Turn your voice into polished text wherever you write on your Mac. Speech recognition and optional writing assistance run locally—no account, no cloud transcription, and no developer analytics.
The microphone stays off between dictations. It opens only after you explicitly start recording and closes when you stop.
Dictate a message, document, note, or prompt in the app where you already work. Speak Easy Privately handles the local transcription and keeps the finished text ready to paste.
Automatic detection keeps multilingual dictation simple, including conversations that move between languages.
Optional on-device writing models can improve punctuation and capitalization without sending your words away.
Choose your shortcut, language mix, recording indicator, personal vocabulary, and optional encrypted history.
Install once and start dictating with the recommended speech and writing models already inside the app. Additional models can be downloaded directly within the app from Settings → Speech Models.
Install the compact app first, then choose which approved local models to download and keep on your Mac.
Both website editions use the same local-first privacy architecture. Each isolated downloader retrieves model files only—it never uploads speech or text.
Speak Easy Privately uses macOS permissions for three clear jobs: capturing audio only after you start a dictation, recognizing your chosen shortcut, and pasting the result. The app and its local inference helpers operate inside the system-enforced macOS App Sandbox.
How the app protects your MacApple’s macOS App Sandbox restricts access to files, hardware, network connections, and other apps.
Audio and dictated text are processed on your Mac, not sent to an inference server.
Each permission supports one visible feature and can be revoked in System Settings.
Optional history and vocabulary use device-protected Keychain keys.
Your voice stays on your Mac. Your workflow keeps moving.
In the website editions, background processing starts recognition while you’re still speaking—not only after you press Stop. Optional live insertion can deliver completed phrases into your writing app before you finish.
Fast dictation means coordinating audio capture, finding useful pauses, scheduling recognition and formatting, and delivering text without repeating earlier phrases.
| Configuration | Median | Maximum |
|---|---|---|
| Recognition only | 0.50 s | 2.22 s |
| With Qwen Fast | 0.66 s | 2.47 s |
| With Gemma | 0.98 s | 2.67 s |
87 of 97 recognition requests finished before Stop in this replay set.
No cloud transcription. No audio upload. All speech recognition and text formatting run locally on your Mac.
Live insertion is optional and off by default in the website editions. It requires Automatic Paste permission and keeping the same app, field and cursor. Editor compatibility varies. The App Store edition does not offer live insertion.
Internal benchmark on build 104, October 3, 2026: ten 45–120-second excerpts from two meeting recordings and one dictation test, replayed at recording speed. Apple M2 Pro, 32 GB memory, warm models, one run per excerpt and configuration, live-insertion processing enabled.
Timing ends when processing completes, not when text appears in an editor. These recordings were also used during development; this is not an independent accuracy benchmark, cloud comparison or latency guarantee. It is not a new benchmark of build 108. Default background processing without live insertion, App Store performance and physical Intel hardware were not measured in this comparison. Results vary with hardware, recordings and settings.
Speak Easy Privately brings the product experience, privacy boundaries, verification, and macOS integration together—but its local intelligence stands on years of openly licensed research and engineering. Without these projects and their contributors, this app would not be possible.
Meet the projects and contributorsThe multilingual speech foundation that turns audio into text locally.
Optional local models for conservative punctuation and capitalization. Gemma is included in Full and available to download in Lite.
The open software that makes efficient on-device inference practical.
