Agent activity and token usage, together on Mac

Your AI at work.
At a glance.

TokenTracker app icon

Bring Codex, Claude Code, Cursor, Kimi and more into one view. Check activity in your menu bar, get timely alerts, and work alongside a hippo companion.

Direct edition 1.0.7 · macOS 14+Apple silicon & Intel

No card required. No automatic charge. Keep it for CNY ¥9.9, one time.

Read the changelog
Your agents and usage, together at a glance.
Current overview: agent activity, hippo companion, usage trends and model breakdown View original

Quick tour

From first task
to final tally.

Open each step to explore real app screenshots and see how it fits your day.

01Before you start, check what is left.

Review supported quota windows and reset times together, then plan your next task.

Quotas, API balances and the usage ledger stay separate. Support varies by source.

Find your tools
Subscription screenshot: Grok and Kimi quota windows, with API balances shown separately
02While agents work, know when to step in.

Glance at activity in the menu bar. Choose alerts for completion, approval, questions or interruptions, then carry on with your work.

Notification settings shown. Enable a supported live hook and macOS notifications; available events vary by tool.

Set up notifications
Notification settings screenshot: quota alerts and four types of agent task notifications
03After the task, trace your usage.

Filter records by tool, model or project and export a CSV for a closer look. Missing prices stay unknown.

Costs are accounting references, not provider charges. Project names are hidden in public screenshots.

Try it on my Mac
Usage records screenshot with project names redacted

A tour using real screenshots, with data from the time of capture. See the FAQ for screenshot and version details.

Agent activity · Fewer window switches

Let your AI work.
Keep your own rhythm.

Write code, read docs, or move to your next task. The menu bar icon responds to activity; expand it for a closer look. The overview also shows sources and concurrent session counts.

Live signals and inferred activity are labeled separately. Sessions awaiting your response take priority in the activity view.

Combine daily tokens, estimated costs, quotas and reset countdowns. Drag to reorder them and choose the modules in your menu bar popover.

Activity detail screenshot: the live label and concurrent Codex sessions
Overview activity detail · State at capture time

Agent notifications · Know when to step in

Know when it is done.
Know when you are needed.

Spend less time checking terminals. Successfully connected agents with live hooks can send macOS notifications for the events they support.

Notification settings screenshot: quota alerts and four types of agent task notifications
Notification settings · Toggle each of four event types
01

Task finished

This round has ended. Come back to review the result.

02

Approval needed

Get a prompt when permission is needed.

03

Answer needed

Your agent has a question. Add the missing detail.

04

Failed or interrupted

Get an alert when the agent reports a problem.

Live hook integrations: Claude Code, Codex, Cursor, Grok, Kimi, OpenCode, pi and oh-my-pi. Event support varies by tool; system notification permission is required. Clicking a notification opens TokenTracker.

Set up notifications

Hippo companion · Give activity some personality

Serious about work.
Room for a little joy.

Working, waiting or resting: your hippo follows the activity. Tap the sidebar companion to interact, turn speech bubbles on or off, and choose light, dark or system appearance.

Current app artwork · State illustrations. The app animates the companion and respects Reduce Motion.

Supported AI tools

Your tools.
One clear ledger.

From Claude Code, Codex and Cursor to Kimi, WorkBuddy, CodeBuddy and Qoder. Bring your mix of global and Chinese tools into one view.

35integrations

Terminals, editors and desktop agents.
Find your everyday combination.

35 integrations. Find the tools you use.

Select a card to explore support
Claude CodeLocal usage

Reads local session records to bring Claude Code token usage into the same ledger.

CodexLocal usage

Reads local sessions and cumulative counters to organize Codex usage. Supported quotas can also be queried after the required authorization.

CursorOnline accounts

Queries Cursor account usage online on demand. Requires a connection and credentials; disabled by default. Account usage may include other devices and cannot be treated as project usage on this Mac alone.

Gemini CLILocal usage

Reads local Gemini CLI sessions and derives usage increments from cumulative counters.

OpenCodeLocal usage

Reads local OpenCode sessions and usage records to show model usage in one place.

KimiLocal usage

Reads local Kimi usage records. Supported quotas and reset times can also be queried after the required authorization.

GitHub CopilotLocal usage

Reads usage from local Copilot sessions or telemetry. The tool must generate the relevant records.

KiroLocal usage

Reads usage from local Kiro data. Available records depend on the tool version.

OpenClawLocal usage

Reads local OpenClaw session records to bring model usage into the ledger.

GrokLocal usage

Reads local Grok usage records. Supported quotas can also be queried after the required authorization.

WorkBuddyLocal usage

Reads local WorkBuddy session usage so you can review it alongside other tools.

CodeBuddyLocal usage

Reads local CodeBuddy sessions and summarizes the recorded token usage.

QoderLocal usage

Reads model usage from the local Qoder database. Qoder and Qoder CN are separate integrations.

Qoder CNLocal usage

Reads the local Qoder CN database as a separate data source from Qoder.

TRAE Work CNOnline accounts

Queries TRAE Work CN account usage online on demand. Requires a connection and credentials; disabled by default. Account-level data may include other devices.

Kilo CLILocal usage

Reads local Kilo CLI database usage separately from the Kilo Code editor extension.

Kilo CodeLocal usage

Reads task usage saved by the Kilo Code editor extension. The relevant task records must still be available.

Roo CodeLocal usage

Reads task usage saved by the Roo Code editor extension to organize model usage.

Zed AgentLocal usage

Reads the local Zed Agent session database and derives usage from recorded counters.

GooseLocal usage

Reads the local Goose session database to organize token usage from session counters.

DroidLocal usage

Reads local Droid records and derives token usage increments from cumulative counters.

piLocal usage

Reads local pi session records and also provides the compatible integration path for Dots.

oh-my-piLocal usage

Reads session records from the separate oh-my-pi directory to summarize model usage.

Prime AgentLocal usage

Reads local Prime Agent sessions and organizes the token usage recorded in them.

Every CodeLocal usage

Reads local Every Code sessions and derives usage from cumulative counters.

Hermes AgentLocal usage

Reads the local Hermes Agent database and derives usage from session counters.

Craft AgentsLocal usage

Reads records in local Craft Agents workspaces and derives usage from cumulative counters.

ReasonixLocal usage

Reads local Reasonix records and derives usage from changes in cumulative counters.

Mimo CodeLocal usage

Reads local Mimo Code database records to bring model usage into the ledger.

ZCodeLocal usage

Reads the local ZCode database to organize token usage in session records.

AnythingLLM DesktopLocal usage

Reads the local AnythingLLM Desktop database and organizes its recorded model usage.

Claude ScienceLocal usage

Reads the local Claude Science database and summarizes usage as a separate source.

DeepSeek HarnessLocal usage

Reads local DeepSeek Harness session records. This integration tracks tool usage; DeepSeek API balance is queried separately.

AntigravityEstimated usage

Estimates usage from local session text. Estimates do not enter the formal ledger and cannot serve as exact token accounting.

DotsVia pi

Uses the Pi-compatible integration path, with no independent usage scanner. Available data depends on compatible session records.

31 local usage sources · 2 online accounts · 2 estimate / compatible integrations. Quota and activity support vary by tool; access depends on version, authorization and record completeness.

About integrations & privacy

Usage ledger · See where your AI usage goes

Start with the total.
Follow each record.

Filter by time, tool, model and project, then narrow by integration, provider or subscription. Explore trends, model shares, source rankings and a heatmap, or export records to CSV.

Input, output, reasoning and cache usage are shown separately when the source provides those fields.

View usage records
Advanced filters screenshot: integration, provider, subscription and project
Records and advanced filters · Export the filtered results

Cost estimates · Understand the numbers

Costs, explained.
Gaps, visible.

Use source-reported costs when available, otherwise calculate from the price table. Unpriced records stay visible. Edit prices, opt into public price updates, and display USD or CNY.

Set a daily budget reminder to keep usage in view. Costs are estimates, separate from actual charges and subscription bills.

Overview showing cost references and records with missing prices; not a provider bill
Cost overview · Missing prices are shown explicitly

Subscriptions and quotas · Check before you start

What’s left.
When does it reset?

See quota windows and reset times for supported, connected accounts. Subscriptions, API balances and local usage stay separate. Enable notifications when quotas approach their limits.

For relay endpoints without official quota data, supported local records can still show usage. Token counts are not converted into an official balance.

Subscription screenshot: Grok and Kimi quota windows, with API balances shown separately
Snapshot at capture time · Available windows and fields vary by service

Integrations and privacy · Choose each capability

Your ledger stays local.
Connections stay in your hands.

Manage local reads, live hooks and online queries separately. The usage database stays on this Mac without conversation bodies. You can turn off storage of project names and anonymized session identifiers.

Local reads start with folder authorization. Live hooks and online queries are opt-in. Hooks add removable tool configuration; online queries use the relevant service endpoints and credentials.

Online usage, quotas, model prices, exchange rates, licensing and updates may use the network. See the privacy policy for details.

Read the privacy policy
Integration center screenshot with a private endpoint redacted and connection states unchanged
Integration center screenshot · Enabled connections still need to report successfully

Meet the maker

Independently built.
Shaped with you.

TokenTracker is maintained by an independent developer. Follow its progress, share your experience or tell me which tools you need.

Start on your Mac

Download today.
Try it for 7 days.

Current public download: 1.0.7. Requires macOS 14 or later. Choose the package for your Mac’s chip.

7 days, every featureStarts on first launch. No card required and no automatic charge.

You choose what happens nextScanning, quota refresh and export pause. Your database and settings stay. Buy and activate to continue.

Trial expiry and activation explained

Apple silicon

For Macs with Apple M-series chips

Download for Apple silicon

Intel

For Macs with Intel processors

Download for Intel

Not sure which to choose? Open the Apple menu  → About This Mac. If you see “Chip: Apple M…”, choose Apple silicon. If you see “Processor: Intel…”, choose Intel.

Screenshot and download version details in the FAQ

A little more to know

FAQ

How do the screenshots compare with the download?

The current direct download is version 1.0.7. Screenshots were taken on September 10, 2026 from development build 1.0.6 (7). The companion section uses current app artwork; interface and animation details may differ from the download.

Does it support every AI tool?

There are 35 listed integrations: 31 local usage sources, 2 online account sources and 2 estimated or compatible integrations. Usage, quota, activity and notification support differ. Explore the Supported tools list for details. Antigravity provides estimates; Dots uses Pi compatibility. Capabilities depend on tool version, permissions and available records.

Can every listed tool send task notifications?

Task notifications require an enabled, connected live hook reporting the relevant event. There are 8 hook integrations and four configurable notification types; event support varies by tool. macOS notification permission is required. Clicking a notification opens the TokenTracker main window.

What can the menu bar show?

Choose up to 3 items: daily tokens, estimated costs, five-hour or weekly quotas, reset countdowns and balances. Drag to reorder or rotate between sources. The icon responds to activity, and popover modules are configurable. Metrics may be hidden when their data is unavailable.

Does the hippo follow agent activity?

Yes. Working, waiting and resting states influence the companion, while source badges and counts help identify activity. Tap the sidebar companion to interact, hide speech if you prefer, and use Reduce Motion to limit animation. This website shows app artwork and screenshots; it does not control your agents.

Are these costs the amount I actually paid?

No. They are cost references reported by the source or calculated from usage and the price table. Subscription quotas, API balances and the usage ledger are tracked separately. Missing prices remain unknown.

Does all data stay on my Mac?

The usage database is local, but online usage, quotas, exchange rates and licensing involve network requests. The whole app is not fully offline. You choose which integrations to enable.

Will I be charged after 7 days? What happens to my data?

No automatic charge. The direct edition includes a 7-day full-feature trial from first launch, with no card required. After it expires, scanning, quota refresh and export pause and the app shows activation controls. Your local database and settings are preserved.

Purchase and activate to continue. The app attempts to backfill usage from raw logs that still exist; deleted logs cannot be recovered.

The direct edition is currently CNY ¥9.9, a one-time purchase for up to 3 Macs per key. The product listing defines the current price and entitlements. Visit Liandong Shop to Buy a license key and activate in the app. See the Terms of Service.

Should I choose the direct edition or the App Store edition?

Choose the direct edition to try it first, or the App Store edition to purchase and update through Apple. The App Store edition is a paid download, without the direct trial or a license key. Purchases and licenses are separate; check your storefront for its price and version.

View original app screenshot

Real app screenshot · Press Escape to close and return

Open image directly (new tab)

Loading original image…