hehehai/voxt

Supports multiple local and remote ASR engines (MLX Audio, Whisper via WhisperKit, Apple Dictation) and separate LLM providers for enhancement/translation/rewriting, with app-specific prompt branching and personal dictionary injection. Architecture uses modular model separation—ASR and LLM providers are independently selectable, enabling users to optimize transcription speed versus translation accuracy. Integrates with macOS 15.0+ for local model inference via MLX and WhisperKit, supports Hugging Face model downloads with China mirror fallback, and works across any macOS input field with context-aware rules per application.

346 stars.