cjpais/Handy

AI Summary: The user observes that the Whisper model remains loaded in VRAM even when the application is inactive, which consumes valuable GPU resources needed for other tasks. They propose a new feature to automatically offload the model from VRAM after a period of inactivity and reload it when processing resumes. This would improve resource management for users who use dictation intermittently.

AI Summary: Debug and fix a threading issue in the Rust backend of the Handy speech-to-text application that causes the end of transcriptions to be occasionally cut off. This likely involves improving the synchronization between the Voice Activity Detection (VAD) filter and the transcription engine, potentially requiring a deeper understanding of Rust concurrency and the application's architecture.