a snapshot is unknown, the widget says stale instead of drawing a confident rail. Numbers mirror from the paired Mac or the account path, and if every host is offline they lag. They do not get invented. Sessions still open from the main iPhone workbench when you need to steer.

04 · Apple Watch

Complications and swipeable limits.

Complications keep a compact percentage on the face across four families. Inside the app, swipeable limit pages flip between providers with per-provider identity, and your page selection persists, so the wrist opens where you left it.

The watch is push-driven, not timer-driven. The phone reloads a complication when the displayed usage actually changes, rather than burning a refresh budget on a schedule that mostly renders the same number. Drop a provider on the iPhone and its complication drops too, instead of freezing on a value that will never move again.

The plan-waiting complication is the exception, and it earns it: it refreshes about once a minute while an approval is actually pending and backs off to roughly half-hourly when nothing is waiting. An approval clears off your wrist promptly instead of sitting stale for half an hour, which is the whole reason to trust it. Approve or interrupt from there and see plan & review for the full surface.

9:41

63%

complication

swipe →

codex

41%

limit pages

Settings

signals

auto-revive

keep-warm for the 5h window, disabled until a provider offers a keepalive that costs nothing

unavailable · no provider qualifies

completion chimes

SF Muni, NYC MTA, Bell, Fanfare, or a system fallback

quiet hours · default 22:00→07:00

05 · Revive & chimes

Stay quiet at night. Never spend behind your back.

Auto-revive was the feature that kept a rate-limit window from going cold. We turned it off. The only implementation we had kept the window warm by sending a tiny real prompt, which spent your quota and left throwaway conversations in your history. No provider currently offers a keepalive that costs nothing, so the control reads unavailable instead of pretending.

Polling is unaffected, and it is worth being precise about why. Gauges read the provider's own non-generative usage endpoint and rate-limit headers. They report your window; they never open a conversation to measure it. A monitor that quietly burns quota to draw a graph is worse than no monitor, which is the trap most usage trackers have to navigate. The same restraint applies to the agent itself: running Claude Code through Continuum spends exactly what your own turns spend, and nothing extra for instrumentation.

Chimes mark turn complete and attention with four packs, falling back to a system sound if the bundled assets are missing. A quiet-hours window suppresses the audible packs so the desk does not ping at 1am, and delivery also honours Do Not Disturb, a muted session, and batching. Visual state still updates the moment you open a surface.

06 · Questions

The honest fine print.

Short answers here, long answers in the docs.

Do widgets work without the Mac open?

Phone and Watch surfaces need a recent sync path (paired host or account mirror). Numbers can lag if every host is offline, and they mark themselves stale rather than inventing usage.

Can I pick which providers show in the menu bar?

Yes. Continuum focuses the compact strip on the gauges you care about; the full panel still shows the wider set. Disabling a provider outright hides it from widgets and complications too.

Is auto-revive safe for billing?

It isn't enabled for anyone, which is the safe answer. Keeping a window warm meant sending a real prompt, so it spent quota. Gauges keep polling for display and send no prompt body.

Do chimes fire during quiet hours?

No. Quiet hours suppress the audible packs, and Do Not Disturb, session mute, and batching apply on top. Visual session state still updates when you open a surface.

Why does the widget disagree with my terminal?

It shouldn't. Every widget on a device reads the same snapshot the app wrote on its last poll, so the Lock Screen, Home Screen, and app always show one number. A gap means the snapshot is stale, and the widget says so.

Is there a spend widget as well as a limit widget?

Yes. Limits and money are separate questions. The Spend widget carries cost today, this week, and this month over a seven-day chart, with repo and model leaderboards at the large size.

Limits withoutopening the app.