onWatch

Codex auto quota-starter (Beta)

Codex 5-hour and weekly limit windows only begin counting once you send your first message after a reset. If you do not use Codex right away, the fresh window - and its reserve - sits unstarted. This release adds an opt-in auto quota-starter that starts it for you (closes #76).

• Detection. An unstarted window reports its reset time as now + full window, so its countdown stays pinned at the full length (a 5h window shows ~4h59m and does not tick down). onWatch checks the reset countdown after each poll; a window that stays pinned is unstarted, one that is decreasing is already started.
• The start. onWatch sends one small but complete Responses-API turn (it asks the model to reply with the string Quota Resumed) and reads the stream to completion. A turn cancelled mid-stream is not counted by the backend, so the full turn is what actually anchors the window. Roughly ~60 tokens per start.
• Controls. Two independent toggles - Auto-start 5h window and Auto-start weekly window - in Settings -> Providers -> Codex. Off by default; changes apply without a daemon restart. Equivalent env vars: CODEX_AUTO_START_5H, CODEX_AUTO_START_7D, and CODEX_STARTER_MODEL (default gpt-5.5).
• Safety. Hard-capped to 5 starts per rolling 4 hours per window, so it can never loop or waste quota. The active card shows an "Auto-start" badge while enabled.

See the Codex Setup Guide for full details and cost.

Fixes

• Codex agent stability. Fixed a churn where the Codex agent restarted every ~30 seconds because the manager's own credential write-backs were seen as external profile changes by the profile scanner.

Docs

• Documented the Antigravity agy CLI data source and ANTIGRAVITY_SOURCE in the README.

Full Changelog: https://github.com/onllm-dev/onWatch/compare/v2.12.3...v2.12.4