VoiceInput Privacy Policy.

The short version

• While you dictate, your voice is streamed to a cloud speech engine (ElevenLabs by default) that sends back the text. VoiceInput keeps no audio from dictation.
• Grammar and Custom vocabulary clean-up are on by default, so the text of each dictation goes to a cloud AI model (Qwen on Cerebras by default) to be tidied.
• Screen context is on by default: with each cleaned-up dictation, the app name, window title, the text around your cursor and the text visible in the window go to that same model.
• Recordings, meeting notes and screenshots are sent to cloud AI services and saved as tasks in your account.
• Your account data is stored with Google Firebase. No ads, no tracking, nothing sold.
• Most of these can be turned off in Settings, described below.

1. Who we are and what this covers

VoiceInput for Mac and VoiceInput for Windows (“VoiceInput”, “the app”) are developed and operated by Sharath K., an individual developer based in Texas, USA (“we”, “us”). You can reach us at sharathck3@gmail.com.

This policy covers the two desktop apps, including their capture shortcuts. VoiceType for iPhone and VoiceType for Android have their own policies, linked at the end. If you use VoiceInput with a VoiceType account, the account data described in those policies is the same account.

2. Your voice when you dictate

• Where it goes. While a dictation is on, VoiceInput streams your microphone audio over an encrypted connection to the speech engine selected in Settings, which sends back the text. The default is ElevenLabs Scribe v2 Realtime. The alternatives are Microsoft (MAI-Transcribe or Azure Speech), Meta (Muse Voice Transcribe), Google Gemini, OpenAI and xAI Grok. Only the audio is sent to the speech engine; your vocabulary and prompt are not.
• No separate consent screen. Unlike the phone apps, VoiceInput does not show an AI consent screen. Cloud speech is how dictation works by default. Starting a dictation with the shortcut is what sends your voice.
• No audio is kept. VoiceInput keeps the words of a dictation, not its audio.
• Built-in dictation. You can pick Built-in dictation in Settings to use the computer's own recognizer, and VoiceInput also falls back to it when the network is unavailable. On the Mac this is Apple's speech framework, which on macOS 14 and 15 may send audio to Apple's servers unless on-device recognition is available. On Windows it is the Windows speech recognizer.
• Windows keeps the microphone ready. After a dictation, VoiceInput for Windows keeps the microphone open for up to 45 seconds so the next dictation starts instantly (the setting Keep microphone ready, on by default). While it is open, about 1.2 seconds of audio is held in memory only. If you start another dictation in that time, that audio becomes its first moment; otherwise it is never saved or sent. Turn the setting off to close the microphone as soon as a dictation ends. The Mac has no such window.

3. Clean-up of the text

Clean-up has four separate switches in Settings › Post-processing. Their defaults are: Grammar on, Custom vocabulary on, Tone off and User prompt off.

When any clean-up step is on, the text of each dictation is sent to the clean-up AI model selected in Settings, together with what that step needs: your custom vocabulary, your grammar instructions, the wording of the tone you picked and, if it is on, your User prompt. The default model is Qwen on Cerebras. The alternatives are Anthropic Claude, OpenAI, Google Gemini and Inception Mercury. The audio is never sent for clean-up.

Clean-up only tidies what you said. It never answers your dictation or writes new content, except with a code tone (such as Java code), which writes the code you describe, whether you pick it as your tone or press the Java code shortcut (⌥J on the Mac, Alt+J on Windows). Picked as your tone, it uses your clean-up model; the shortcut always uses Anthropic's Claude, whichever model you picked.

To keep a dictation's text off the cloud clean-up models, turn all four clean-up switches off (and don't use the Java code shortcut).

4. Screen context

Screen context is on by default and is used whenever a clean-up step runs. With that dictation, VoiceInput reads:

• the name of the app you are dictating into;
• the window or browser tab title;
• the text around your cursor in the field you are typing in;
• the text visible in the window, read from the screen (on-screen text recognition).

It sends this to the clean-up model with the dictation so that names, code and terms match what you are working on. It never reads a password field. VoiceInput does not keep it; the provider processes it under its own terms. Turn it off in Settings › Post-processing. The Java code shortcut always reads the screen, even when this switch is off.

5. Capture shortcuts

Recordings, meeting notes, screenshots and adding tasks need you to be signed in; on Windows, reading selected text aloud also works signed out. What each one does with your data:

• Record audio and single-speaker recording capture your microphone and your computer's sound, so other people on a call are recorded too. The original audio file is uploaded to your account's storage unless you turn off Keep the original audio; an Audio only recording is always uploaded, because it is the only copy.
• Who transcribes a recording. Record audio is transcribed by your account's recording engine: ElevenLabs by default, or Microsoft, Google Gemini, OpenAI or xAI Grok if you set another in the To-Do List app. That engine is used whatever speech engine VoiceInput uses for dictation. A single-speaker recording is transcribed by the engine that matches your dictation speech engine (by your recording engine when you dictate with Meta or Built-in dictation), and while you record it also streams live words to Microsoft, or to ElevenLabs when ElevenLabs is your speech engine.
• Titles and notes. The transcript of a recording goes to your Notes AI model, which writes the task's title and notes: Qwen on Cerebras by default, or Google Gemini, OpenAI or Anthropic Claude if you set another in the To-Do List app. If the default Qwen model fails, Google Gemini is the fallback.
• Meeting notes capture your microphone and your computer's sound and stream them live to Meta's speech engine, which labels the speakers. The notes are kept as text.
• Screenshot to task captures the whole screen or a region you drag. Google Gemini reads it, and the image is saved with the task.
• Speak selected text sends the selected text to Microsoft Azure text-to-speech, which returns the audio you hear.
• Add selected text as a task saves that text to your to-do list.
• Load audio and Load image send a file you pick through the same path as a recording or a screenshot, and the result becomes a task.

Every result is saved as a task in your account's to-do list, the same list the To-Do List app shows. If you record other people, tell them, and get their consent where the law requires it. You are responsible for having the right to record.

6. Who processes your data

Which provider receives a request depends on the feature: dictation and clean-up use the models picked in VoiceInput's Settings, while recordings, meeting notes, screenshots and read-aloud use the providers named in section 5. Each provider processes a request under its own terms and privacy policy.

We do not use your data for advertising, we do not sell it, and we do not train our own AI models on it. AI providers may keep request data for a limited time under their own policies, for example to monitor abuse.

7. What is stored, and where

In your account, with Google Firebase (the same project as the VoiceType apps):

• your email address and sign-in (Google or email, handled by Firebase Authentication; we never see your password);
• your custom vocabulary, settings, grammar instructions, tones and User prompt, with earlier versions so you can restore them;
• your dictation history: the text you dictated (as recognized and after clean-up), with the time, the speech engine and which device it came from. Never the audio of a dictation;
• the tasks created by the capture shortcuts, with screenshot images and, unless you turned it off, the original audio of recordings.

On your computer: a local copy of your dictation history, the settings that belong to that computer, recordings waiting to be transcribed, and the screenshots you captured (each kept for 30 days after it is saved to your account). Each transcript is also copied to your clipboard.

Signing in and the subscription. VoiceInput for Mac checks with our service whether your account holds an active VoiceType Monthly subscription, which you buy on iPhone through the App Store. When you are not signed in, VoiceInput (Mac or Windows) may create an anonymous Firebase account only to fetch its service configuration; nothing else is stored with it. The same sign-in is shared with the developer's other apps, VoiceType, To-Do List and GenAI Studio.

8. How long we keep it

Data in your account is kept until you delete it, delete your account, or ask us to delete it. Deleting the account in the VoiceType phone app removes the account and its VoiceType data, but the recordings, screenshots and tasks made with VoiceInput are deleted only when you email us (section 10). Data on your computer stays there until you delete it or remove the app and its data. Copies in our providers' backups are removed on their normal backup cycles.

9. Your choices

• Speech engine: pick a different engine, or Built-in dictation, in Settings.
• Clean-up: turn Grammar, Custom vocabulary, Tone and User prompt on or off in Settings › Post-processing.
• Screen context: turn it off in Settings › Post-processing.
• Keep microphone ready (Windows): turn it off in Settings.
• Keep the original audio: turn it off so recordings are not uploaded (an audio-only recording is always uploaded).
• Permissions: on the Mac, Microphone, Accessibility and Screen & System Audio Recording can be turned off in System Settings › Privacy & Security. On Windows, the microphone can be turned off in Settings › Privacy & security › Microphone.
• Signing out stops syncing and turns off the capture shortcuts; on the Mac it also locks the app until you sign in again.

10. Deleting your account and data

VoiceInput for Mac and Windows has no delete button. You can:

• delete the account in the VoiceType app on iPhone or Android, from its account settings. That removes the account and its VoiceType data (settings, vocabulary, dictation history), but not the recordings, screenshots and tasks made with VoiceInput; or
• email sharathck3@gmail.com from your account's address with the subject “Delete my VoiceType account”. We will delete the account and its data, including the recordings, screenshots and tasks created by VoiceInput, and confirm by email.

If you used the capture shortcuts, email us even if you deleted the account in the phone app: recordings, screenshots and tasks made with VoiceInput are deleted only on request.

Because the sign-in is shared, deleting it deletes that account in the VoiceType apps, To-Do List and GenAI Studio too. Deleting your account does not cancel an App Store subscription; cancel that on your iPhone in Settings › [your name] › Subscriptions.

11. Security

Audio, text and account data travel over encrypted connections and are encrypted at rest by Google Cloud. Database rules let only your signed-in account read its own data. No system is perfectly secure, so please use a strong password for an email sign-in.

12. Your rights

You can ask us for a copy of your data, a correction or its deletion by emailing sharathck3@gmail.com from the address on your account. We may need to confirm the request comes from you, and we will respond within the time the law requires. Depending on where you live (for example the EEA, the UK or a US state with a privacy law), you may have further rights, including to complain to your data protection authority. Your account data is stored in the United States, and providers may process requests in other countries where they operate.

13. Children

VoiceInput is not directed to children under 13, and we do not knowingly collect personal information from them. If you believe a child has given us personal information, contact us and we will delete it.

14. Changes to this policy

If we change this policy, we will post the new version on this page with a new effective date. If a change materially affects how we use your information, we will tell you in the app or by email before it takes effect.

15. Contact

Sharath K. · Texas, USA · sharathck3@gmail.com

The other policies: iPhone Privacy Policy · Android Privacy Policy · Privacy, in plain words