Not our game? Play someone else's.

The strongest free option that genuinely covers both macOS and Windows.

Zero setup: no API keys, no proxies. Works if you have run Claude Code or Codex once. Live cost figure in the menu bar or system tray, full dashboard on click.

A glanceable widget for the three main coding CLIs without any account setup.

Floating always-visible widget plus tray icon. Tray colour follows the highest visible provider: green, yellow, orange, red.

Chat-interface users rather than CLI users.

Zero setup, reads browser cookies automatically. Live usage limits for both consumer chat products in the menu bar.

Teams that want full code access, OpenTelemetry standards, and the lowest per-unit cost at very high volume.

Lightest self-host in this category. Runs in a notebook or a single Docker container. Hierarchical tracing with a turnkey setup.

Pay-as-you-go API users, and anyone with home-grown tooling other monitors cannot see.

Transparent local proxy that catches usage from any client, including custom scripts that write no standard logs. 7-day token usage chart.

The lightest native option for Claude Code, and the only free one reading server-truth limits.

Native Swift on NSStatusItem, NSPopover and SwiftUI. Roughly 3 MB, no Electron, no Node, near-zero idle overhead. Server-truth limit gauges from Claude Code's OAuth usage endpoint, so it sees usage from your other devices.

The default answer. Anyone who lives in a terminal and wants historical truth about where tokens went.

Daily, weekly, monthly, per-session and 5-hour block reports. Live monitoring mode with automatic token limit detection.

Statistical rigour rather than a live gauge. The deepest post-hoc analysis of Claude Code spend.

Daily cost broken down by token type: input, output, cache creation, cache read. Per-model statistics across all Claude models used.

The fuel gauge. Keep it in a side pane during heavy sessions so you see the limit coming.

Real-time terminal dashboard with ML-based burn-rate prediction. P90 calculator infers your actual limits from usage history.

Windows users who want to design the widget rather than accept a default.

Lives on the taskbar rather than the tray alone. Multi-monitor taskbar placement so the widget sits on the screen you prefer.

People who want charts rather than terminal tables, without leaving the editor.

Local dashboard with charts, cost estimates, session timelines and model breakdowns. Same UI available as a VS Code extension in the activity-bar sidebar.

A genuinely cross-platform tray gauge with pure Go on Linux.

Built on fyne.io/systray for a native look on all three platforms. Rolling 5-hour window progress bar.

A draggable dashboard widget rather than a tray icon.

Session and weekly limit progress bars with configurable warning thresholds. Circular countdown timers for elapsed time in the current window.

Claude subscription users wanting a portable executable.

Session usage on the 5-hour rolling window, weekly caps, extra usage and API credits. Several tray indicator styles: battery, progress bar, percentage and more.

Windows users wanting a legible glance value and matching official numbers.

Reads the same 5-hour and weekly limits that Claude Code's /usage command uses, so numbers match exactly. Remaining percentage drawn large in the icon centre, legible at 16 to 32 px.

Name collision. Four separate projects publish a claude-usage-tray, from four different GitHub accounts: Bortlesboat, apexlocal-jz, jjolmo and ksmaster03. This card links ksmaster03. Check you have the one you want.

Teams needing the same tool on mixed-OS machines including headless boxes.

Cross-platform system tray with colour-coded icon showing current percentage. GUI dashboard with gauges, stats cards and daily output charts.

Name collision. Four separate projects publish a claude-usage-tray, from four different GitHub accounts: Bortlesboat, apexlocal-jz, jjolmo and ksmaster03. We cannot tell which one the PyPI package comes from, so check the repo before installing.

Anyone who wants template control over what reaches the model.

Rust CLI, substantially faster than Node alternatives. Displays token count of the generated prompt.

Answering whether a codebase fits in a given context window before you try.

Total token count for a local or remote git repo. Breakdown by file extension and by technology.

People who want the cookie-free authentication path.

OAuth setup rather than cookie scraping. Real-time limits in the menu bar.

Codex-only users who want the smallest possible trust surface.

Uses OpenAI's documented Codex app-server protocol, asking the local Codex process for rate-limit data. Does not read .codex/auth.json, handle OAuth tokens, or send its own requests to OpenAI.

The default macOS recommendation. Widest coverage, sensible permission model, no cost.

Separate session, weekly and monthly windows tracked per provider. Two-bar menu icon: top bar is the 5-hour window, bottom is the weekly quota.

The fastest path to per-call cost visibility when you do not want to touch application code.

Zero instrumentation: swap your base URL and traffic flows through the gateway. Async OpenLLMetry mode for teams who prefer not to proxy.

Teams instrumenting their own applications who need cost attributable to a user, prompt template and model version.

Per-generation token and cost tracking with automatic capture from most integrations. Trace tree view: full span hierarchy with inputs, outputs and scores.

Enforcement rather than reporting. The only tool here that can stop spend, not just measure it.

Per-key and per-user budgets that can actually reject requests over limit, unlike pure observability tools. Cost attribution across providers through one gateway.

Counting tokens on text you cannot paste into a website.

Offline token counter shipping as web, CLI and Windows desktop builds. Zero knowledge, nothing leaves the device.

The only serious cross-platform option with history and forecasting. The right pick on Linux.

Go background daemon polling every 60 seconds into local SQLite. Browser dashboard at localhost:9211, so no desktop environment needed.

Anyone wanting to embed live quota into an editor status line, custom dashboard or automation.

Local HTTP API on 127.0.0.1:6736 so any script or status line can read quota via curl. Unified JSON format across providers, no auth handling needed by the consumer.

Windows users who want OpenUsage's local HTTP API for status lines and scripts.

Faithful Tauri port sharing the upstream codebase, only OS-specific pieces changed. Borderless transparent always-on-top panel anchored above the tray icon, draggable.

Managing tokens on the way in. Roughly 26k stars and 255k npm downloads a month, the category leader.

Packs a repo or remote GitHub URL into one file with per-file token counts. Tree-sitter compression mode cuts roughly 70% of tokens while preserving structure.

People who want a polished native app and will pay once to skip configuration.

Native macOS menu bar app, no Electron. Auto-detects installed tools, only shows tabs for what you have.

Improving how you use the tool, rather than budgeting for it. Different question from every quota monitor here.

Analyses where Claude Code makes mistakes, not just what it cost. Walks through all your instructions so you can review and share them.

Claude-only users who want always-visible monitoring plus Raycast access.

Real-time burn rate in the menu bar. Per-project cost breakdown.

Scripting token counts into shell pipelines and CI.

POSIX-style CLI that behaves like wc or grep, pipes cleanly. Exact offline tokenization for OpenAI and Gemini with embedded tokenizers.

People running agents on more than one machine who want one combined picture.

Electron desktop widget, local-first by default. Optional multi-device sync through a hub: separate process, Cloudflare Worker, or hosted inside one widget.

Optimising a specific system prompt rather than tracking spend.

Line-by-line breakdown showing exactly where a prompt's tokens go. Chars-per-token efficiency metric per line.

Prompt authors working in an editor rather than a browser tokenizer.

Token count updates live in the status bar as you type. Colour-coded usage warnings.

Seeing at a glance which directories are eating your context budget.

Desktop treemap visualisation of a codebase by token count. Also maps size, complexity, hotspots and refactor priority.

People who find the free monitors bloated and want one clean indicator.

Distinctive visual design and a compact indicator. Simple setup.

Developers who want a full call log rather than a quota gauge, without instrumenting their code.

Records every API call: tokens, latency, cost, model, full request and response. Setup is one line, changing base_url to localhost.

Windows power users who want their monitor to trigger automation, not just display numbers.

Portable single EXE, zero configuration. Event commands run a custom shell command on quota reset, threshold crossing, startup or tray double-click. Use it to push to your phone, resume an agent, start a fresh 5-hour session, or trigger any workflow.

Claude subscribers who want both limit windows visible at once in one small icon.

Dual-ring menu bar icon showing 5-hour and 7-day limits simultaneously. Colour shifts per limit: green to red for 5-hour, cyan to deep purple for 7-day.

App Store distribution, older macOS versions, and no setup acrobatics.

Mac App Store distribution for people who prefer that channel. Local-first by default.

Wayland users on tiling window managers, where tray apps are awkward.

Native Waybar module, so it belongs to your status bar rather than fighting it. 5-hour and 7-day rate limits.

Windows users wanting output attribution alongside cost.

Local-first tray app. Quota windows, token totals, API-equivalent cost estimates.

Mac users who want the same feature set as the Windows build.

Menu bar edition of the Windows app. Quota windows, token totals, API-equivalent cost estimates.

The default Windows recommendation, especially with many providers.

The widest provider coverage of any tool in this catalogue. Tauri and React shell over shared Rust provider logic.