**WARNING: This version introduced ML Kit that is closed-source from Google to improve time and entity extraction for Chinese notifications. If you prefer a fully open-source version, please switch to the FOSS version.**
## Core enhancements
- Introduced Google ML Kit engine: added ML-based time and entity extraction, significantly improving recognition accuracy for times, locations, and event titles in complex Chinese notifications.
- Parsing engine integration: optimized settings logic to support seamless switching among "Built-in rules", "xk-time", and "ML Kit", and automatically synchronize time parsing with the event extraction engine.
- Experimental AI support: preliminary support for local GGUF models (based on llama.cpp) (experimental).

## UI/UX improvements
- Real-time debug log viewer: added a scrollable log area on the main screen to view notification processing, parsing results, and debug info in real time; includes auto-scroll and capacity limits to prevent lag.
- Settings UI upgrades:
    - Parser engine selector now displays detailed descriptions.
    - Added "Event reminder time" setting, supporting custom minutes-before reminders.
    - Improved interaction logic for the app selector.
- Background log persistence: supports loading the most recent 50 processing records from NotificationCache for easier tracing.

## Stability and compatibility fixes
- Android 14+ adaptation: updated foreground service type (specialUse) to resolve potential system termination of the data sync service on Android 14.
- ANR prevention: moved test parsing logic to a background thread to avoid UI freezes when handling complex text or loading AI models.
- Memory and performance optimizations:
    - Added per-log entry length limit (1200 characters) and a total count limit.
    - Improved TextView rendering efficiency by using batch updates to reduce layout redraws.
    - Rule engine enhancements: improved built-in rules for "countdowns" (e.g., “还有X小时”) and "relative offsets" (e.g., “3天后”).