VoiceInk is a native macOS voice-to-text app from an independent developer who signs the project as Pax, positioned in its own repository description as an open-source alternative to subscription dictation tools like Superwhisper and Wispr Flow. You hold a configurable global shortcut or use push-to-talk, speak, and the transcript lands in whatever app is in front of you. What separates it from the crowded field of macOS dictation utilities is the breadth of its engine layer: the repository organizes transcription behind a shared provider contract with four local backends, namely Apple's own Speech framework, whisper.cpp, FluidAudio for NVIDIA Parakeet models, and a Swift wrapper around transcribe.cpp for local GGUF models, alongside optional cloud providers including AssemblyAI, Cartesia, Deepgram, ElevenLabs, Gemini, Groq, Mistral, Soniox, Speechmatics and xAI plus an OpenAI-compatible path for custom endpoints, so the README's fully-offline privacy pitch describes the local model path rather than every configuration. Around transcription it layers workflow features: Modes that detect the frontmost app or URL and apply the matching pre-configured settings, screen-context awareness feeding the AI cleanup step, a personal dictionary for custom terminology and text replacements, rewriting modes tuned to different writing styles, and a built-in conversational assistant mode. Version 2.13, released on 2026-08-27, added Gemini 3.5 Transcribe and SenseVoice Small support, improved cancellation handling, and fixed the recorder panel and microphone routing in clamshell mode. Licensing is open-core rather than purely free: the source is GPL-3.0 and can be built yourself following BUILDING.md, while the packaged download from tryvoiceink.com is a paid license with a free trial that funds development and adds automatic updates and priority support, and the app is also published as the Homebrew cask voiceink. The maintainer states the project is not accepting pull requests, directing contributors to issues and forks instead. It requires macOS 14.4 or later.