Privacy Policy

Account & billing

When you create an OpenWhispr account, we collect the minimum needed to manage it:

• Email/password sign-up: email, name, and a hashed password (we never store your password in plain text)
• Google, Microsoft or Apple sign-in: email, name, and basic profile from that provider. We never receive your password for it.
• Paid plans: billing details processed by Stripe. We store your customer ID and subscription state, never your full card number.
• Session records for authentication and automatic session refresh, including the IP address and browser or device type of each signed-in session, which we use to keep your account secure
• Website cookies for analytics and advertising, subject to your cookie choices (opt-in in the EEA, UK and Switzerland, opt-out elsewhere). See our Cookie Policy.

Voice, audio, and transcriptions

The desktop app offers three transcription modes; the mobile app offers local processing and OpenWhispr Cloud. You pick the one that fits your needs and can change it at any time in Settings.

• Local processing. OpenWhispr can run speech-to-text fully on your device using local open-source models. When you choose this mode, no audio is transmitted anywhere.
• Bring your own API key. If you provide your own API key for a third-party transcription provider, your audio is sent directly to that provider for processing under your account. Their privacy policy applies. We don't see, store, or proxy that audio.
• OpenWhispr Cloud. Our managed transcription service. Audio is sent to our API, or straight to one of our speech-to-text providers for live transcription, processed in real time, and not stored by us. We keep a usage record of each transcription (word count, provider, model and length) to run and bill the service, but not the text itself unless you turn on cloud backup. We don't use your data to train AI models. See our Security page for training opt-outs; the current provider list is in the Subprocessors table below.

Audio Retention (desktop app, on by default). The desktop app keeps recorded dictation audio on your device for 30 days by default, so you can review or re-process it, then deletes it automatically. You can choose a shorter or longer period (up to 90 days) or turn it off in Settings. Retained audio is never synced to OpenWhispr Cloud; it leaves your device only if you choose to re-transcribe a recording, when it goes to the transcription service you have selected. Meeting recordings are not kept (see System audio capture below).

Transcribed text is stored in a local SQLite database on your device. In the desktop app, if you turn on cloud backup (off by default, available on paid plans), your transcriptions — the final text and the raw text before cleanup — are also stored in our cloud database so they appear on all your devices. The mobile app keeps dictation history on your device only.

Custom dictionary. If you add words or phrases (technical terms, names, jargon) — or OpenWhispr learns them from your corrections — they're stored on your device and sent to your transcription and AI cleanup provider as a hint. If cloud backup is on, your dictionary also syncs to our cloud database. They're not used for anything else.

Notes and AI agent conversations

Notes are stored on your device in a local SQLite database. A note can contain a title, body content, AI-enhanced versions (if you use the enhancement feature), and — for meeting notes — the meeting transcript and participant list pulled from your calendar invite.

Cloud backup and sharing (Optional). If you turn on cloud backup (available on paid plans), your notes, folders, dictionary and snippets — and, in the desktop app, your agent conversations and transcriptions — are stored in our cloud database (Neon, encrypted at rest) so you can access them across devices. Backup is off by default in the desktop app and on by default in the mobile app, and you can turn it off at any time. Notes you share, and notes in team spaces, are stored in our cloud whatever this setting says. When a note is in our cloud, we send its text to OpenAI to index it for search. Audio is never synced. Insights sync, a separate opt-in, uploads content-free counters (word counts, the date and time of each dictation, durations and model types).

The AI Agent (Optional). When you use the AI agent, your prompt and any tool results (notes it reads, calendar events it queries, web searches, clipboard contents you ask it to handle) are sent to whichever AI provider you've configured, or stay on your device if you run it on a local model. Whichever model you use, web searches the agent runs go through OpenWhispr to our search provider, Exa, and if cloud backup is on, note searches are answered by our cloud, which uses OpenAI to process the search. The agent no longer keeps a long-term memory of your conversations.

Local semantic search. Notes are indexed for search on your device by a small embedding model that runs locally, and that on-device index never leaves your machine. Notes stored in our cloud are also indexed there, as described above.

Calendars (Optional)

Connecting a calendar is optional. You can connect Google Calendar or a Microsoft Outlook / Microsoft 365 calendar, or, on a Mac, let OpenWhispr read the Calendar app. OpenWhispr reads your calendar list and upcoming events (read-only — we never create, edit, or delete anything) and uses that data to:

• Notify you when a scheduled meeting is about to start
• Suggest a title for your meeting note based on the event name
• Attach the calendar invite participants to that note

Upcoming events are cached locally alongside your notes and transcriptions. The connection runs directly between your device and your calendar provider, and we don't store your calendar on our servers, share it, or use it for ads or AI training. Event details reach our servers and your AI provider only when you ask the AI agent about them, and a meeting note's title and participant list sync with the note if it is backed up or shared.

Disconnect from OpenWhispr settings to delete the local cache, or revoke access anytime via your Google Account permissions or your Microsoft account's app permissions.

Limited Use. OpenWhispr's use and transfer of information received from Google APIs to any other app adheres to the Google API Services User Data Policy, including the Limited Use requirements.

Meeting detection signals

OpenWhispr can detect when a meeting is happening so it can offer to take a note. To do this it monitors two things on your device only:

• Whether known meeting apps (Zoom, Teams, Webex, FaceTime) are running on your computer
• Whether any application is currently using your microphone (on macOS, Windows, and Linux)

OpenWhispr does not capture what those other apps are doing, what they're recording, or who's on the call. It only knows the binary fact that "a meeting app is open" or "the mic is in use." These signals are held in memory only and never sent anywhere (if you turn on debug logging, the app's local log file notes which meeting app was detected). You can disable meeting detection in Settings.

System audio capture (Optional). OpenWhispr can record both your microphone and the audio coming from other apps so it can transcribe both sides of a meeting. This works on macOS 14.2 or later, where you must first grant macOS's "System Audio Recording" permission, and on Windows and Linux. Meeting audio, from your microphone and from other apps, is processed as you record and is not kept. To tell speakers apart, the app may write a temporary working copy to your computer, which it deletes after processing.

Accessibility-based features

The first time you use OpenWhispr to paste a transcription, it asks for your operating system's accessibility permission. We use this for two things:

• Auto-paste: pastes your transcribed text into whatever text field you have focused, so you don't have to copy and paste manually.
• Correction learning: for up to 30 seconds after a paste, OpenWhispr watches the focused text field for edits you make to the text it pasted. If you fix a misspelled word, OpenWhispr adds the corrected word to your custom dictionary (see above) so future transcriptions are more accurate. The text it watches is never sent to OpenWhispr or any third party. You can turn this off with "Auto-learn from corrections" in Settings.

You can revoke accessibility permission in your OS settings at any time, which disables both features.

Service logs (OpenWhispr Cloud)

When you use OpenWhispr Cloud, our servers record technical details about each request so we can run, bill, support and fix the service:

• Audio length, size and format
• Speech-to-text provider, model, processing time, language and word count
• AI model provider, model, processing time and token counts
• Round-trip timing and error information, including error messages from our providers
• Client information (client type and app version)

These records are linked to your account and to a random ID for each app session, and they never include your audio or the text you dictate. We use them to run, support and improve OpenWhispr and to see which marketing channels bring active users; they are never sold or shared with advertisers. They are recorded for OpenWhispr Cloud requests whatever your settings — the "Usage analytics" setting in the desktop app does not currently turn them off.

Referral program (Optional)

If you participate in the referral program, we collect:

• Your unique referral code and the invitations you send
• The email addresses of people you invite (uploaded to our servers so we can track conversions and prevent abuse)
• Visits to your referral link (IP address, user agent, referring URL)
• Whether a referral results in a sign-up or subscription

We only email a recipient when you initiate the invite. Referral data is used to attribute rewards and prevent abuse — nothing else.

Mobile apps: analytics, attribution and in-app purchases

Our mobile apps include a small number of third-party software kits (SDKs) that the desktop app does not. They help us understand how the apps are used and which marketing works, and they process in-app purchases. We use what they collect to improve the product and to measure and grow our marketing. None of them ever receives your audio, transcriptions, notes or agent conversations.

• AppsFlyer (attribution and app analytics). Tells us which ads and campaigns lead to installs and subscriptions, and how the app is used, so we can improve the product and spend our marketing budget well. It receives device and advertising identifiers, IP address, device and OS details, and app events such as install, app open, tutorial completed, upgrade screen viewed, transcription completed (the source and provider only, never the text) and subscription started, together with a small set of non-content properties such as which upgrade screen was shown and which product was bought. On iOS, AppsFlyer runs only if you allow tracking when Apple asks; if you decline, nothing is sent to it. AppsFlyer may also be used to measure marketing for our desktop app, where it receives the same kind of data and never dictation content.
• RevenueCat (in-app purchases). Records and verifies purchases made through the App Store or Google Play so your plan stays in sync across your devices and we can understand subscription revenue. It receives a random billing identifier for your account (not your email), the store purchase receipt, your subscription status, device type and operating system, and IP address.
• Superwall (upgrade screens). Powers the upgrade screens (paywalls) in the app and lets us test which designs and prices work best, so we can improve the upgrade experience. It receives the same random billing identifier, device, OS and app version, your plan and usage status (plan, trial state, words used and remaining, processing mode) and upgrade-screen events such as shown, dismissed and purchased.
• Your controls. Turn off Usage Analytics in the app's Privacy & Data settings to stop analytics collection. On iOS you can also change the tracking permission at any time in Settings → Privacy & Security → Tracking. RevenueCat and Superwall are needed to buy and manage in-app purchases and cannot be switched off separately.