← IOKIGopen source · macOS / windows / linux · MIT
Tokenrunway icon — a pale runway stretching away under an amber gauge arc, on a navy rounded square.cleared for takeoff
v0.1.0 · free

Tokenrunway

How much AI runway do you have left?

A menubar and system-tray app that shows every AI subscription and API account in one place: how much of each allowance is left, when it resets, and whether you'll run out before it does.

It warns you before you hit the wall, writes the request for more to your admin, and reads your agents' own logs to show what the tokens went on. No account, no server, no telemetry.

The Tokenrunway popover, Now tab, in dark mode. Shortest runway 1h 26m. A Claude work (Enterprise) card shows its 5-hour window at 78% with 'Out at 13:05, 1h 15m before reset' and a Request more button, the weekly window at 64% on track, and Weekly Opus at 83%, tight. A ChatGPT personal (Pro) card follows below.
now · sorted by urgency

what it does

ONE GLANCE

Every account in one popover, sorted by whichever is closest to running out. The menubar can show the shortest runway as a number, or nothing at all.

EVERY WINDOW

5-hour, weekly and monthly windows, each with how much is used and when it resets, in your local time.

A FORECAST

A tick on each bar marks where you'd be at an even pace, and a hatched projection shows where you're heading. It says plainly if you'll run out before the reset, and when.

A WARNING FIRST

A notification at 75% and 90%, and one when your pace says you'll run out before the reset. Pause them for an hour, or until you're ready.

ASK FOR MORE

One click writes the request to your admin, with the numbers and the reset time filled in, by email, Slack or Teams. Before you're stuck, not after.

SEVERAL ACCOUNTS

Work and personal, side by side, with several accounts per provider. Each one keeps its own admin contact and its own alerts.

reading the bars

Find out at 2 pm, not at 4:55.

The solid part of each bar is what you've used. The white tick is where you'd be if you spread the window evenly. The hatched part is where your current pace takes you by the reset.

When the projection runs past the end, Tokenrunway tells you the time you'll run out and how long that leaves before the reset, and offers to ask for more while there's still time to get it. Until it has a couple of weeks of history, it says the forecast is rough.

5-hour · 78%out at 13:05, 1h 15m before reset
weekly opus · 83%tight · ~97% at reset
weekly · 64%on track
used even pace projected at reset

accounts · who it watches

CLAUDEPro · Max · Team · Enterprise5-hour and weekly windows, per-model weekly limits and their resets, read with your Claude Code login.SUPPORTED
CHATGPT / CODEXPlus · Pro · Business · EnterpriseCodex 5-hour and weekly windows and credits, from Codex's own session files or your Codex login.SUPPORTED
AWS BEDROCKOn-demandMonth-to-date Bedrock spend against your AWS Budget, through the AWS CLI and your existing profile.SUPPORTED
ANTHROPIC APIOrganisationCost per model and day from the Admin API, with an admin key you paste in.SUPPORTED
OPENAI APIOrganisationCost per model and day from the organisation usage API, with an admin key.SUPPORTED
GITHUB COPILOTPro · Pro+ · FreePremium requests left this month, and Free-tier chat and completions, with your GitHub CLI login.SUPPORTED
CURSORPro · BusinessPlan usage and on-demand spend this billing cycle. Sign in inside Tokenrunway.SUPPORTED
GEMINIGoogleDaily quota per model, read with your Gemini CLI login.EARLY
GROKSuperGrok · xAI APIQuery limits on grok.com (sign in inside Tokenrunway), or an xAI API key.EARLY

Several accounts per provider. Claude Code running on Bedrock still shows up in your stats, even without AWS access on your laptop.

stats · where the tokens went

Your agents already keep a record. Claude Code, Codex and opencode write a log of every turn to your computer. Tokenrunway reads them as they grow and turns them into numbers no provider dashboard shows you.

SPENDPer day, week, month and lifetime, at API prices, next to what you were actually billed where the provider says.
MODELSWhich models you use most, and what each one cost.
PROJECTSWhich repositories the tokens went to.
KIND OF WORKBuild, bug fix, discovery, tests, refactoring, review, docs. Worked out on your computer from the tools the agent used, not from reading your prompts back to you.
BY HARNESSClaude Code, Codex and opencode side by side: spend, sessions, cache hits and each one's favourite model.
The Stats tab for the last 7 days: $412.80 at API prices against $46.20 actually billed on Bedrock, 48.2M tokens, a daily spend chart peaking on Wednesday, top models led by Opus 5.5, top projects, and a kind-of-work split led by Build at 31%.
stats · last 7 days
The Harness tab comparing Claude Code, Codex, opencode and Gemini CLI over 7 days: share of spend, tokens, sessions, average session length, cost per session, cache hit rate, top model, top kind of work and a small daily chart for each.
harness · side by side

the dashboard · for a longer look

Daily spend by model, harness or kind of work; a burn-down of the tightest window with its forecast; every upcoming reset across your accounts; a calendar of tokens per day; and every session, newest first.

The Tokenrunway dashboard over 30 days: $1,704 at API prices, $212.60 actually billed, 83.6M tokens, 312 sessions, ran out 3 times. A stacked daily spend chart by model, a burn-down of the Claude work 5-hour window projecting a run-out at 5:48 PM before the 7:02 PM reset, a sessions table, upcoming resets and a tokens-per-day heatmap.
dashboard · sample data

ask for more · before you run out

Providers offer “ask your admin” once you're already locked out. Tokenrunway offers it at 85%, or as soon as your pace says you'll run out before the reset.

Set your admin's email, Slack or Teams once per account. Request more writes the message with the numbers in it, opens it in your mail or chat app, and you press send. Each request is noted on the account card, so you know you asked.

to: your adminemail · slack · teams

I'm at 86% of my Claude Enterprise weekly allowance (resets Mon 09:00). At my current pace I'll run out Thu around 15:00. Could you raise my limit for this cycle? Most of it is Claude Code: building (48%) and bug fixing (30%).

rattleboard · on the board across the room

Got a Rattleboard? Tokenrunway can flip its alerts onto your split-flap board and keep a live “AI runway” card in the board's rotation, on this computer or a board on your network.

Only window names, percentages and times reach the board. Never prompts, costs or keys.

Rattleboard→

private · stays on your computer

NO ACCOUNT

Nothing to sign up for. There is no Tokenrunway server.

NO TELEMETRY

No analytics and no crash reports. It talks only to the providers you add, and to GitHub for updates.

KEYCHAIN ONLY

Keys you add live only in the macOS Keychain, Windows Credential Manager or your Linux keyring. Your CLI logins are read, never copied.

NO PROMPTS KEPT

Stats are counts, models, projects and categories. Your prompts are never stored.

The details are in the privacy policy.

details

Where do the numbers come from?Allowances come from the providers. Some have no public usage API for subscriptions, so Tokenrunway reads what their own CLIs already keep on your computer, or calls the same endpoints their own apps call, for your own account only. Those can change without notice. When one does, Tokenrunway keeps the last good reading and says the source needs an update.
And the stats?From the session logs your agents already write: Claude Code, Codex and opencode. They're read on your computer, a few seconds after each turn. Costs use a bundled table of public API prices.
Does it get in the way of my agents?No. It isn't a proxy and never sits between an agent and its provider. It only reads.
Does it cost anything to run?Not with Claude, ChatGPT or the API organisations. On AWS, each Cost Explorer query costs one cent, so Bedrock spend is cached and checked at most four times a day.
Does it log me out of Claude Code or Codex?No. It reads their logins without changing them and never refreshes a token itself. On a Mac, reading Claude Code's keychain entry shows a system prompt once.
Is it official?No. Tokenrunway is not affiliated with or endorsed by Anthropic, OpenAI, Amazon, xAI, Google, GitHub or Cursor.

Get it

Free and MIT licensed. Version 0.1.0. Pick your platform; every download is on the releases page. Tokenrunway checks for new versions once a day, shows what's new, and updates in place when you say so. If it saves you from a 4:55 lockout, you can buy me a coffee on Ko-fi.

macOS
macOS 11+ · Apple Silicon and Intel
  1. Install with Homebrew (newer Homebrew asks you to trust the tap first), or download the DMG and drag Tokenrunway to Applications.
  2. Open it. It lives in the menu bar, with no Dock icon.
  3. Click + to add an account. It finds your Claude Code, Codex and AWS logins for you.
brew tap slastrina/tap
brew trust --tap slastrina/tap
brew install --cask tokenrunway
Windows
Windows 10 or 11 · x64
  1. Download the setup .exe (or the .msi for managed installs).
  2. Run it (if SmartScreen appears, choose More info → Run anyway). Tokenrunway sits in the system tray, by the clock.
  3. Windows can't show text in the tray, so the shortest runway is in the tooltip.
Linux
Ubuntu 22.04+, Debian 12+ · x86_64 and arm64
  1. Download the .deb and install it with apt, or use the AppImage on any distribution.
  2. The tray needs libayatana-appindicator; on GNOME, the AppIndicator extension.
  3. Logins go in your keyring (GNOME Keyring or KWallet).
sudo apt install ./Tokenrunway_0.1.0_amd64.deb

Homebrew installs update with brew upgrade --cask tokenrunway.

Claude is a trademark of Anthropic; ChatGPT and Codex of OpenAI; AWS and Bedrock of Amazon; Grok of xAI; Gemini of Google; GitHub Copilot of GitHub; Cursor of Anysphere. Tokenrunway is not affiliated with or endorsed by any of them. Screenshots show sample data.