Free desktop app · Windows
TokenTank. A gas gauge for your AI tools.
A needle on an E-to-F dial, a live countdown to your next session reset, and your burn rate in dollars per hour — in your system tray, updated every 15 seconds. TokenTank reads the transcripts Claude Code already writes to your disk. It never connects to your account, and nothing leaves your machine.
Live demo
This is the popover. The countdown is real.
Sample data, actual interface. In the app, the same dial lives behind a tray icon that drains as you work — white when you have headroom, amber under half, vermillion when you're close to the limit.
TokenTank
Remaining usage, estimated from your local transcripts.
tank refills in 3h 12m 45s
refills in 3d 14h 6m
How it works
Local files in, fuel gauge out.
01 · Reads
Claude Code writes a transcript of every session to your own disk. TokenTank tails those files locally — the same data, no API, no account, no upload.
02 · Computes
Two windows, modeled the way Claude actually enforces them: the 5-hour session block that opens with your first message, and the weekly quota anchored to your real reset time.
03 · Warns
The tray tank drains toward E as you burn. Hover for the countdown; click for the dial, burn rate, and your plan multiple. No more guessing whether the afternoon is safe.
Calibration
Accurate without touching your account.
Anthropic doesn't publish quota numbers, so TokenTank calibrates against ground truth instead of guessing. The first-run wizard takes about a minute: pick your plan, copy the "% used" figures from the Claude app's own usage panel, and your ceilings compute from your real consumption. From then on, every genuine limit event re-teaches the gauge where the wall is.
Measured against Anthropic's official panel in our own use, the calibrated gauge matched to the percentage point: the dial read 39% of the session used, the panel said 39%. The weekly tank landed within three points.
Download
Windows now. macOS next.
One installer, about 2 MB, no dependencies. The v0.x builds are not yet code-signed, so Windows SmartScreen will warn on first run — "More info", then "Run anyway". Verify the download against the checksum if you want certainty, or build from source.
TokenTank_0.3.0_x64-setup.exeSHA-256: 18aeb10412458e5231c636f82e7bc8cf7acd86155ce5c965c4c579bac0a4ffc8
All versions on the releases page.
Weekly digest
Get the macOS build the day it ships
One email when TokenTank lands on macOS or adds a new tool (Codex and Gemini CLI are next), plus the weekly digest of what changed in agent land.
Common questions
What people ask before installing.
Does TokenTank connect to my Claude account?
No, by design. There's no login, no API call, no token read. TokenTank reads the transcript files Claude Code already writes to your own disk, does the math locally, and draws the gauge. That privacy stance is the reason the app is safe to run on a work machine and small enough to audit before you install it.
The trade-off: it can't look up your plan automatically. A 60-second setup wizard handles that instead.
How does it know my plan limits?
Anthropic doesn't publish subscription quotas as numbers, so TokenTank calibrates instead of guessing. On first run, the wizard asks which plan you're on and has you copy the "% used" figures from the Claude app's own Settings → Usage panel; your real ceilings compute from your actual consumption. After that, every time you genuinely hit a limit, the gauge re-learns your ceiling from what preceded it.
In our own testing the calibrated gauge matched Anthropic's official usage panel to the percentage point: the dial read 39% of the session used and the panel said 39%.
Is TokenTank free?
Yes. Free download, MIT-licensed, full source on GitHub. No account, no trial, no paid tier. It's published by Agent Shortlist, the publication behind the reviews and pricing data on this site.
Windows warns me when I install it. Why?
The v0.x builds aren't code-signed (signing certificates cost hundreds of dollars a year; we'll sign at v1.0), so Windows SmartScreen shows its "unrecognized app" warning. Click "More info", then "Run anyway".
If you want to verify the download first: check the SHA-256 checksum published on this page and on the GitHub release against the file you downloaded, or build from source — it's a standard Tauri project.
Which AI tools does it support?
Claude Code today, including the plan check for Pro, Max 5x, Max 20x, and API pay-as-you-go. OpenAI Codex and Gemini CLI are next on the roadmap — they write local session logs the same way, so each one is a new parser, not a new app. Windows first; the macOS build is in progress (same codebase, needs packaging and notarization).
Related
Tool
Claude usage dashboard
The CLI sibling: per-project history, plan verdicts, and a share card, from one Python file.
Guide
Claude Pro vs Max vs API
The tier-by-tier decision guide behind the plan check, with the breakeven math.
Guide
How to avoid Claude usage limits
If the tank keeps hitting E, start here — most limit pain is fixable habits.
Claude and Claude Code are trademarks of Anthropic, PBC. TokenTank is an independent open-source project by Agent Shortlist, not affiliated with or endorsed by Anthropic. See the software terms and privacy policy.