Show HN: TypeWhisper – speech-to-text with multiple engines, profiles

Hey HN, I'm Marco, the creator of TypeWhisper.

TypeWhisper is a free, open-source speech-to-text app for macOS and Windows. Everything runs locally on your machine - no cloud, no telemetry, no data collection. Your voice never leaves your device.

What makes it different:

- Multiple engines, your choice: On macOS: WhisperKit, Parakeet TDT, Apple SpeechAnalyzer. On Windows: Parakeet TDT, Canary 180M Flash. All run on CPU, no GPU needed. You can also plug in cloud APIs (OpenAI, Groq, Deepgram) if you prefer.

- Context-aware profiles: Set per-app and per-website overrides for language, engine, and AI prompts. TypeWhisper detects which app you're in and adapts automatically.

- AI text processing: Built-in prompts for rewriting, translation (20 languages), summarization. Works with on-device Apple Intelligence or cloud LLMs.

- Plugin system: Build custom transcription engines, LLM providers, post-processors, or action plugins with the Swift SDK (macOS) or .NET SDK (Windows). Community plugin catalog included.

- File transcription: Drag-and-drop audio/video files, export as SRT/WebVTT subtitles.

- Local HTTP API: Automate everything via REST endpoints.

Windows users: v0.3.3 just shipped with real-time WebSocket streaming, Fluent Design UI, three dictation modes (toggle, hold, hybrid), audio ducking, and whisper mode for quiet environments. CPU-only inference, runs on x64 and ARM64.

Download: - macOS: brew install --cask typewhisper/tap/typewhisper or https://github.com/TypeWhisper/typewhisper-mac/releases - Windows: https://github.com/TypeWhisper/typewhisper-win/releases

Website: https://www.typewhisper.com | Discord: https://discord.gg/pUFR4a65SD

Happy to answer any questions!