Is Spokenly on-device?
Short answer: it can be, and the answer depends on a choice you make inside the app. Spokenly runs free local Whisper and Parakeet models offline on Apple Silicon, and it also supports cloud engines including OpenAI, Deepgram and Groq. So whether your audio stays on the Mac is determined by which model is selected, not by the product as a whole. Spokenly does provide a Local Only Mode that blocks all network requests, which turns a conditional answer into a definite one. This is the same local-by-configuration pattern described in the primer on on-device speech recognition.
Local by configuration versus local by architecture
Parakeety has no model picker, because it ships exactly one: the NVIDIA Parakeet TDT 0.6B v3, on the Apple Neural Engine. There is no cloud engine to select by accident and no mode to verify, since the product contains no path for audio to take off the machine. $25 once, on macOS 14 or later on an Apple Silicon Mac (M1 or later). The trade is flexibility: where Spokenly lets you reach for a larger cloud model when a recording is hard, Parakeety gives you one model and no fallback.
Why the answer is conditional
Spokenly is a multi-engine app. Its free local models, Whisper and Parakeet, run offline on Apple Silicon and keep audio on the machine. Its cloud engines, including OpenAI, Deepgram and Groq, do what cloud engines do, which is receive your audio.
That makes on-device a property of your current selection rather than of the app. Both configurations are legitimate, and a tool that offers both is more flexible than one that does not. It simply means the question is Spokenly on-device has no single correct answer that holds regardless of setup.
It also means the answer can change without you deciding it should. Selecting a different model for one difficult recording changes where the next recording goes too, unless you change it back.
Local Only Mode is the part worth knowing about
Spokenly ships a Local Only Mode that blocks all network requests. That is a stronger guarantee than choosing a local model, because it does not depend on remembering which model is active.
For anyone whose reason for asking is confidentiality rather than curiosity, this is the setting that matters, and it is the one almost nobody mentions when answering this question. A model choice is a preference. A mode that blocks network requests is an enforcement.
The distinction is the same one that separates a policy from a control anywhere else. If the consequence of getting it wrong is meaningful to you, prefer the control.
Which shape suits which job
A multi-engine app with a local mode is the right tool when your work varies: mostly routine dictation, occasionally a recording where a larger cloud model earns its keep, and a switch to flip when the material is sensitive.
A single-model local app is the right tool when the variation is the problem. If what you want is that the question never arises again, and you will trade the ability to escalate a hard recording for that, the narrower product is the better fit.
Neither is a better piece of software. They answer different questions about how much configuration you want to own.
| Configuration | Where audio goes | How certain is it |
|---|---|---|
| Local model selected | Stays on the Mac | Depends on the current selection |
| Cloud engine selected | Sent to the provider | Definite, by design |
| Local Only Mode on | Stays on the Mac | Enforced, all network requests blocked |
FAQ
- Does Spokenly send my audio to the cloud?
- Only if a cloud engine is selected. Spokenly supports cloud providers including OpenAI, Deepgram and Groq alongside free local Whisper and Parakeet models that run offline on Apple Silicon. With a local model chosen, audio stays on the Mac. Because it is a setting rather than a fixed property, the answer follows your configuration.
- What is Spokenly Local Only Mode?
- A mode that blocks all network requests, so nothing leaves the Mac regardless of which model happens to be selected. It is a stronger guarantee than picking a local model, because it does not depend on remembering the current selection. If your reason for asking about on-device processing is confidentiality, this is the setting that matters.
- Which Spokenly models run offline?
- The free local models, Whisper and Parakeet, run offline on Apple Silicon with unlimited use and no account. The premium managed cloud models on the Pro tier, such as GPT-4o Transcribe, Deepgram Nova 3 and Soniox, require a connection because the transcription happens on the provider's infrastructure rather than on your Mac.
- Is Spokenly private by default?
- Its free local models keep audio on the machine, and Local Only Mode blocks network requests entirely. But because Spokenly ships both local and cloud engines, privacy is a function of configuration rather than something the architecture guarantees on its own. That is a meaningful difference from an app with no cloud path at all, and it is worth knowing which sort you are running.
Try it
Parakeety is the narrow version of this trade. One model, on the Neural Engine, no picker and no mode to verify, because there is no cloud path in the product to switch off. Hold the section key, talk, release, and the words paste at the cursor in whichever app you were typing into. There is a free 7-day trial with no card required, and after that it is $25 once.