Parakeety vs VoiceInk

The difference is not on-device versus cloud, because both run transcription on the Mac. The difference is open and configurable versus closed and decided. VoiceInk hands you the source, a choice of speech models, and optional cloud cleanup; you assemble the setup you want. Parakeety makes the decisions for you and ships a single app that runs the moment you install it.

Neither approach is wrong. They serve different people. If you like reading code, tuning models, and owning the whole stack, the open path is genuinely appealing. If you want to hold a key, talk, and have the words appear, the decided path is less to think about.

VoiceInk is built in public on GitHub under the GNU General Public License v3.0, maintained by a solo developer who has been on it full-time. The whole codebase is there to read, audit, and compile. That is a real strength: you can verify exactly what it does, fix something yourself, or run a build no one else controls. The trade is that building from source is a developer task, and a self-built copy does not get the automatic updates and support that come with the paid version.

Parakeety is closed source. You do not get the code, and that is a fair thing to weigh if auditability is what you care about. What you get instead is a signed and notarized app that Gatekeeper trusts, no compiler step, no API keys to paste, no model files to fetch by hand. You download it, grant microphone and accessibility permission, and dictate. The trade is opacity for zero setup.

VoiceInk runs whisper.cpp for the Whisper family and also supports a Parakeet model, so you can pick the engine that suits a given language or speed need. The pre-built version adds an optional cloud enhancement step that sends the transcribed text, not the audio, to an AI provider to tidy it up. All of that is configurable, and configurable means there is a right setting to find.

Parakeety runs one model: Parakeet TDT v3, a transducer model on the Apple Neural Engine. On the standard benchmarks it posts a 6.32% word error rate against Whisper Large V3's 7.44%, transcribes ten minutes of audio in around 0.2 seconds, and the transducer architecture does not hallucinate phrases on long silences the way Whisper sometimes can. There is no model picker and no post-processing to enable. One model, tuned for one job. The cost of that is no flexibility: if Parakeet does not cover your language, there is nothing to switch to.

With VoiceInk you have two front doors. You can clone the repo and build it yourself, which suits a developer who wants the free, fully open route and does not mind the compiler and the manual model setup. Or you can download the paid pre-built app, which removes the build step and adds updates and support. Either way you then choose a model and decide whether to turn on cloud cleanup.

With Parakeety there is one door. Install the app, grant two permissions, hold the section key (§) below Esc, talk, release, and the words paste at the cursor in whatever app you were already in. No model choice, no provider keys, no build. That is the whole setup.

So VoiceInk fits people who value open source, want to tune the engine, and are happy to assemble their setup. Parakeety fits people who want dictation that works on first launch and never asks them to configure anything. If push-to-talk at the cursor is the specific behavior you are after, the push-to-talk versus always-on dictation comparison goes deeper on why that loop matters.

• Open source. GPL v3, built in public; read it, audit it, fork it, run a build no one else controls.
• Free to build yourself. If you compile from source, there is no cost at all.
• Model choice. whisper.cpp Whisper models plus a Parakeet option, so you can match the engine to the task or language.
• Optional cloud cleanup. A text-enhancement step you can switch on when you want it, off when you don't.
• Lower entry price. The paid app starts at $25 for one Mac.

• No setup. A signed, notarized app that runs on first launch. No build, no API keys, no model files to fetch.
• One fast on-device model. Parakeet TDT v3 on the Apple Neural Engine, around 3,333x real-time, with no model picker to get wrong.
• Push-to-talk at the cursor. Hold the section key, talk, release, paste in any app. The whole product is that loop.
• No silence hallucination. The transducer architecture does not invent phrases on quiet stretches.
• $30 once, everything included. Every feature and every future update, no tiers to navigate.
• Solo-maker support. Parakeety is a one-person operation; mail the person who builds it and you reach the person who builds it.

For other points in this comparison family, the Parakeety vs SuperWhisper piece covers another Mac model-menu competitor, Parakeety vs MacWhisper covers file transcription versus dictation, and Parakeety vs Wispr Flow is the deeper read on cloud versus local. If you are still mapping the category, the explainer on what on-device speech recognition actually is and the overview of voice to text on Mac are good starting points.

Parakeety is a Mac menu-bar app. Hold the section key (§) below Esc, talk, release; your words paste at the cursor in whichever app you were typing into. Audio never leaves the machine. There is a free 7-day trial with no card required. After that it is $30 once, all features and all future updates included.

Try Parakeety free →