Download Latest Version AIUsageTracker_Setup_v2.4.6_win-x86.exe (14.5 MB)
Home / v2.4.3
Name Modified Size InfoDownloads / Week
Parent folder
AIUsageTracker_Setup_v2.4.3_win-arm64.exe 2026-07-13 14.2 MB
0
AIUsageTracker_Setup_v2.4.3_win-x64.exe 2026-07-13 14.9 MB
0
AIUsageTracker_Setup_v2.4.3_win-x86.exe 2026-07-13 14.3 MB
0
AIUsageTracker_v2.4.3_win-arm64.zip 2026-07-13 17.0 MB
0
AIUsageTracker_v2.4.3_win-x64.zip 2026-07-13 17.8 MB
0
AIUsageTracker_v2.4.3_win-x86.zip 2026-07-13 16.9 MB
0
appcast.xml 2026-07-13 1.1 kB
0
appcast_arm64.xml 2026-07-13 1.1 kB
0
appcast_beta.xml 2026-07-13 1.1 kB
0
appcast_beta_arm64.xml 2026-07-13 1.2 kB
0
appcast_beta_x64.xml 2026-07-13 1.1 kB
0
appcast_beta_x86.xml 2026-07-13 1.2 kB
0
appcast_x64.xml 2026-07-13 1.1 kB
0
appcast_x86.xml 2026-07-13 1.1 kB
0
release-checksums-v2.4.3.txt 2026-07-13 622 Bytes
0
README.md 2026-07-13 2.3 kB
0
v2.4.3 source code.tar.gz 2026-07-13 5.9 MB
0
v2.4.3 source code.zip 2026-07-13 6.3 MB
0
Totals: 18 Items   107.3 MB 0

Fixed

  • Aggregate percentage now sums only cards from the latest refresh batch — When a provider changes plans (e.g., from dual-window burst+weekly to single-window weekly), old cards from the previous plan stayed in the 24h database window and were summed into the provider-level aggregate alongside the current card. This caused wrong provider-level percentages (e.g., 44% shown instead of 11%).
  • OpenAI/Codex providers now detect window type from limit_window_seconds — The API changed its response format for some plans (e.g. "prolite"). Window type is now read dynamically, supporting both old dual-window and new unified weekly format.
  • Tooltip per-window lines for Codex/OpenAI dual-bar providers — Widened type check to QuotaProviderUsage so per-window lines appear for ModelScopedProviderUsage objects.
  • Tooltip pace calculation uses correct window values — Pace target now uses rolling window values for all three terms instead of mixing burst and rolling windows.
  • Monitor no longer writes port: 0 to monitor.json on hibernate — Port file is left untouched across hibernate cycles.

Changed

  • Group summary aggregates across all available quota entries — RequestsUsed, RequestsAvailable, and UsedPercent on group summaries now reflect the sum of all quota entries, giving accurate totals for providers with multiple windows.
  • Simplified CodexProvider window handling — Removed separate code paths for unified vs dual-window detection. Primary window always produces one card; secondary window (when present) produces a second card.
  • NormalizeUsage now mutates in place — No allocation per entry, no user-visible difference.

Removed

  • Spark card from OpenAI Codex provider — The erroneous codex.spark sub-model card is no longer generated. Existing history preserved.
  • Plan type from card descriptions — Descriptions no longer include "| Plan: plus" suffix.
  • CachedGroupedUsageProjectionService — Redundant in-memory cache removed.
  • SettingsAdditionalProviderIds provider discovery — Settings now uses a flat set of ProviderDefinition entries.
  • SelectPrimaryUsage owner-matching fallback — Groups use the most recent usage row directly.

Full Changelog: https://github.com/rygel/AIUsageTracker/compare/v2.4.2...v2.4.3

Source: README.md, updated 2026-07-13