Home / Guides / Best AI Usage Trackers for Mac

Best AI Usage Trackers for Mac (2026)

On a Mac the choice is settled by four documented facts before taste enters it: the macOS version a tool requires, whether it runs on your architecture, which providers it covers, and what it needs access to. This page lines those up from each project's own source, with the date each source was read. It names no winner and ranks nothing.

Last updated August 20, 2026 · By Soren Starck · Sources carry their own read dates below; only sources re-read for this correction carry the update date.

How to read this page

  • Every claim carries its own source and its own read date. There is no single page-wide verification date, because the sources were read on different days and re-reading one does not re-date the others.
  • External rows are labelled “External publisher/project claim — not independently tested”. They report what each publisher or project documents about itself.
  • Our own row is labelled “First-party SessionWatcher claim — not independently tested”. SessionWatcher is our product and its row is a disclosed first-party description, not a comparison result.
  • Recommendations and inferences are labelled editorial. Anything not traceable to a dated source is our judgment and is marked as such.
  • Nothing here is a ranking. No safety ranking, no quality ranking, no “winner”. Documented minimums, coverage lists and prices move between releases: open the linked source before relying on a row.

Row by row, with sources and read dates

Each row answers the same nine questions from that tool's own documentation, and each answer links the source it was read from with the date it was read. Nothing in these rows is a test result.

ClaudeBar

External publisher/project claim — not independently tested

tddworks, open-source project

A menu-bar quota monitor across a long list of coding assistants, documented against a higher macOS floor than the rest of this page.

Documented macOS minimum
The README states macOS 15 or later. Homebrew’s own cask metadata for ClaudeBar, generated August 14, 2026, records the same macOS 15 or later requirement.ClaudeBar README (tddworks), verified August 14, 2026 · Homebrew cask metadata for ClaudeBar, verified August 14, 2026
Intel or Apple SiliconSource is silent
Not stated in the source we read. That is unknown, not a “no”.ClaudeBar README (tddworks), verified August 14, 2026
Provider scope
Described as “a macOS menu bar application that monitors AI coding assistant usage quotas”, listing Claude, Codex, Gemini, GitHub Copilot, Antigravity, Z.ai, Kimi, Kiro, Amp, OpenCode Go, Oh My Pi and Grok.ClaudeBar README (tddworks), verified August 14, 2026
Menu bar, reports, alerts
Documents session, weekly and model-specific usage percentages, colour-coded indicators, configurable auto-refresh and threshold notifications.ClaudeBar README (tddworks), verified August 14, 2026
Access and authentication
Documents that the provider tools must be installed and connected on the same Mac.ClaudeBar README (tddworks), verified August 14, 2026
Outgoing dataSource is silent
Not stated in the source we read: it documents no outgoing-data, analytics or telemetry policy, so this page makes no claim in either direction.ClaudeBar README (tddworks), verified August 14, 2026
Signing and notarization
Its installer states that distributed builds are code-signed and notarized.ClaudeBar README (tddworks), verified August 14, 2026
Install path
A Homebrew cask (claudebar) per Homebrew’s cask metadata, alongside the project’s own GitHub releases.Homebrew cask metadata for ClaudeBar, verified August 14, 2026 · ClaudeBar README (tddworks), verified August 14, 2026
Current status
Release v0.4.79 was published on August 13, 2026, and repository metadata showed the same day. A repository timestamp is a supply signal, not a page-level publication date and not quality evidence.ClaudeBar v0.4.79 release, verified August 14, 2026

TokenBar

External publisher/project claim — not independently tested

Nanako0129, open-source project

The broadest documented agent list of the projects here, with an Apple Silicon requirement and an unnotarized build.

Documented macOS minimum
The README states macOS 14 or later.TokenBar README (Nanako0129), verified August 14, 2026
Intel or Apple Silicon
The README states that Apple Silicon is required, so an Intel Mac is out of scope as documented.TokenBar README (Nanako0129), verified August 14, 2026
Provider scope
Described as a native macOS menu-bar monitor across “25+ AI coding agents”, naming Claude Code, Codex, Cursor, OpenCode and Gemini CLI among them.TokenBar README (Nanako0129), verified August 14, 2026
Menu bar, reports, alerts
Documents a menu-bar reading with no Dock icon. The source we read does not document threshold alerts.TokenBar README (Nanako0129), verified August 14, 2026
Access and authentication
Documents reading usage on-device from your local session logs, and states “no account”.TokenBar README (Nanako0129), verified August 14, 2026
Outgoing data
States “No Dock icon, no telemetry, no account”. That is the project’s own statement about its build, not an audit result.TokenBar README (Nanako0129), verified August 14, 2026
Signing and notarization
States that its release is ad-hoc signed and not notarized.TokenBar README (Nanako0129), verified August 14, 2026
Install path
A downloaded release build from the project.TokenBar v1.13.2 release, verified August 14, 2026
Current status
Release v1.13.2 was published on August 9, 2026; repository metadata was updated on August 13, 2026.TokenBar v1.13.2 release, verified August 14, 2026 · TokenBar README (Nanako0129), verified August 14, 2026

Codex Limits

External publisher/project claim — not independently tested

thrr87, open-source project

A Codex-only specialist that separates what it knows from what it estimates, and says so in its own documentation.

Documented macOS minimum
The README states macOS 14 or later.Codex Limits README (thrr87), verified August 20, 2026
Intel or Apple Silicon
The README states that stable releases include a universal app for Apple Silicon and Intel.Codex Limits README (thrr87), verified August 20, 2026
Provider scope
Codex only. The README calls the project “an independent, unofficial project”, so it is not an OpenAI product.Codex Limits README (thrr87), verified August 20, 2026
Menu bar, reports, alerts
Documents separated account facts, local facts and derived estimates, each carrying coverage and confidence. The source we read does not document a menu-bar item or threshold alerts.Codex Limits README (thrr87), verified August 20, 2026
Access and authentication
Documents a signed-in, Homebrew-managed Codex CLI on the same Mac as a prerequisite, and states that credentials are not copied or stored.Codex Limits README (thrr87), verified August 20, 2026
Outgoing data
Its privacy section states that analytics are local by default, and that a user-selected “Analyze” action discloses the content type before sending it to Codex. The outgoing path is opt-in per action as documented, not absent.Codex Limits README (thrr87), verified August 20, 2026
Signing and notarization
States that the stable release is not Developer ID signed and not notarized.Codex Limits README (thrr87), verified August 20, 2026
Install path
The project’s stable release, with the Homebrew-managed Codex CLI as the documented prerequisite. The source we read does not document a Homebrew cask for the app itself.Codex Limits README (thrr87), verified August 20, 2026
Current status
Release v0.2.8 was published on August 5, 2026; repository metadata was updated on August 13, 2026.Codex Limits v0.2.8 release, verified August 14, 2026 · Codex Limits README (thrr87), verified August 20, 2026

ClaudeMeter

External publisher/project claim — not independently tested

eddmann, open-source project

Claude.ai plan limits in the menu bar, with an authentication caveat the project raises itself.

Documented macOS minimum
The README states macOS 14 or later.ClaudeMeter README (eddmann), verified August 20, 2026
Intel or Apple SiliconSource is silent
Not stated in the source we read.ClaudeMeter README (eddmann), verified August 20, 2026
Provider scope
Claude.ai plan limits only.ClaudeMeter README (eddmann), verified August 20, 2026
Menu bar, reports, alerts
Documents tracking Claude.ai plan limits in the macOS menu bar, configurable warning and critical threshold alerts, and native macOS notifications at those thresholds.ClaudeMeter README (eddmann), verified August 20, 2026
Access and authentication
Documents an unofficial browser-authentication method, carrying the project’s own disclaimer that the method “may violate Anthropic’s Terms of Service”.ClaudeMeter README (eddmann), verified August 20, 2026
Outgoing dataSource is silent
Not stated in the source we read.ClaudeMeter README (eddmann), verified August 20, 2026
Signing and notarization
States that the app is signed and notarized by Apple.ClaudeMeter README (eddmann), verified August 20, 2026
Install path
Documents Homebrew installation and a manual GitHub Releases download that is unzipped and moved to Applications.ClaudeMeter README (eddmann), verified August 20, 2026
Current status
Repository metadata was updated on August 11, 2026, and no page-level publication date was shown.ClaudeMeter README (eddmann), verified August 20, 2026

cc-bar

External publisher/project claim — not independently tested

nanvon, open-source project

A Codex-, Claude Code-, and Antigravity-aware menu-bar app whose documented data path includes local credentials, and whose build is not notarized.

Documented macOS minimum
The project README states macOS 14 or later.cc-bar README (nanvon; Chinese), verified August 20, 2026
Intel or Apple SiliconSource is silent
Not stated in the source we read.cc-bar README (nanvon; Chinese), verified August 20, 2026
Provider scope
Documents remaining quota for Codex, Claude Code and Antigravity, plus local token and cost statistics for Codex, Claude Code and Pi.cc-bar README (nanvon; Chinese), verified August 20, 2026
Menu bar, reports, alerts
Documents a macOS menu-bar app, a floating HUD, quota timelines, and usage statistics. The source we read does not document threshold alerts.cc-bar README (nanvon; Chinese), verified August 20, 2026
Access and authentication
States that it reads Codex and Claude local credentials and local session logs.cc-bar README (nanvon; Chinese), verified August 20, 2026
Outgoing dataSource is silent
Not stated in the source we read about outgoing data. The project README is in Chinese; this page does not draw an outgoing-data conclusion from that silence.cc-bar README (nanvon; Chinese), verified August 20, 2026
Signing and notarization
States that the released app is ad-hoc signed and not notarized.cc-bar README (nanvon; Chinese), verified August 20, 2026
Install path
A released app from the project.cc-bar README (nanvon; Chinese), verified August 20, 2026
Current status
Repository metadata was updated on August 13, 2026, and no page-level publication date was shown.cc-bar README (nanvon; Chinese), verified August 20, 2026

Claude Code /usage (Anthropic, official)

External publisher/project claim — not independently tested

Anthropic, the provider itself

The one official view kept on this page: it is authoritative for Claude Code, and the caveats every third-party Claude reading inherits come from it.

Documented macOS minimum
Not a separately distributed Mac app: /usage is built into Claude Code, so Claude Code’s own requirements apply rather than a macOS minimum documented here.Anthropic, Manage costs effectively, verified August 15, 2026
Intel or Apple Silicon
Not applicable, for the same reason.Anthropic, Manage costs effectively, verified August 15, 2026
Provider scope
Claude Code only.Anthropic, Manage costs effectively, verified August 15, 2026
Menu bar, reports, alerts
Documents a Session block and a plan usage breakdown on Pro, Max, Team and Enterprise with a 24-hour and 7-day toggle, plus a continuously visible status line. No threshold alerts are documented on that page.Anthropic, Manage costs effectively, verified August 15, 2026
Access and authentication
Uses your existing Claude Code sign-in; the breakdown is computed from local session history on that machine.Anthropic, Manage costs effectively, verified August 15, 2026
Outgoing data
The page documents where the figures come from rather than an outgoing-data policy: the breakdown is approximate, computed from local session history, and excludes usage from other devices and from claude.ai.Anthropic, Manage costs effectively, verified August 15, 2026
Signing and notarization
Not applicable: a built-in command, not a separately distributed Mac app.Anthropic, Manage costs effectively, verified August 15, 2026
Install path
Already present in Claude Code; run /usage inside a session.Anthropic, Manage costs effectively, verified August 15, 2026
Current status
Read on August 15, 2026. Anthropic publishes this as current guidance, and product documentation changes without a version number.Anthropic, Manage costs effectively, verified August 15, 2026

SessionWatcher

First-party SessionWatcher claim — not independently tested

SessionWatcher, our own commercial product

Our own product, described here only as its publisher and its dated release manifest describe it. It is on this page because it is a Mac menu-bar tracker, not because anything on this page placed it above another tool: nothing here is ordered.

Documented macOS minimum
Our own site states macOS 14 (Sonoma) and later. Our own Homebrew cask records depends_on macos: :sonoma, which matches.sessionwatcher.com (our own site), verified August 14, 2026 · our own Homebrew tap README, verified August 14, 2026
Intel or Apple Silicon
Our own site states both Apple Silicon and Intel.sessionwatcher.com (our own site), verified August 14, 2026
Provider scope
Our release-pinned manifest for app 7.1.0 records 13 tools: Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Antigravity, opencode, Devin, Grok, OpenRouter, Z.ai, Kimi / Kimi Code workflows, and Qwen. Qwen is available only on the Pro tiers. When we read our own homepage on August 14, 2026 its provider strip named 12 of these and did not name Qwen; the manifest is the dated record quoted here, and the two disagree.our release-pinned product manifest, app 7.1.0 (first-party, not public), verified August 11, 2026 · sessionwatcher.com (our own site), verified August 14, 2026
Menu bar, reports, alerts
Our own site states a live macOS menu-bar reading of usage, costs and rate limits, a background refresh of about every 90 seconds with an instant refresh when the menu opens, and notifications at 70% and 90% by default, with optional every-10% and reset alerts.sessionwatcher.com (our own site), verified August 14, 2026
Access and authentication
Setup differs per tool in the same manifest: some tools use an existing local login (Devin, Grok), some require a provider API key (OpenRouter, Z.ai), Qwen uses a provider session cookie (an Alibaba console session cookie), and Kimi is recorded as mixed. There is no single credential-free setup covering all 13 tools.our release-pinned product manifest, app 7.1.0 (first-party, not public), verified August 11, 2026
Outgoing data
The same manifest records that some monitors query provider APIs from your Mac, and that license validation sends a license key and device metadata. It records that code and conversation content are not read. These are our own dated records rather than an independent audit, and this page makes no local-only claim for our product.our release-pinned product manifest, app 7.1.0 (first-party, not public), verified August 11, 2026
Signing and notarization
Our own site states the app is signed and notarized by Apple. Notarization is a distribution check, not an Apple endorsement, review or security audit.sessionwatcher.com (our own site), verified August 14, 2026
Install path
A direct download from our own site, or our Homebrew tap (brew install --cask sessionwatcher). Note the discrepancy: that tap README describes the app as a monitor for “Claude, Codex, and Cursor usage”, which is narrower than the 13 tools in the release manifest.our own Homebrew tap README, verified August 14, 2026 · sessionwatcher.com (our own site), verified August 14, 2026
Current status
App 7.1.0 in our release-pinned manifest. SessionWatcher is a commercial product; its current terms live on our own site and are deliberately not restated on this page.our release-pinned product manifest, app 7.1.0 (first-party, not public), verified August 11, 2026

What we could not verify

  • No hands-on comparison. We did not install, launch or use any tool on this page, including our own. Nothing here is a test result.
  • No security audit. No privacy, telemetry, credential or data-path statement above was audited. Every one is a publisher or project statement, ours included.
  • No accuracy benchmark. We did not compare any tool's numbers against a provider's billing or plan data, so this page makes no accuracy claim about any of them.
  • No provider-by-provider live test. Coverage lists were read, not exercised. A named provider does not prove a working connection on your Mac, your plan and your account.
  • Source gaps are marked, not filled. “Not stated in the source we read” means unknown. cc-bar's README is in Chinese; this page does not fill an outgoing-data gap from that silence. Repository timestamps and release dates show that a project is being maintained; they are not quality, popularity or market evidence.
  • No safety ranking exists on this page. Signing and notarization differences are provenance and install-friction facts. We did not audit any build, and we do not order these tools by safety, privacy or quality.

Three different numbers, three different sources

Trackers put percent and dollar signs next to figures that are not interchangeable:

  • A plan-limit percentage comes from the provider and describes your subscription window.
  • A token total can be aggregated from local session files. Anthropic states of its own /usage breakdown that the figures are approximate, computed from local session history on that machine, and do not include usage from other devices or from claude.ai. This page does not extend Anthropic's caveat to any other tool's local-file reading.
  • A dollar figure computed locally is an estimate. Anthropic notes Claude Code prices its session figure at standard list rates, so it does not reflect promotional pricing or contracted discounts and may differ from your actual bill. Your account billing view is authoritative.

Source: Anthropic, Manage costs effectively, verified August 15, 2026.

How we would narrow this down

Editorial judgment — our reading of the sources above, not a test result

Start by eliminating on the two facts you cannot negotiate with: your macOS version and your Mac's architecture. That alone removes tools from the list before any feature comparison. Then match provider scope to the tools you actually pay for, rather than to the largest advertised count — a bigger list does not mean your provider is in it. Then read the access and outgoing-data rows, because a browser-authentication method, a provider API key, a session cookie and a local log read are different commitments, and only you can decide which you are willing to make. Signing and notarization come last, as an install-friction question. If two candidates survive all of that, the honest next step is to try them yourself: this page cannot make that call for you, because it did not test them.

When SessionWatcher is not the right answer

Editorial judgment — our reading of the sources above, not a test result

SessionWatcher is macOS-only and commercial, so it is the wrong answer for Windows or Linux, and Anthropic's free built-in /usage is enough if Claude Code is all you track and you only check in occasionally. If your stack is a single provider, a specialist such as Codex Limits for Codex may fit your question more precisely than a multi-tool menu bar. If your real question is account billing or team administration, a vendor dashboard answers it and none of these menu-bar tools do.

Frequently asked questions

What is the best AI usage tracker for Mac?

This page does not name one, and it does not rank these tools by quality, privacy or safety. We did not install, test, benchmark or audit any of them, including our own, so a ranking would not be supportable. What it does instead is line up each project’s own documented answers to nine questions that decide the choice on a Mac: the macOS minimum, whether Intel Macs are covered, which providers are in scope, what appears in the menu bar and whether alerts exist, what it needs access to, what it says leaves your Mac, whether the build is signed and notarized, how it installs, and how current it is. Pick the tool whose documented answers fit your Mac and your providers, and open its source before relying on it.

Which macOS versions and which Macs do these trackers document?

They differ, and the difference disqualifies tools outright. ClaudeBar documents macOS 15 or later, so a Mac on macOS 14 is out of scope as documented. TokenBar documents macOS 14 or later but requires Apple Silicon, so an Intel Mac is out of scope as documented. Codex Limits documents a universal stable release for Apple Silicon and Intel. ClaudeMeter and cc-bar do not state whether Intel Macs are supported. SessionWatcher is ours: our own site describes macOS 14 (Sonoma) and later on both Apple Silicon and Intel, which is a first-party claim we did not independently test. All of these were read on the dates shown beside each row, and minimums move between releases.

Do these Mac trackers only read local files?

No, and treating them as interchangeable on that point would be a mistake. The documented data paths differ materially. TokenBar states it reads on-device from local session logs and states “no telemetry, no account”. cc-bar states it reads local credentials and local session logs for Codex and Claude. Codex Limits states analytics are local by default and that a user-selected “Analyze” action discloses the content type before sending it to Codex. ClaudeMeter documents an unofficial browser-authentication method and carries its own disclaimer that the method may violate Anthropic’s Terms of Service. For SessionWatcher, our own release manifest records that some monitors query provider APIs from your Mac and that license validation sends a license key and device metadata, so we make no local-only claim for it either. Every one of those is a publisher or project statement, not an audit result.

Do these trackers show exactly the same numbers?

No, and you should not expect them to. A plan-limit percentage read from a provider, a token total aggregated from local session files, and a locally computed dollar estimate are three different figures. Anthropic notes that its own /usage breakdown figures are approximate and computed from local session history on that machine, so usage from other devices or claude.ai is not included, and that the session dollar figure is computed at standard list rates and may differ from your actual bill. Your account billing view is what settles what you pay. We did not compare any tool’s numbers against a provider’s billing data, so this page makes no accuracy claim about any of them.

First-party SessionWatcher claim — not independently tested

One macOS menu bar for the thirteen tools in our release manifest.

SessionWatcher is our own commercial product, described here as our own site and dated release manifest describe it: a signed and notarized menu-bar app for macOS 14 (Sonoma) and later, on Apple Silicon and Intel. Setup differs per tool, Qwen is Pro-only and uses a provider session cookie, some monitors query provider APIs from your Mac, and license validation sends a license key and device metadata. Treat this as a disclosed first-party pitch, not the verdict of this comparison — there is no verdict.

Download for Mac

Product facts from our own site, read August 14, 2026, and our release-pinned manifest for app 7.1.0, verified August 11, 2026.

Related single-tool guides