§ Local AI
Local AI models
Speech-to-text runs through Whisper (via the WhisperKit engine) or Parakeet (via the FluidAudio engine), pick per task. Enhanced polish runs through our own bespoke MLX model, fine-tuned for cleaning spoken input rather than general chat; meeting analysis runs through bundled MLX too. Advanced users can point Dettivo at compatible local MLX folders or cached Hugging Face MLX repos.
Who this is for
Anyone who wants to run Dettivo fully offline, anyone evaluating which local models do the work, and anyone trying alternative MLX models beyond the curated set.
Where local models are used
Dictation STT
Whisper tiers (via WhisperKit) and Parakeet (via FluidAudio), downloadable in Settings › Models. Pick the speed/accuracy tradeoff per task.
Dictation polish (Enhanced)
Our bespoke MLX polish model, fine-tuned for cleaning up spoken input, not a general-purpose chat model. Local rewrite without leaving your Mac.
Meeting analysis
Bundled MLX analysis model. Summaries, action items, decisions, all on-device.
Speaker diarization
Downloadable diarization model. Runs after meeting stops.
Advanced local model sources
Settings › AI Processing › Advanced Local MLX
- Source picker
- Curated (recommended), Local folder, or Hugging Face MLX ID.
- Local folder path
- A local MLX-format model directory you have prepared.
- Hugging Face owner/name
- Pulls a cached MLX repo. Validate before Download.
- Validate / Download / Load Test
- Three actions to confirm the model is usable before live runs.
- Reset to Recommended
- Bails back to the curated default if a custom model misbehaves.
Just launched
Just launched on your Mac.
One-time license, three active Macs, lifetime updates. $45 once, forever yours.