VoiceInk

Start dictation from a shortcut and adapt output to the active app

VoiceInk can start recording from configurable global keyboard or mouse shortcuts and works naturally as a push-to-talk dictation layer. Modes can react to the active application or URL, allowing different writing behavior for chat, email, documentation, or technical work. The Personal Dictionary can add names, domain terminology, and text replacements so recognition can be tuned around vocabulary that generic speech systems often miss.

Sources: [1]

Use local transcription or opt into cloud speech and AI enhancement

The README lists local technologies including whisper.cpp, Parakeet through FluidAudio, transcribe.cpp-based models, and SenseVoice. The current onboarding code also exposes cloud transcription providers including AssemblyAI, Cartesia, Deepgram, ElevenLabs, Soniox, Speechmatics, xAI, Mistral, Groq, and Gemini, while enhancement configuration includes providers such as OpenAI, OpenRouter, and Anthropic. VoiceInk is therefore neither inherently cloud-only nor guaranteed to keep every enabled feature offline: the data path depends on the providers you configure.

Sources: [1][2]

For Mac users who want speech input across many applications

VoiceInk fits people who spend substantial time composing email, chat messages, documents, prompts, or development notes across macOS applications. It is especially relevant when the goal is immediate dictation into the current workflow rather than uploading recordings for later transcription. Sensitive material can stay on a local-model path, while users who prioritize recognition speed or provider-specific capabilities can opt into supported cloud services.

Sources: [1][2]

Review screen permissions, cloud data paths, and build differences

Context Awareness uses the macOS Screen Recording permission so VoiceInk can read visible screen content for transcription context. Environments with sensitive on-screen information should decide deliberately whether that permission and feature are appropriate. Cloud transcription or AI-enhancement providers also change the privacy boundary because relevant data is sent to the selected external service rather than being processed only on the Mac.

VoiceInk requires macOS 14.4 or later. It can be built from source with make local, but the BUILDING guide states that local builds do not include iCloud dictionary synchronization or automatic updates. Purchasing the distributed build provides conveniences such as automatic updates and support, while the source itself is published under GNU GPL v3. Modified or redistributed versions should be evaluated against the GPLv3 obligations.

Sources: [3][2][4][5]