A Local-First Voice App for Mac Users Who'd Rather Not Send Audio to the Cloud
You've probably used a dictation or transcription app at some point, and you've probably also wondered where all that audio actually goes. For a lot of tools, the answer is "somewhere else." MacParakeet takes a different approach: it's a free, open-source voice app for Apple Silicon Macs that does its speech recognition locally, on your machine.
What It Does
MacParakeet bundles three things that usually ship as separate products: system-wide dictation, file and media transcription, and meeting recording. It's a native Mac app built in Swift 6, and it requires macOS 14.2 or later on Apple Silicon hardware.
The speech engine situation is more nuanced than "it just works." Parakeet v3 is the standard local engine, but the first-run setup is locale-aware—if your Mac doesn't have a preferred English language and leans toward Korean, Japanese, Chinese, or Cantonese, it selects WhisperKit instead. Beyond that, Parakeet v2/Unified, Nemotron Beta, Cohere Transcribe, and WhisperKit are all selectable, which gives you room to trade off language support, latency, timestamp quality, and accuracy depending on what you're doing. There's also an optional macparakeet-cli for agent automation, and a Transforms feature that works on selected text.
Why It's Cool
-
Local-first isn't a marketing line here. The README is explicit: all speech recognition happens on your Mac, and networked AI features are separate and opt-in. If privacy is the reason you've avoided these tools, this is the architecture you want.
-
The engine choice is genuinely thoughtful. Rather than picking one model and calling it done, the locale-aware first-run setup means people who aren't English-first users don't get a degraded default experience. That's a small design decision with real consequences.
-
It covers the whole workflow. Dictation while you type, dropping in a file, recording a meeting live with a floating pill and a Notes/Transcript/Ask panel—these are different jobs, and having them in one app beats stitching together three tools that each want a subscription.
-
The 0.8.10 release shows active polish. Speaker-count selection, transcript find and replace, additional dictation shortcuts, and meeting-capture recovery improvements are the kind of features that only show up when someone is actually using the thing daily. The source revision also adds opt-in Beta meeting recording reminders for the next release.
-
GPL-3.0, free, with a notarized DMG. You don't have to build it from source to try it, and you can read every line if you want to verify the local-only claim yourself.
How to Try It
The stable channel is the notarized DMG, currently version 0.8.10 (published October 14, 2026), which bundles CLI 5.1.0.
- Grab the DMG from the project's download link or the releases page.
- Install and launch. The first-run setup will pick an engine based on your Mac's language preferences.
- Start with the Transcribe tab—drop in a file or paste a YouTube link—or try system-wide dictation to see how it feels in your normal workflow.
- If you want automation, look at
macparakeet-cli, which ships with the 0.8.10 build.
The source lives at github.com/moona3k/macparakeet, and there's a DeepWiki badge if you want to ask questions about the codebase before diving in. Note the hardware requirement: this is Apple Silicon only, so Intel Macs are out.
Final Thoughts
MacParakeet is best suited to people who already live on Apple Silicon and want dictation and transcription without handing their audio to a server—or paying a monthly fee for the privilege. It's not trying to be everything to everyone, and the README's own line ("There are many voice transcription/dictation apps, but this one is mine") sets an honest tone about what this is: a well-built personal tool that happens to be open source. If that matches your setup, it's worth an install.
Follow @githubprojects for more developer tools and open source projects.