Skip to content

Soufflé v0.8.7

Choose a tag to compare

@github-actions github-actions released this 02 Sep 12:55
· 4 commits to master since this release
v0.8.7
bcba7d2

Soufflé now checks for updates on its own, and you pick which engine writes summaries.

Fixed

  • Summaries and dictation polish could silently run a different engine than the one you thought you had chosen, or do nothing at all, with nothing in the log. A leftover Ollama model name decided the provider, and an unavailable one fell through to whatever was first in the list. There is now an explicit choice in Settings — Automatic, Apple Intelligence, or Ollama — and an explicit choice is never traded for the other engine. Code models such as CodeLlama are no longer offered; they passed the polish prompt just well enough to look like they were working and then handed the dictation back unchanged.
  • The log file recorded what you said in the Settings microphone test, and the title of a meeting started from a calendar event. Both are now kept out of INFO, and the diagnostics bundle strips any remaining speech lines before you paste them into an issue. Paths are rewritten to ~ so they no longer carry your account name.

Improved

  • Soufflé checks GitHub once a day for a newer release and offers the notes with a Download button. Nothing is downloaded or installed; you still fetch the DMG or run brew upgrade. The switch is in Settings > About, on by default.

Changed

  • The README now states the two macOS floors (13 for dictation, 14.4 for system-audio capture), the permissions Soufflé asks for, and the three outbound connections it makes.

Notes for testing

  • With polish enabled, pick Ollama while Apple Intelligence is available and confirm dictation uses Ollama, not Apple. Pick Apple Intelligence with Ollama down and confirm Settings says so rather than offering a download button.
  • A stored Ollama model from before this release should still use Ollama, not silently switch to Apple Intelligence.
  • Leave the daily update check on, or trigger it from Settings > About, and confirm the dialog appears for a newer tag without downloading anything.
  • Generate a diagnostics bundle after a dictation and confirm the transcript text is not in the attached log.

Download the .dmg below for Apple Silicon Macs.

Requires macOS 13 or newer on Apple Silicon (M1 or newer).