ai usage and quota tracker for Windows 11 · v0.4.1 · MIT
Windows installer · portable ZIP · setup · provider guide

track Claude Code, OpenAI Codex, Grok CLI, Google Antigravity, and API providers without opening each service. a tray meter tracks quota pressure; the native flyout brings limits, reset times, pace, balances, and what your tokens would cost at API prices together. choose Dashboard to switch to a detailed page inside the same popup. readings are marked live, cached, or unavailable, so an older result stays visibly older.

_{v0.4.1 native flyout; sample data. readings depend on your connected accounts.}

dashboard

click Dashboard at the top of the flyout, or right-click the tray icon and choose Dashboard. the popup switches to the dashboard page, with all reported provider limits, reset times, API balances/spend, and local Claude Code/Codex usage. scroll for more detail; Back, Refresh and Settings stay at the top.

local history includes 24-hour and 14-day activity, token/cache breakdowns for today, 7 and 30 days, and up to 20 models and projects from the last 7 days. only Claude Code and Codex supply this local history. these are API-price estimates, separate from subscription limits and actual API account spending. coverage notes flag compressed Codex archives that have not been indexed.

sample data rendered from the Dashboard page; no personal account or project data.

download

v0.4.1 · Python included	getting started
Windows installer	per-user install; no administrator prompt; optional start at sign-in
portable ZIP	extract and run Fleet.exe

release notes and SHA-256 checksums · latest release

Warning

binaries are unsigned; Windows SmartScreen may show a reputation warning. download from this repository's releases and check the published checksums.

to upgrade, rerun the installer or quit Fleet before replacing portable files. keep _internal beside Fleet.exe when upgrading a portable copy.

what you can track

source	available readings
local clients	Claude Code, OpenAI Codex, Grok CLI, and Google Antigravity quotas, using existing client credentials
cost estimates	what Claude Code and Codex tokens would cost at public API prices, from local logs: today, 7 and 30 days, plus burn rate. estimates only; subscriptions are not billed per token
API providers	OpenRouter, DeepSeek, Kimi/Moonshot, MiniMax, Z.AI, and LiteLLM balances or usage; OpenAI organization cost with an admin key
Claude API	free connection check for a regular key; this month's and today's (UTC) organization spend for an admin key. the included Max/Team credit balance is shown only in Claude Console, which Open billing links to
opt-in probes	rate-limit headers from OpenAI, Anthropic, and other supported APIs; real inference requests that may cost money
custom providers	map an HTTPS JSON usage endpoint for a gateway, proxy, or self-hosted service

balances stay in the flyout; repeated zero-use quota windows collapse. support and metrics vary by provider. see the provider matrix and limitations, including unsupported Gemini API keys and console-only dashboards.

quick start

1. install or extract Fleet, then launch Fleet.exe.

2. install and sign in to the local clients you use on the same Windows account. keep Claude Code on PATH for login renewal.

3. for API providers, set the matching key, then choose Refresh now from the tray menu. for example:

   setx OPENROUTER_API_KEY "your-key"

left-click the icon to toggle the flyout; right-click for Refresh now or Quit. check Windows' tray overflow if the icon is hidden.

for a Claude API account, open Settings → Claude API, choose Connect, and paste a key. Fleet keeps it in Windows Credential Manager for your Windows user, never in config or cache files; environment variables still take precedence. a regular key only lists models (free); organization spend needs an admin key.

other API usage endpoints activate when their key is available. rate-limit probes are off by default: "probe": true sends a one-token inference request on each poll, may cost money, and appears in the provider's logs.

custom endpoints and overrides go in %LOCALAPPDATA%\Fleet\providers.json (packaged) or providers.json beside fleet.py (source). keep keys in environment variables, not the config file. see the setup guide and example configuration.

privacy and troubleshooting

Fleet has no telemetry or hosted Fleet service. it reads local credentials and contacts providers directly; authentication flows may refresh those credentials. saved Claude API keys use Windows Credential Manager and admin keys are used only for read-only cost reports. runtime state lives in %LOCALAPPDATA%\Fleet\.cache (packaged) or .cache (source). an authenticated local loopback bridge lets T3 Code read supported usage; it does not expose Claude API admin keys.

for missing or stale readings, check sign-in or key setup, then Refresh now. OpenRouter management keys show account credits; regular keys use a key budget/spend fallback. see the provider guide.

report a bug with Fleet version, install type, Windows build, provider, and expected versus actual behaviour. redact diagnostics; never post tokens, credential files, .cache contents, or provider response bodies. report security issues privately through SECURITY.md.

this is an early public Windows 11 release. provider APIs and credential formats can change without notice. Fleet is independent of the listed providers; their names and artwork remain theirs. see THIRD_PARTY_NOTICES.md.

run from source

requires Windows 11, Git, and 64-bit Python 3.13 (used by CI). the flyout uses Windows' .NET Framework/WPF. configure providers as above.

git clone https://github.com/okisooo/fleet.git
cd fleet
py -3.13 -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
.\fleet-tray.cmd

diagnostics without opening the flyout:

.\.venv\Scripts\python.exe fleet.py --providers
.\.venv\Scripts\python.exe tray.py --dry

--providers lists API-key configuration; --dry polls and prints readings. for the native dashboard and key setup, install a .NET SDK and run dotnet build native\FleetPanel\FleetPanel.csproj -c Release before launching from source. without that build the Python launcher uses the compatibility flyout. the release script builds the packaged copy; see building from source. see CONTRIBUTING.md and CHANGELOG.md.

more from OKISO

Fleet is no.02 in OKISO's little tools. no.01, Backdrop, adds custom images and animated GIF wallpapers to Claude Desktop, Codex Desktop, and T3 Code, with blur and readability controls. no.03, practice room, offers short Japanese practice rounds. more projects by OKISO.