Pacer

Know what Claude Code is costing you — and how close you are to your limits — right from your Mac's menu bar.

Pacer is a free, open-source macOS app that keeps an eye on your Claude Code usage: the tokens you're burning, what they cost, how close you are to your 5-hour and weekly rate limits, and where the spend is going — by project, by model, by day. It sits quietly in your menu bar, keeps itself up to date, and keeps all of your data on your own Mac.

Menu bar & click-down popover

Home-screen & Notification Center widgets

_{Share your 5-hour & 7-day burn charts as an image}

More screenshots — models, projects & collections, six-month history

_{Models — share, trend, and per-model table; group by model or class}

_{Projects & Collections — group projects, scope totals to a collection}

_{History — lifetime totals, six-month heatmap, monthly spend}

Heads up — early days. Pacer is pre-1.0 and under active development. It works and it's useful today, but expect the occasional rough edge, and your saved history may need to be rebuilt between 0.x versions.

Download & install

1. Download the latest version → (grab the Pacer-x.y.z.dmg file under "Assets").
2. Open the downloaded file and drag Pacer into your Applications folder.
3. Launch Pacer. It lives in your menu bar — look for the little gauge icon up top (there's no Dock icon). Click it to open the dashboard.
4. The first time, macOS asks permission for Pacer to read Claude Code's files. Click Allow — that's how Pacer sees your usage. Nothing leaves your Mac.

Or with Homebrew:

brew install --cask ericandrechek/tap/pacer

That's all. Pacer keeps itself up to date automatically: when a new version ships, it offers to install it for you — no re-downloading, no reinstalling.

You'll need macOS 15 (Sequoia) or newer, on either Apple Silicon or Intel. Tested on macOS 15, 26 (Tahoe) and 27. One build serves all of them: on macOS 26 and newer Pacer adopts the system's Liquid Glass design, and on macOS 15 it looks exactly as it always has.

What you get

• Rate-limit pacing. Your 5-hour, weekly, and any per-model windows — the scoped caps Anthropic reports for your account, like a weekly Fable limit — as easy-to-read pace charts. Are you ahead, on track, or about to hit the wall? Color bands (behind / on track / ahead / nearly maxed) tell you at a glance, and every window gets the same forecast: projected fill, time-to-limit, and calibrated bands. Refreshed every few minutes from Anthropic's usage data.
• Costs, your way. See spend the way Claude Code reports it, or have Pacer price it from tokens itself — switchable in Settings. Daily, monthly, and all-time totals included.
• Where it's going. Break usage down by project — grouped into your own non-destructive collections — and by model, with every Claude model in its own color and groupable by class. Drill into any single day, and see a GitHub-style activity heatmap of the last six months.
• Live "today" view. Your current burn rate plus a running "at this pace, today will end at about $X" projection.
• Multiple accounts. Signed into more than one Claude account? Every token and dollar is attributed to the account that earned it, and a toolbar menu scopes the whole app — dashboard, history, projects, models, widgets, menu bar — to one account or all of them. Rate limits always show the account you're signed into, since that's what your next message is billed against.
• Works with claude-swap. Switch accounts with cswap — or /logout, or a CLAUDE_CONFIG_DIR profile — and Pacer follows, because it watches Claude Code's own record of who is signed in rather than any one tool. It also reads cswap's usage cache instead of asking Anthropic the same question with the same token: that budget is per-token, and two clients polling it independently rate-limit each other.
• At a glance, always. A configurable menu-bar readout — an activity-ring icon, percent chips, and an explicit choice of which window drives the icon — plus home-screen-style widgets, each pickable to whichever window you care about (5-hour, weekly, or a per-model cap) for cost and pacing.
• The window stays where you put it. Park the dashboard on whichever display and Space suits you and it comes back there — across relaunches, updates and a monitor that was asleep at login. Summoning it from the menu bar takes you to it rather than dragging it to you, and a dashboard you closed stays closed.
• Optional nudges. Local notifications when you cross a rate-limit threshold (50 / 75 / 90%) — on any window, including a per-model cap — or blow past a daily spending limit you set. Off by default.
• Pace your agents, not just yourself. Pacer ships a Claude Code skill. Install it from Settings and Claude can read every window before it fans out a hundred subagents — per-model caps and your other account included — pause and checkpoint when one is nearly full, and wait out the reset rather than losing the run. One shared read serves the whole fan-out, and the skill updates when Pacer does. One optional line in your CLAUDE.md makes it the default for long runs, and because that line is a pointer rather than a copy, it never needs re-pasting.
• pacer in your terminal. pacer report shows every window for the account you're on; pacer gate and pacer wait are what the skill paces with, for your own scripts too. It reads Pacer directly, so the local API can stay off. A Homebrew install puts pacer on your PATH; with the download, Settings → Integrations → Command line does it (no password if /usr/local/bin is already yours).
• Export. Send daily totals, daily-by-model, or per-project numbers to a CSV for your own spreadsheets.

How Pacer compares

You've got options for keeping an eye on Claude Code usage — Claude Code's own /usage, the popular ccusage CLI, and several menu-bar apps — the closest being Token Pacer, a paid, closed-source namesake, plus Claude God and ccseva. They're good tools; here's the honest lay of the land.

	Pacer	Token Pacer	Claude God	ccseva	ccusage	CC /usage
Form factor	Menu-bar app	Menu-bar app	Menu-bar app	Menu-bar app	CLI	In-terminal
Native macOS (not Electron)	✓	✓	✓	– (Electron)	–	–
Always-on, glanceable	✓	✓	✓	✓	–	–
Live limit % from Anthropic's API	✓	✦	✓	✦	✦	✓
Per-model (scoped) limit windows	✓	–	–	–	–	–
Pace vs. ideal-burn line	✓	–	–	–	–	–
End-of-day spend projection	✓	✓	✓	✓	–	–
Activity heatmap (6 months)	✓	–	–	–	–	–
Per-project & per-model breakdown	✓	✓	✓	✓	✓	–
Beyond Claude Code (Codex, opencode, …)	–	✓	–	–	–	–
Home-screen / Notification Center widgets	✓	–	✓	–	–	–
Threshold / budget notifications	✓	✓	✓	✓	–	–
CSV export	✓	✓	✓	–	✓	–
ROI: cost vs. git commits	–	–	✓	–	–	–
Claude Code plugin marketplace	–	–	✓	–	–	–
Price	Free	$29 one-time	Free	Free	Free	Included
Free & open source	✓	–	✓	✓	✓	–

_{✦ estimates the windows from your local JSONL logs rather than reading Anthropic's usage API. Token Pacer is a paid, closed-source app, so its column is read from its public site rather than its source — and some of its features (CSV export, custom model pricing) are Pro-tier. Best-effort as of June 2026 — these tools all move fast, so corrections are welcome via an issue or PR.}

The gist:

• vs. /usage — Claude Code's built-in view is a snapshot you ask for in one terminal; Pacer is the always-on, zoomed-out companion that remembers every session. The speedometer in one car vs. the dashboard that logs every trip.
• vs. ccusage — a great CLI for scriptable numbers (Pacer's tests even cross-check their scanner against it); Pacer is the GUI you glance at instead of a command you re-run, and it reads your actual limit % from Anthropic rather than estimating from logs.
• vs. Claude God / ccseva — the closest rivals, and genuinely nice. Pacer leans into pacing (your windows against an ideal-burn line, "will I run out before the reset?"), native + quiet-by-default, and signed/notarized self-update; Claude God goes further on ROI/git correlation and a plugin marketplace, and ccseva on its glassy UI. Pick the one that thinks about your usage the way you do — they coexist happily.
• vs. Token Pacer — the close namesake, and the nearest thing to a direct rival: also a native, local-first macOS menu-bar tracker, and it reaches wider — covering Codex and opencode alongside Claude Code, with "next action" nudges. The trade-offs: it's $29 and closed-source, where Pacer is free and open; Pacer reads your real limit % from Anthropic and adds the ideal-burn pacing line, the 6-month heatmap, and widgets. Multi-agent and willing to pay? Fair pick. Claude Code-first and value open source? Pacer.

Privacy

Pacer is local-first. It reads only the files Claude Code already writes to your own Mac, and it sends nothing about your usage anywhere:

• ~/.claude/projects/*.jsonl — your session token usage, read on your machine.
• The Claude Code login token in your macOS Keychain — used only to ask Anthropic for your rate-limit status, the same way Claude Code itself does.

The only network connections Pacer makes are:

• api.anthropic.com — to check your rate-limit windows (~12 small requests an hour while it's running).
• github.com — to check for, download, and install app updates (at launch, then every 6 hours; installs happen in the background when automatic updates are on).

No analytics, no telemetry, no third parties. Your data stays on your Mac in ~/Library/Group Containers/…/pacer.sqlite, and it persists across app updates.

---

For developers

Pacer is SwiftUI + SwiftData + Charts, signed with Developer ID and notarized, auto-updating via Sparkle. Contributions welcome — see CONTRIBUTING.md.

Components

Component	Role
Pacer.app	Main UI + menu-bar item. Data collection (FSEvents JSONL scan + OAuth poll) runs in-process inside this binary — there is no separate daemon.
PacerWidgets	WidgetKit extension — reads the shared App Group store directly.
PacerCore	Shared Swift package — parsers, models, scan coordinator, recomputers.
Skills/pacer	The Claude Code skill, copied into Pacer.app/Contents/Resources and installed into ~/.claude/skills/ on request. ClaudeSkillInstaller re-syncs it after an app update.
PacerCLI.app	The pacer command line, at Pacer.app/Contents/Helpers/PacerCLI.app/Contents/MacOS/pacer: an app-like bundle signed into the App Group, so it can read the store with the local API off. Its logic is the PacerPace library in PacerCore.

Building from source

For everyday use, the released DMG is what you want — this section is only for hacking on Pacer.

Requirements: Xcode 26 or newer and xcodegen (brew install xcodegen). CI and releases build on macos-26 runners; this is developed against Xcode 27.

The deployment target is macOS 15.0 whichever SDK you build against, so one binary runs everywhere from Sequoia up — the SDK never decides compatibility. It does decide appearance: built against the macOS 26+ SDK the app adopts Liquid Glass on macOS 26+, and there is no opting back out (UIDesignRequiresCompatibility is ignored from the 27 SDK on). macOS 15 users are unaffected, because their AppKit has no new design to draw.

Older Xcodes still compile the project — the PacerCore tests run on macOS 15 with Xcode 16 in CI — but an app built that way renders in the old appearance and will not match a release. The Xcode project is generated from project.yml — never edit the .xcodeproj directly.

make verify       # unsigned compile-only check (no Apple account needed)
make test         # PacerCore unit + ground-truth tests
make install      # signed + notarized build → /Applications/Pacer.app
make screenshots APPROVED=1  # local preview of the README screenshots (CI renders the real ones — see docs/screenshots.md)
make help         # everything else

make verify and make test need no signing setup and are all that CI runs. To build a runnable app you need your own Apple Developer account — point the install at it with PACER_SIGN_IDENTITY="Developer ID Application: <Name> (<TEAMID>)" make install. See CONTRIBUTING → "Building and running it yourself" for the full story (and why an App Group ties signing to your Team ID).

AGENTS.md is the deep architectural guide (performance invariants, the SwiftData schema, the recomputer pattern); docs/ covers the design, perf-tuning, screenshot generation, and release process.

Releasing

Pushing a vX.Y.Z tag triggers .github/workflows/release.yml, which builds, signs, notarizes, packages a DMG, signs the Sparkle update, publishes the GitHub Release, and updates appcast.xml on the gh-pages branch. See docs/releasing.md for the secrets and the cut-a-release checklist.

License

MIT. Pacer reads data from Anthropic's Claude Code product but is not affiliated with or endorsed by Anthropic.