Usage Limits 0.3.1

Usage Limits 0.3.1 improves app icons, account navigation, widgets and provider usage displays.

• Added six selectable app icons and transparent Android launcher artwork.
• Changed Overview to show the account that resets next, and fixed provider switching after Needs Attention.
• Added credit-provider logos and links to each provider’s key-management page.
• Added Grok product usage segments, removed duplicate Claude weekly rows, and updated Antigravity requests to the current Hub identity.
• Improved widget alignment, colors, ring text and provider selection. Shortened the documentation and added an MIT license and security policy.

Android launchers control any extra icon frame or background. These transparent icons do not supply monochrome themed variants. The iOS download is an unsigned archive and requires signing before installation.

Usage Limits 0.3.0

Usage Limits 0.3.0

New

• Z.ai GLM Coding Plan and OpenCode Go: connect with an API key and see five-hour, weekly and monthly limits. Z.ai works with z.ai and bigmodel.cn.
• Kimi International: Kimi now works with kimi.ai as well as kimi.com. You pick the region when you add the account; existing accounts stay on kimi.com.
• Credits: a new section and home-screen widget for prepaid API balances on OpenRouter, DeepSeek, Moonshot, xAI, fal.ai and Novita. Balances are shown in each account's own currency, with a low-balance warning you can set per account. Anthropic and OpenAI are listed as not possible, because neither offers a balance endpoint for API keys.
• Redesign of the screens and widgets on Android and iOS: neutral colours around your data, Terracotta only on bars and rings, the limit closest to running out at the top of the overview, and clear loading, empty and retry states.

Fixed

• Sign-in on Android 14 and later: signing in to Claude, Codex, Antigravity or Devin could hang on the last browser page, because Android paused the app while the browser was open. A short "Sign-in" notification now keeps the app awake until the login finishes.
• Wrong or stale percentages (about 25 fixes, Android and iOS), for example:
  • a limit that has reset since the last refresh no longer shows its old "used up" value as current, in the app or in widgets;
  • Kimi reset times are read again, and each Kimi limit gets its real window length instead of always "5 hours";
  • Antigravity and Devin quotas that are fully used no longer show as "unknown";
  • tapping refresh can no longer be swallowed by a pending background sync;
  • iOS widgets update their verdict and chosen window at every reset, not only when the app republishes.
• Kimi and Z.ai accounts from different regions no longer overwrite each other.

Checks

• Android: 710 unit tests, lint and release build; API 36 release-upgrade smoke test on CI.
• iOS: 637 kit tests on Linux and macOS; app built with the real SDK on CI.
• Every change was reviewed (Codex) until no major findings remained.

Not yet verified

• No live provider accounts were used in testing; provider responses are covered by fixture data.
• The Android 14+ sign-in fix and the redesign have not yet been checked on a physical device or iOS simulator.
• Z.ai, OpenCode Go and several credit endpoints are undocumented by their providers and may change.

Full Changelog: v0.2.2...v0.3.0

Usage Limits 0.2.2

Usage Limits 0.2.2

Changes

Codex windows (Android + Swift kit)

• Classification had one short and one long slot, so weekly + monthly or two 5-hour windows lost one. Unslotted windows now keep a position id (<prefix>-primary / <prefix>-secondary); existing -short/-long ids are unchanged for the common pairs.
• A family flagged allowed: false / limit_reached: true with no child window is kept as an unmeasured exhausted window (<prefix>-blocked, no percentage, duration or reset invented). A healthy code-review family can no longer make a globally blocked account read healthy.
• available_count / applicable_available_count accept only whole, non-negative values that fit an Int. Negative, fractional, oversized, bool values are unknown instead of rounded or saturated; a real zero stays zero.

Codex reset credits (Android)

• A failed optional credit request with no embedded copy produced resetCreditCount = 0 on an OK snapshot. The count now stays null and the snapshot is PARTIAL with a message; usage windows remain current. Rows count only when a source actually lists them. Cancellation is rethrown instead of being absorbed by the fallback.

xAI (Android + Swift kit)

• The deprecated monthly allowance is model scoped when a positive on-demand cap proves a second pool, but the newer credit view of the same allowance stayed global. Exhausted credits beside an unused on-demand pool read as EXHAUSTED instead of LIMITED. Merge now gives the credit window the same scope when the monthly window is model scoped or the on-demand window is measured (which needs a positive cap). An unknown on-demand row proves nothing. Cached snapshots are normalised on load.

Widget colour (Android + iOS app)

• The full-quota teal was chosen before the row's own severity, so an explicitly exhausted row still reporting 100 % drew as untouched. Teal now applies only to a healthy row; data validity still outranks both.

What's Changed

• Fix scoped quota status, refresh diagnostics and widget consistency by @ZepiGit in #15
• Reduce API 36 emulator load before release smoke by @ZepiGit in #16
• ci: select SwiftShader for API 36 startup by @ZepiGit in #17
• Fix Codex window loss, unknown credits, xAI pool scope and blocked row colour by @ZepiGit in #18

Full Changelog: v0.2.0...v0.2.2