Added
- Guided first-run onboarding. A welcome window walks you through your language and the permissions one at a time, flips each switch on with you, advances on its own the moment access lands, and finishes with a live "try it" so you know it works before you start.
- Silero VAD as the speech gate. A tiny on-device voice-activity model now decides whether you actually spoke before anything is delivered. It catches pulsed noise and quiet speech that pure energy detection missed (RMS stays as an automatic fallback), so the hallucination guard is sharper still.
Changed
- One permission instead of two for push-to-talk. Talkink now uses Accessibility (which also enables auto-paste) rather than Input Monitoring. It shows up in the list on its own, you just switch it on, and it takes effect live: no more dragging the app in, no more quitting and reopening.
Fixed
- Long status messages no longer get cut off. The floating pill wraps and resizes to fit, so disk-space and memory notices read in full.
- Sturdier disk pre-flight before a model download, with real headroom so a "there's room" verdict is honest.
100% on-device. Open source. Notarized by Apple. Install with the DMG below, or brew install --cask hasso5703/tap/talkink.