Rhino 0.1.2

• "Open System Settings" in the permissions banner now takes you straight
  to the exact privacy pane (Microphone, then Accessibility) instead of
  the Privacy & Security front page.

Rhino 0.1.1

• Hands-free mode: double-press your dictation shortcut to lock
  recording on — nothing to hold — and press it again to stop. On by
  default; toggle in Settings → Dictation.
• Dictation failures now say what went wrong (like "Model not loaded —
  check Settings → Models") instead of a bare "Transcription failed".
• Home shows a banner when no speech model is installed, with a button
  that takes you straight to the model downloads.
• Setup now verifies your chosen model actually loads before finishing,
  and interrupted model downloads are detected and retried instead of
  silently breaking every dictation.
• Cleanup now restores short words the engine dropped ("Schedule review"
  → "Schedule the review") when the optional cleanup model is enabled.

Rhino 0.1.0

• Rhino is born: new name, icon, menu-bar rhino, and a redesigned home
  with your dictation stats.

• Hold Fn to dictate (default), release to insert. Launch at login and
  live in the menu bar, on by default.

• Radically simplified: removed per-app rules, the SenseVoice and Apple
  Speech engines, translation, mouse/latch/submit trigger modes, the
  indicator layout editor, and the advanced decoding knobs. One shortcut,
  two engines (Parakeet + Whisper), sane defaults.

• Faster first dictation: the speech engine and cleanup model now load at
  launch instead of on your first recording.

• New Rhino app icon and a rhino menu-bar icon.

• Dictation history is on by default, stored only on this Mac; the main
  window now shows your stats (words, time saved, speed), history, and
  the dictionary.

• Silence no longer produces text: dictating nothing (or a dead mic) now
  inserts nothing, instead of an occasional hallucinated word.

• AI cleanup now always runs on the built-in on-device model; the Ollama
  server option and the post-record shell hook were removed.

• All-local, provably: removed the Remote (OpenAI-compatible/Groq)
  transcription engine and remote LLM cleanup entirely — every engine and
  cleanup model now runs on your Mac. Existing remote configurations fall
  back to Whisper / the built-in cleanup model, and stored remote API keys
  are removed from the Keychain.

• Removed the GitHub release-check ping; updates go only through the
  signed Sparkle feed, checked over HTTPS.

• Ollama cleanup connections are now restricted to this machine
  (localhost) — remote Ollama URLs are refused.

• New hygiene test gate: every push must transcribe end-to-end with zero
  non-localhost network connections.

• Initial skeleton from ~/app-template.