Glimpse v0.9.4

Features

• Glimpse now supports all whisper models.
• Onboarding is now a quick guided setup: tell Glimpse which language you dictate in and whether to prioritize quality, balance, or size, and it picks and downloads the right local model for you.
• New model browser for choosing local models: search by name, filter by category, and see each model's size and supported languages at a glance. Quantization options (Q5, Q8, Full) are grouped under each model so switching between them is simpler.

Improvements

• Much of the UI has been cleaned up and improved to feel cleaner.
• Words spoken on your account card is now a running lifetime total, it counts everything you've dictated and is no longer based on the current amount of transcripts.
• Model downloads are now verified after they finish, so an interrupted or corrupted download won't leave you with a broken model.
• The transcription language list now clearly separates the languages your current model supports from the ones that need a different model.
• Auto Dictionary now works with online transcription providers, not just local models.

Fixes

• Fixed cancelling a dictation and immediately starting a new one occasionally pasting the cancelled text and glitching the recording pill.
• Fixed the Nemotron Streaming model downloading mismatched files, which left it broken and showing the wrong size.