Changelog

Notable changes in OneSentence. Binary updates use Sparkle (onesentence.app/appcast.xml). Release assets: GitHub Releases.

3.23.0 August 2026

• Settings → Permissions shows live Microphone and Accessibility status, with the same repair steps as first-run when a switch was turned off.

3.22.0 August 2026

• Opening Settings → Models after an update no longer freezes on Checking local model integrity.

3.21.0 August 2026

• First-run no longer finishes when models are on disk if Microphone or Accessibility is still off.
• The download HUD says Models downloaded, not Ready, and stays quiet during first-run.
• Launch at login is on by default for new installs.
• Settings, Help, and first-run copy is shorter. Accessibility is required for Listen.

3.20.0 August 2026

• Long transcribing HUD copy on macOS 13.3 keeps rounded capsule ends.

3.19.0 August 2026

• First-run model download shows progress instead of staying on a spinner.
• HUD capsules keep rounded ends on macOS 13.3, including transcribing and unknown-language toasts.
• Reading HUD Cancel works on macOS 13.3.
• Transcription finishes even if Accessibility is not granted.
• History loads on macOS 13.3 instead of staying on Loading History.

3.18.0 August 2026

• Reading HUD no longer crashes when its message updates.
• Cancelling a paste restores the previous clipboard contents.
• Interrupted model downloads resume from the correct byte offset.

3.17.0 August 2026

• Status HUD is closer to the old compact size.
• The capsule grows with the message instead of clipping it.
• Recording Cancel sits farther from the waveform.

3.16.0 August 2026

• Status HUD while recording, transcribing, or reading, with Cancel (and pause/resume while speaking).
• HUD position is Top or Bottom in Settings → General. New installs default to Top.
• Visual alerts use a glass capsule with a thin colored outline.
• Headset pause toggles pause/resume. Leaving Voices no longer stops Read.

3.15.0 August 2026

• Factory reset erases local data and keeps the app. Uninstall erases leftover models and settings, then shows OneSentence in Finder so you can drag it to the Trash.
• Quit after reset or first-run setup actually quits.
• Installer disk image shows drag-to-Applications.

3.14.0 August 2026

• Insert transcription into the active field after Listen (on by default; turn off in Settings → General).
• Default speech-to-text is Parakeet. Whisper is optional and downloads only when you tap Download on a model.
• Voice quality is Fast / Balanced / Quality. Speaking rate is synthesis tempo (0.5x–2.0x).
• First-run download is about 1 GB.
• Update checks wait until first setup finishes.

3.13.0 April 2026

First public MIT snapshot. No license key. Models download from Hugging Face. Updates use GitHub Releases and this site’s appcast.

• Duck other applications while speaking or recording.
• Now Playing widget for active playback, including media keys.
• Clearer recordings via Voice Processing (VPIO) and echo cancellation.
• Application source MIT; model weights keep upstream licenses (see Credits).