Tokenrunway v0.3.0

What's new

Added

• This cycle's spend on the Now tab: each allowance window shows what the work since its
  last reset would have cost at API prices ("$413 this cycle"), from the account's own agent
  (Claude Code for Claude, Codex for ChatGPT).
• $ left in the menubar: a new label shows the dollars left in a window: the real amount
  for budgets and credit limits, and for plan windows an estimate at API prices from this
  cycle's work ("~$590"). Settings → Show picks what the label follows: auto (the lowest of
  everything), one account's lowest window, or one window.
• tokenrunway guard: a check for scripts and agent loops. It exits quietly with 0 when every
  window is under your limit (90% by default, --max 80 to change it) and 1 with a one-line
  reason when one isn't, so tokenrunway guard && claude -p "…" stops a long batch before it hits
  the wall. Narrow it with --account and --window, add --runout to also stop when a window
  is forecast to run out before it resets, and --json for the details. When the app isn't
  running it can't tell, and stops (exit 3) unless you add --allow-unknown.
• Use it or lose it: turn on Settings → Notifications → Use it or lose it and Tokenrunway
  reminds you when a weekly or monthly allowance (Claude, ChatGPT, Copilot and other seats)
  resets in the next 12 hours with 30% or more still unused, even at your recent pace. It goes to
  your Rattleboard too, with its own switch under Settings → Rattleboard.
• New models priced within a day: Tokenrunway fetches the latest price list from GitHub once a
  day, so a new model gets its price without waiting for an app update. The prices built into the
  app stay as the fallback: behind a strict firewall it quietly uses those, and Settings → Fetch
  the latest prices shows which prices are in use (turn it off to always use the built-in ones).

Fixed

• Weekly forecasts that cried wolf: an hour of heavy use early in the week no longer reads as
  "Out Monday" at 2% used. The last hour's pace now counts for the next few hours; after that,
  the forecast uses your pace across the window so far, or the recent pace if you've slowed down.
• Check for updates: turning it off in Settings now stops the daily check (it kept checking).
  Check now still works.
• Stats per login: with two Claude (or two ChatGPT) accounts, the dashboard's account filter
  now shows each login's own sessions instead of sharing them. Codex says which account a session
  ran under, so that goes back through your history; for Claude Code, Tokenrunway notes which
  login is signed in as it reads the logs, so it counts from when you update (the spend tile says
  "since" that day). With one account per agent nothing changes.
• Session drill-down: click a session in the dashboard to see why it cost what it did: its
  cost over time, the tools it ran, and every step in order with its tokens and cost. The most
  expensive steps and the cache misses (a big prompt written to the cache again after a break)
  are marked. Only tool names are kept, never what they were given, and they're recorded from
  this version on.

Download the installer for your platform below.

macOS 11+ (Apple Silicon and Intel): open the .dmg and drag Tokenrunway to Applications. Or install with Homebrew: brew install --cask slastrina/tap/tokenrunway
Windows: run the -setup.exe (or the .msi). If SmartScreen warns you, choose More info → Run anyway.
Linux (x86_64): install the _amd64.deb (Debian, Ubuntu 22.04+) or run the _amd64.AppImage (chmod +x first). The tray needs libayatana-appindicator (and on GNOME the AppIndicator extension); logins need a Secret Service keyring such as GNOME Keyring or KWallet.
Linux (arm64): the _arm64.deb or the _aarch64.AppImage.