AppleJuice turns your voice into typed text in any app. This page walks through everything it can do — setup, every menu option, and what to do if something goes wrong.
The first time you use AppleJuice, macOS will ask your permission for two things. Both are required for the app to work at all:
Needed to record your voice when you start dictating. You'll be asked the first time you tap the talk key.
Needed so AppleJuice can type the transcribed text into whatever app you're using. Grant it under System Settings → Privacy & Security → Accessibility.
If you ever deny a permission by mistake, you can turn it back on any time in System Settings → Privacy & Security.
AppleJuice is tap-to-talk, not hold-to-talk — you never need to hold a key down.
While you're recording, a small pill-shaped window follows your cursor and shows:
It disappears once transcription finishes and your text has been typed in.
Don't like Right Option? Pick a different one from the tray menu under Talk Key. The available choices are limited on purpose to keys that never type a character by themselves — so switching your talk key can never interfere with your normal typing:
| Modifier keys | Other keys |
|---|---|
| Left/Right Option, Command, Control, Shift | Caps Lock, F13, F14, F15 |
AppleJuice can transcribe your voice two different ways. Switch between them any time from the tray menu's Transcription submenu.
Sends audio to OpenAI's Whisper API for transcription. Very accurate, works instantly, needs an internet connection and your own OpenAI API key.
Transcribes entirely on your own Mac — audio never leaves your machine, works with no internet connection at all. Requires downloading a model first (see below).
Cloud mode needs your own OpenAI API key (AppleJuice never ships with one built in — that's a cost only you control).
platform.openai.com (requires an OpenAI account with billing set up).sk- — not a whole line like OPENAI_API_KEY=sk-....Your key is stored securely on your Mac (encrypted via the macOS Keychain) and is never sent anywhere except directly to OpenAI.
Local mode needs a speech-recognition model downloaded to your Mac first. Pick one from the tray menu's Model submenu — clicking an undownloaded model starts the download automatically, with progress shown in the floating indicator.
| Model | Download size | Best for |
|---|---|---|
| Tiny | ~75 MB | Fastest, good for quick notes |
| Base | ~140 MB | A good balance of speed and accuracy |
| Small | ~465 MB | Noticeably more accurate |
| Medium | ~1.5 GB | High accuracy, slower |
| Large | ~2.9 GB | Best accuracy, slowest |
Nothing downloads without you clicking it — AppleJuice never fetches anything in the background without asking first.
An optional toggle in the tray menu. When on, AppleJuice automatically trims dead air from the very start and end of a recording before transcribing — the small pause between tapping the talk key and actually speaking, for example. This can reduce the odd stray word Whisper sometimes hallucinates from silence.
It only trims the edges, never pauses in the middle of your speech — those are left alone since they help with natural sentence breaks and punctuation.
Turning it on for the first time downloads a small (under 1 MB) speech-detection model.
A checkbox in the tray menu, available in both Cloud and Local mode. Speak in another language, and AppleJuice types out an English translation instead of a transcription in the original language.
It automatically detects whatever language you're speaking — there's nothing else to configure. Quality is generally best for well-represented languages (French, Spanish, German, Japanese, and similar) and best on Cloud mode or larger Local models.
A handful of spoken phrases get turned into real formatting instead of being typed literally, similar to how dictation software like Dragon NaturallySpeaking works:
| Say | Result |
|---|---|
new line | Inserts a line break |
new paragraph | Inserts a blank line |
upper case ... lower case | UPPERCASES everything said in between |
uppercase ... lowercase | Same as above (one-word phrasing also works) |
For example, saying "remember to say, upper case, urgent, lower case, in the subject" types out: remember to say URGENT in the subject.
If privacy matters more than convenience for a particular dictation, switch to Local mode first.
Cloud mode couldn't connect at all. Check your Wi-Fi/network connection, or switch to Local mode if you need to keep working offline.
Your OpenAI API key is missing or invalid. Re-open Set API Key… and make sure you've pasted just the key itself (starting with sk-), not a full line like OPENAI_API_KEY=sk-....
AppleJuice didn't hear anything usable in the recording. Check your microphone is picking up sound (System Settings → Sound → Input), and that you're not muted.
This almost always means the Accessibility permission hasn't been granted. Check System Settings → Privacy & Security → Accessibility and make sure AppleJuice is switched on in the list.
You're in Local mode but haven't picked a model yet. Choose one from the tray menu's Model submenu.