macOS dictation that works offline, then gets clever online

What it does

TypeWhisper is a native macOS dictation and transcription utility that runs eleven different speech-to-text engines locally on Apple Silicon or Intel by default, keeping audio data on the machine. It can also fall back to cloud APIs like Groq, OpenAI, or xAI/Grok when speed matters. Once text is captured, the app passes it through customizable workflows for translation, rewriting, formatting, or app-specific automation, and can inject the result into any application via global hotkeys.

The interesting bit

The project treats dictation as a programmable platform rather than a simple input method. It ships with a plugin SDK, a local HTTP API, a CLI, watch folders, and desktop widgets, effectively turning your microphone into an automation trigger that can spin up local MLX-based models or route text to Linear issues and webhooks.

Key highlights

• Eleven STT engines under one roof, from on-device WhisperKit and Parakeet to cloud Groq and OpenAI Whisper, with per-request engine overrides.
• Workflows triggered by specific apps, websites, global hotkeys, or clipboard content, running through local Gemma 4 or cloud LLMs.
• A bundled plugin registry and SDK for adding custom transcription engines, TTS providers, and post-processors.
• Deep personalization: dictionary term packs that auto-learn corrections, snippet expansion with placeholders, and searchable transcription history.
• Full local automation surface: REST API, command-line tool, watch-folder batch processing, and menu-bar widgets.

Caveats

• Several advanced features—Apple Intelligence, Apple SpeechAnalyzer, and local Gemma 4 prompt processing—require macOS 26 or newer or specific Apple Silicon configurations, while the verified local LLM path is currently limited to 4-bit quantized models.
• The app is GPL v3 licensed, but some bundled local model plugins require accepting third-party licenses (e.g., Supertonic’s OpenRAIL-M) or Hugging Face tokens before downloading weights.

Verdict

Mac power users, accessibility advocates, and automation nerds who want system-wide voice control without surrendering their audio to the cloud will find a lot to like. If you just need quick, occasional transcription and do not care where the audio goes, the feature sprawl and model-management overhead may feel like overkill.