Faster transcription, stable offline AI, and cleaner layout control

🔄 Improvements

• Improved slides and video layout behavior: main area visibility is now consistent across modes, with a single control for the whole media block Demo
• Faster transcription start: reduced delay before recording by optimizing offline preparation and keeping models warm between sessions Demo
• Faster transcription by default: audio processing now uses lower quality settings to prioritize speed Screenshot

✨ New Features

• Added Settings button to the top-right panel for quicker access to configuration Screenshot
• Clearer error messages for broken or inaccessible attachments, with actionable feedback instead of silent failures

🛠 Fixes

• Fixed: inconsistent layout visibility between slides and video modes
• Fixed: Deepgram API key issue affecting transcription

And other improvements