v2.7.0

v2.7.0

This release refreshes the on-device speech models, replaces Nemotron with a new Parakeet Unified provider, and makes Apple Speech more resilient. If you use a cloud cleanup model, it now stays on zero-retention OpenRouter providers.

Highlights

• Nemotron is gone; its slot is now taken by Parakeet Unified, a simpler single-model option.
• Parakeet TDT v3 has been replaced with Parakeet Ultra for better transcription quality.
• Cloud cleanup models no longer run reasoning, keeping cleanup output focused and predictable.
• Cleanup requests are routed only to OpenRouter providers with zero-retention policies.
• Apple Speech now falls back to DictationTranscriber when the preferred transcriber is unavailable.

Added

• Parakeet Unified transcription provider.
• Parakeet Ultra model option, replacing Parakeet TDT v3.
• Automatic fallback in Apple Speech so dictation keeps working if the primary transcriber cannot be used.

Changed

• FluidAudio updated to 0.18.
• Parakeet models are downloaded to the locations where FluidAudio actually loads them, avoiding mismatched or duplicate model files.
• Old model downloads are cleaned up at app launch instead of during ModelManager initialization, so startup no longer does that work inline.
• Cloud cleanup models are restricted to zero-retention OpenRouter providers.

Fixed

• Cloud cleanup models could previously emit reasoning output; they are now configured to stop reasoning.
• Apple Speech no longer fails outright when DictationTranscriber is needed.

Upgrade Notes

• Nemotron has been removed. If you had selected Nemotron, switch to Parakeet Unified or another available transcription provider after upgrading.
• Parakeet TDT v3 is no longer available. Select Parakeet Ultra and let it download the new model on first use; old TDT v3 downloads will be cleaned up automatically.
• If your cloud cleanup setup relied on a non-zero-retention OpenRouter provider, you may need to pick a different provider or model.

Full changelog: v2.6.1...v2.7.0