The maintenance release fixes a packaging regression that disabled key Windows integrations while improving Antigravity quota reporting, model-based cost estimates and usage-history retention.
Usage v0.33.10 is a targeted maintenance release that restores broken Windows hook behaviour and tightens several parts of the tool’s AI usage tracking. According to the release notes, a packaging regression introduced in version 0.33.2 placed hook scripts in the wrong directory, which meant the installer could not find them. That disrupted first-run setup, session resume, Terse Mode and Quota-Aware Mode on Windows, while also preventing existing hooks from receiving updates. macOS users were not affected. The update corrects the installation path and brings hook updates back into line.
The repair matters because hooks are central to how Usage fits into coding workflows. The application is designed to track AI tool consumption, quota headroom and estimated costs across products including Claude Code, Claude Desktop, Codex and Antigravity. When those integrations fail without warning, developers can be left believing automation is active when it is not. Restoring hook installation therefore improves not only reliability, but also the trustworthiness of the workflow features built on top of it.
Recent changes to Antigravity quota tracking add further detail to the release history. The Antigravity CLI documentation explains that the /usage and /quota commands open an interactive panel showing active model limits and remaining allowance, which makes accurate reporting especially useful for developers trying to keep within provider caps. Usage has been updated to query that information more reliably, including Gemini weekly-quota reporting and support for switching between Gemini, Claude/GPT and Gemini 4 Argon quota groups. The distinction matters because Argon uses a weekly limit rather than the five-hour limit applied elsewhere, and the interface now reflects that difference more clearly.
The broader pricing work in the application has also continued. Earlier fixes aligned Grok cost estimates with xAI pricing rather than treating usage as free, while other updates corrected offline pricing for Claude Sonnet 5.5 cache reads and made Codex sessions that switch models mid-conversation charge each turn against the model actually used. Those changes are significant because token counts alone do not describe the real cost of AI-assisted development. Model-specific pricing gives a more accurate basis for comparing tools and understanding API-equivalent spend.
Usage has also been working to preserve more historical data for longer. Version 0.33.6 added background saving of recent usage history every six hours and retained the last 45 days even when reports are not opened regularly. That helps offset the default transcript-deletion behaviour of Claude Code, which removes older sessions after 30 days. The application can fill in missing days from saved records for totals and trend reporting, although some fields still cannot be reconstructed when the underlying transcript details are unavailable. Users moving to another machine are still advised to copy the ~/.usage/ directory.
Performance and Windows stability have improved in parallel. Earlier releases reduced background refresh CPU time sharply, trimmed retained memory from previously viewed panels and addressed Windows-specific issues around native calls and tray-icon handling. Other changes reduced unnecessary temporary WebView2 folders, preserved session-resume behaviour across multiple open Claude Code windows and stopped background-agent activity from clearing Beginner Mode selections too early. Taken together, the release line suggests a continued focus on making Usage a light-weight monitoring layer rather than a resource-heavy dashboard. For Windows users who depend on hooks, v0.33.10 is the clearest update to install first.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
