AeriVoice for Mac / Documentation
A few steps.
Then just speak.
Set up dictation that goes where you write. This guide takes you from installing AeriVoice to making it your own.
01 / Start here
Getting started
Install the app
Download the latest DMG, open it, and drag AeriVoice into Applications. Launch it from Applications; AeriVoice lives in your Mac’s menu bar.
Allow Microphone and Accessibility
Microphone access lets AeriVoice hear your dictation. Accessibility access lets it send the finished text to the field you’re using. Follow the setup prompts to enable both.
Choose how to transcribe
Connect your own cloud provider accounts, or set up an on-device model and enable Offline mode. Cloud providers may charge for usage; Offline mode needs no API keys.
Check your shortcut
Open AeriVoice Settings and choose a shortcut that feels natural. Hybrid is the default behavior: tap to toggle recording, or hold and release to finish.
New to providers? The default cloud setup uses Soniox for transcription and OpenRouter for AI cleanup. Add a Soniox key and an OpenRouter key in Settings → Providers.
02 / Everyday use
Your first dictation
Click into a text field in the app you want to write in. Use your shortcut, speak naturally, and finish the recording. AeriVoice transcribes your speech, applies your cleanup settings, and sends the result to that field.
- Hybrid Default
- Tap to start and tap again to finish, or hold the shortcut while speaking and release to finish.
- Toggle
- Press once to start recording, then press again to finish.
- Hold
- Hold the shortcut while speaking. Release it when you’re done.
Keep the destination field focused as you finish. If automatic Paste is unavailable, AeriVoice can copy the result and explain why. A green check means the Paste shortcut was sent; check the destination to confirm your text arrived.
03 / Your setup
Providers & API keys
Transcription turns audio into text. AI cleanup then edits that text. You choose these separately, so you can mix providers to suit your workflow.
| Stage | Options |
|---|---|
| Cloud transcription | Soniox, Meta, Grok (xAI), or Cartesia |
| On-device transcription | NVIDIA Nemotron 3.5 or Apple Speech |
| AI cleanup | OpenRouter, Cerebras, or Groq (experimental) |
Add keys under Settings → Providers. Choose your transcription engine in Dictation, and your cleanup provider and model in AI Cleanup. API keys are saved in macOS Keychain.
OpenRouter offers recommended models and a searchable catalog of compatible text models. Available reasoning settings depend on the model; more reasoning can add waiting time.
Grok transcription is available from 0.1.0-beta.10 and requires an xAI API key with API credits. Provider accounts, usage limits, and language coverage vary.
Cartesia transcription is available from 0.2.3 and requires a Cartesia API key with credits. Choose Cartesia Ink 2, the stable model, or Cartesia Ink Preview, which can change without notice. Both transcribe English, French, Hindi, Japanese, and Spanish only.
From 0.2.4 you can choose a Spoken language under Dictation → Transcription, or keep Auto-detect, the default. Soniox restricts recognition to the chosen language; Grok and Meta use it as a hint and can still follow another language. Cartesia detects its languages itself, Nemotron transcribes English, and Apple Speech keeps its own language setting.
04 / Make it sound like you
Cleanup styles
Choose a style in Settings → AI Cleanup → Cleanup style.
- Faithful
- Removes speech clutter and fixes punctuation while staying close to your wording.
- Polished Default
- Improves grammar and phrasing while keeping your meaning and details.
- Compose Experimental
- Organizes lists and paragraphs, handles spoken corrections, and refines the current recording.
- Custom Experimental
- Adds your own instructions for tone, terminology, translation, or formatting on top of Polished cleanup.
With Compose
“Meet on Monday, sorry, Wednesday.”
Meet on Wednesday.
An example of the intended behavior. Results depend on your selected model.
Custom instructions support up to 2,000 characters. They stay saved when you switch styles and are sent to your cleanup provider only when Custom is selected. Changes made during recording apply to the next dictation.
AI cleanup can make mistakes, especially with experimental styles. Review text that matters. Offline mode skips all AI cleanup.
05 / Stay on your Mac
Local & offline dictation
Choose Local in Dictation settings. Manage models and language support under Providers → Local models → Manage.
- NVIDIA Nemotron 3.5
- An English on-device model with an approximately 611 MB download. Interrupted downloads can be resumed or removed.
- Apple Speech
- Uses macOS speech recognition and system-managed languages. It reuses installed language support; missing languages need a download from Apple.
Local transcription and Offline mode are different. Selecting Local keeps transcription on your Mac, but cloud cleanup may still receive the transcript. Enable Offline mode to use local transcription with no cloud cleanup or cloud keys.
Finish downloading your model or language assets before enabling Offline mode. You can turn it on from General settings, the menu bar, or local onboarding. Your usual provider choices are preserved for when you switch back.
AeriVoice does not silently fall back to cloud transcription if local assets are unavailable.
For setup, a disconnected-network check, and troubleshooting, follow the offline dictation on Mac guide.
06 / Names & specialist terms
Your dictionary
Add names, acronyms, and specialist vocabulary to the Dictionary. Supported transcription engines use these entries as recognition hints; they do not guarantee exact spelling.
Grok accepts up to 100 eligible entries, with a maximum of 50 Unicode code points per entry. Cartesia accepts up to 100 entries totaling 1,200 characters, and fewer in non-Latin scripts. Excluded terms remain saved, and Dictation settings show the exclusions.
07 / Know where your words go
Privacy & permissions
- Cloud transcription: microphone audio and vocabulary hints go to the provider you select.
- Cloud cleanup: your transcript and cleanup instructions go to your selected cleanup provider.
- Offline mode: uses your local engine and skips cloud cleanup. AeriVoice blocks new app-initiated network requests while it is enabled.
- Credentials: provider API keys are stored in macOS Keychain. You do not need an AeriVoice account.
Some providers prepare connections in advance. For example, Grok can receive your key and eligible dictionary hints before dictation starts, but no microphone audio is sent during that preparation.
Optional diagnostics are off by default. When enabled, they stay on your Mac and exclude audio, transcripts, clipboard contents, and credentials. Manage them in Settings → Privacy & Data.
Settings → Stats shows local word counts, dictations, recording time, and speaking pace. Only daily totals are saved, without transcripts. You can turn collection off or clear these totals in Privacy & Data.
You can review or revoke Microphone and Accessibility access in macOS System Settings → Privacy & Security. Each cloud provider’s own retention and account policies apply. Read the full privacy details
08 / If something gets in the way
Troubleshooting
I can’t start recording.
Check that AeriVoice is running in the menu bar, review your shortcut in Settings, and confirm Microphone permission is enabled in macOS System Settings → Privacy & Security.
AeriVoice uses the wrong microphone.
Choose an input in Settings → Dictation → Microphone. System default follows the input selected in macOS; if a chosen microphone is disconnected, AeriVoice falls back to the system default. From 0.2.5, a chosen microphone that is connected but can’t be started also falls back, and AeriVoice shows “Using the default microphone”. From 0.2.1, switching inputs during a dictation continues recording on the new input.
The result didn’t appear in my app.
Confirm Accessibility permission is enabled and the destination is an editable text field. Keep that field focused when finishing dictation. Secure fields and unsupported editors may require manual Paste. Follow AeriVoice’s status message: if it copied the result, use ⌘ + V to paste it yourself.
If you copy something else while dictation is finishing, AeriVoice preserves your new clipboard content; the transcript may not be copied over it.
When an app can’t confirm the paste, such as a terminal that redraws its screen, AeriVoice restores your previous clipboard after the delay chosen in Settings → General → Clipboard (5 seconds by default, or Never).
My provider won’t connect.
Check the key saved in Providers, your internet connection, and the provider account’s API credits or usage limit. Make sure the key belongs to the selected provider. Cloud providers are unavailable while Offline mode is enabled.
Local dictation isn’t ready.
Open Providers → Local models → Manage and check the selected engine’s download or language status. Finish setup while online before enabling Offline mode. Local transcription won’t switch to a cloud provider automatically.
Cleanup changed more than I wanted.
Try Faithful for edits closer to your wording. If you use Custom, review its instructions or clear them to return to Polished behavior. You can also try another cleanup model. Review important text before sending it.
Still stuck? Open an issue on GitHub with your app version, macOS version, and steps to reproduce the problem. Keep API keys and private dictation out of your report.
09 / Keep up to date
Updating AeriVoice
From 0.2.0-beta.1, use Settings → General → Check for Updates to install a newer version. Automatic checks run daily, update alerts are optional, and you choose when to install. Offline mode blocks update checks and downloads.
If you are updating from an older version without the built-in updater, download the latest DMG, quit AeriVoice, and drag the new app into Applications to replace it. Your existing settings and saved credentials are retained. Future updates can then be installed inside AeriVoice.
The download button follows the release marked Latest on GitHub. For release notes, earlier versions, or matching checksum files, visit GitHub Releases .
Download for Mac