One click per batch
Nothing starts on its own. The window shows what’s spare now and next session, and checks the batch against it.
Pace tells you whether your current Claude Code pace will hit your plan limit before it resets — and when.
For Claude Pro & MaxmacOS 14 or laterv0.7.4Not affiliated with Anthropic

The pace bar
Every window in Pace is drawn the same way, so the question “can I keep going at this rate?” takes one glance.
The Day Ring
For every window Pace computes one number. You can only go as fast as the tightest limit allows, so the arrow follows the window with the smallest factor.
factor = sustainable ÷ current
Ease off×0.68
Ease off to 68% of your current pace, or you hit the limit before the reset. That is exactly the at risk condition.
Session (5h)A continuous ring: the session is the tight one.
Week (7d)Seven segments, one per day: the week is.
One status scale
Every window gets one of four states. The menu bar, the cards and the notifications all come from the same value, so they never disagree.

Projected below 85% at reset. Keep going.

Projected to end at 85% or more. Worth a glance.

At this rate you hit 100% before the reset, and Pace says when.

At the limit. Pace shows when you're back.
Dashboard
⌘D opens the dashboard: each window’s burn rate now, over the last hour, 5 hours, 24 hours and the window average, all against the sustainable rate, and usage over the window with the even-pace line and where this pace lands.


Projection New in 0.7
A burst, a lunch break and a heavy afternoon don’t average into one honest rate. Drag across the chart to pick the stretch that looks like what comes next, and compare four ways to read it. Try it below; in the app, pinch zooms either axis.
Drag across the chart to pick a stretch.
| Method | Rate | At reset | Band | Limit |
|---|---|---|---|---|
| Average | – | – | – | – |
| Linear trend | – | – | – | – |
| Robust trend | – | – | – | – |
| Recent-weighted | – | – | – | – |
Resets New in 0.7
When a reset is announced early, the right pace changes. Tell Pace and every forecast aims at it instead.


Spend New in 0.7
When Pace sees a session or a week ending with quota unused, it offers work that’s always worth more of: docs, reviews, tests, ideas. Pick tasks and projects, and Pace hands them to Claude Code or Codex agents in Herdr, now, at the next reset, or overnight.
Nothing starts on its own. The window shows what’s spare now and next session, and checks the batch against it.
Each job works on its own pace/* branch in a fresh worktree, never touches your changes, and never merges, pushes or deploys.
Jobs survive a relaunch; Pace tracks each one to Done, flags any that need you, and can wait for the reset or a time you choose.
What you get
Built for people who run Claude Code all day, often several agents at once.
Pace runs Claude Code's own /usage command locally ($0, no tokens spent) and listens to its status line for updates after every response.
Token counts in your local Claude Code transcripts calibrate the burn rate, so it reacts within ~30 s of a burst instead of waiting for the next reading.
Session (5h), Week (7d) and per-model week windows, each with used %, an even-pace marker, projected % at reset, time to limit and sustainable rate.
A dashboard (⌘D) with burn rate across horizons, usage over the window you can zoom and project from, activity by project, and Claude Code's own findings, like “63% of usage was at >150k context”.
A notification when you're on track to hit a limit, while there's still time to slow down. Another when a limit is reached, and when a tight window resets.
⌃⌥P opens Pace from anywhere; from there ⌘D is the dashboard and ⌘E is Spend. Launch at login. A monochrome menu bar that only takes colour when something needs your attention.
Private by construction
Pace reads your limits through Claude Code itself: its local /usage command and its status line. Claude Code makes its own request, as it would anyway. Pace only reads the result.
What’s new
Pace moves quickly. Here’s what changed, newest first.
status.json, and pace-bridge status --json prints it, so Claude Code, Codex or a script can check what's left before starting big work./usage and status line. No cookies, no tokens, nothing leaves your Mac.No. Pace reads your limits through Claude Code itself: its local /usage command and its status line. It never touches a credential and never calls an Anthropic endpoint.
No. /usage is a local command: no model turn, no tokens, $0. The status line relay only copies what Claude Code already reports after each response.
Claude Pro and Max, with Claude Code installed and signed in. Pace follows the 5-hour session, the week and the per-model week windows your plan has.
It's your pace factor: sustainable rate ÷ current rate. Level means you're on pace, tilted up means there's room to go faster, tilted down means ease off before the limit. It follows whichever limit is tightest.
Pace accepts a confirmed drop of 10 points or more as an early reset and restarts its rates from there. Add an announced reset on the dashboard and every forecast aims at it; keep your banked count there too.
No. Spend suggests work for quota that would otherwise go unused, and every batch starts with your click. Jobs run in Herdr on your Mac, each on its own pace/* branch, and never merge, push or deploy.
Settings → Disconnect puts your original status line back exactly as it was. Then quit Pace and move it to the Bin; its data lives in ~/Library/Application Support/Pace.
Pace lives in your menu bar and stays quiet until there's something worth knowing.