Privacy Policy
Effective date: September 18, 2026
ShiftPearls is designed for private professional reflection and learning. It is not a patient record. Do not enter patient names, medical record numbers, exact dates, contact details, exact locations, images, or any other information that could identify a patient or another person.
Data stored on your iPhone
Your saved reflections, optional shift-experience ratings, tags, learning questions, local AI usage counts, app preferences, and reminder settings are stored on your iPhone. ShiftPearls does not provide an account or cloud sync in this version.
On-device voice transcription uses Apple's speech recognition on the device and does not create a saved audio file. Optional cloud medical dictation is separate and is described below.
You can introduce yourself with a local nickname, professional role, experience, learning interests, and feedback preferences. Personal memory is enabled by default: saved reflections, including existing saved reflections, automatically contribute selected source excerpts and themes from your own words. Readable profile.json and memory.json files remain inside this app's protected local storage and are excluded from backup. The nickname is never included in AI requests. Settings and Insights let you correct, pause, or forget memory. The app does not infer a clinical ability score, diagnosis, or personality label.
When data leaves your device
Optional AI content uploads send reflection text, transcript text, or audio only for the feature you choose and only after the relevant in-app confirmation. RevenueCat subscription services are separate from AI content uploads: the RevenueCat SDK may connect when the app starts or when purchase/settings surfaces load so the app can fetch offerings, purchase status, and anonymous entitlement data. The app uses transport encryption for network requests. Before text AI processing, ShiftPearls checks text on your device for common identifier patterns and asks you to confirm that the text is de-identified. Automated checks can miss information, so you remain responsible for reviewing the text before sending it.
The ShiftPearls gateway is hosted on Vercel, whose infrastructure processes the text or audio you choose to send while relaying the request to the named AI provider. The gateway is designed not to persist or log request or response bodies. Vercel may also process technical connection information, such as IP address, user agent or client information, requested URL, request time, and operational logs, to provide hosting, security, routing, and troubleshooting.
The gateway may receive an anonymous installation identifier and, when RevenueCat is configured, an anonymous RevenueCat app user identifier so it can enforce free-use limits, rate limits, and Pro access. The gateway may send the RevenueCat app user identifier to RevenueCat to verify entitlement status. These identifiers are not forwarded to MiniMax or Deepgram by the ShiftPearls gateway, are not a ShiftPearls account, and are not used to publish or sync your journal.
AI reflection generation with MiniMax
If you choose Create reflection after the privacy check, ShiftPearls sends the following content through the ShiftPearls gateway to MiniMax, a third-party AI provider:
- the current de-identified reflection text;
- the locale;
- if you leave personal context enabled for that request, your safe professional introduction and up to five relevant or recent local memory entries shown on the consent screen.
The anonymous installation identifier is used by the ShiftPearls gateway for rate limits and free-use counts. The optional RevenueCat app user identifier is used by the ShiftPearls gateway and RevenueCat for entitlement checks. ShiftPearls does not forward these identifiers to MiniMax.
MiniMax uses the submitted text to return structured reflection fields, such as summary, decision points, strengths, uncertainties, learning questions, tags, optional encouragement, and optional wind-down suggestions. Audio is not included in this MiniMax reflection request. AI output is a draft that you must review and may edit before local save.
MiniMax's China Open Platform privacy policy describes processing submitted API data to produce results and investigate reported problems. It also permits use of de-identified information for analysis and commercial use, including analysis and commercialization of databases containing that information. This additional use is disclosed before you consent to either MiniMax feature. De-identification does not mean that your text is never retained or used again.
MiniMax states that it retains personal information for the shortest period necessary for the relevant purpose, subject to legal and other stated exceptions, then deletes or anonymizes it. Its policy permits continued use of anonymized information. ShiftPearls does not promise zero retention or a prohibition on all model-improvement uses. Do not submit patient-identifying or confidential information. These practices also apply to the optional transcript corrections described below.
Optional transcript correction suggestions with MiniMax
After dictation, you may choose Review dictation with AI. This runs a local privacy check and requires a separate confirmation before the displayed transcript text is sent through the ShiftPearls gateway to MiniMax.
This request includes:
- the current transcript text;
- the locale.
The anonymous installation identifier is used by the ShiftPearls gateway for rate limits and free-use counts. The optional RevenueCat app user identifier is used by the ShiftPearls gateway and RevenueCat for entitlement checks. ShiftPearls does not forward these identifiers to MiniMax.
This request does not include audio, nickname, profile, personal memory, or saved reflection history. MiniMax returns possible wording corrections. Suggestions are shown alongside the original wording and start unselected. Only changes you select and confirm replace the current editor text. MiniMax cannot verify the audio or reliably reconstruct missing speech; you must check medical terms, quantities, negations, and clinical meaning manually.
Optional cloud medical dictation with Deepgram
Cloud medical dictation first records a temporary M4A audio file locally. Uploading requires a separate explicit audio confirmation after recording. If you confirm upload, the original audio recording is sent through the ShiftPearls gateway to Deepgram, a third-party speech-to-text provider, using the Nova-3 Medical model.
This request includes:
- the original audio recording;
- audio content type and request metadata needed for transcription.
The anonymous installation identifier is used by the ShiftPearls gateway for rate limits and free-use counts. The optional RevenueCat app user identifier is used by the ShiftPearls gateway and RevenueCat for entitlement checks. ShiftPearls does not forward these identifiers to Deepgram.
Audio cannot be automatically de-identified before upload. Do not include identifiers in recordings. The gateway forwards transient audio bytes without intentionally storing audio files or logging bodies. ShiftPearls sets Deepgram's mip_opt_out=true request option for these transcription requests. This opts the request out of Deepgram's Model Improvement Program under Deepgram's current documentation, but it does not create a healthcare business associate agreement or a blanket zero-retention guarantee.
The app keeps the temporary local recording while you review the transcript, so you can replay it and edit misheard words. It removes the recording when you use the transcript, discard it, or close the recording screen. After a network error a local copy remains for an explicit retry. Unfinished foreground recording is discarded when the app enters the background. You review and can edit the transcript before any separate text privacy check or MiniMax reflection request.
Purchases with RevenueCat and Apple
RevenueCat processes an anonymous app user identifier, purchase status, product identifiers, receipt and entitlement information, and related subscription events so ShiftPearls can provide and restore Pro access. The RevenueCat SDK may connect when the app starts and when purchase/settings surfaces load, even if you do not start an AI request. Apple processes App Store payments. ShiftPearls does not receive your full payment-card details.
Reflection and sharing
Generated case summaries describe only the events and management you supplied. Reflection, encouragement, optional learning habits, and wind-down suggestions are editable AI interpretations, not a clinical assessment. Personal memory remains tied to your original words; generated suggestions are not automatically treated as actions you took or goals you chose.
Share card creates a front with the case and described management and a back with your reviewed insight and next-shift action. Your optional shift rating and the private encouragement and wind-down sections are not included. A privacy check runs before exporting. No raw transcript, nickname, profile, or memory history is automatically added to the card. You choose the recipient or destination in the iOS share sheet. Images and a portable interactive card file are generated locally; the app does not publish a public link. Copies you share leave the app's private storage and are controlled by the destination you choose.
Notifications
If you enable learning reminders, notification text is generic and does not contain your reflection or learning question. Reminder scheduling occurs on your device.
Analytics and tracking
This version does not include advertising or cross-app tracking SDKs. Product counts used by the app are stored locally and do not contain reflection text.
Deleting data and withdrawing consent
You can delete all local reflections, learning prompts, your introduction, and personal memory from Settings. This also clears local AI usage and pending learning notifications. Purchase records held by Apple or RevenueCat are governed by those providers and may be retained for transaction, fraud-prevention, tax, and legal purposes.
Forgetting a memory excludes that source from future automatic updates and AI context but leaves its original reflection in Library. Deleting a reflection also removes its associated memory. Turning memory off preserves existing local memory but stops accumulation and its use in AI requests. Previously saved comparisons remain historical reflection content until you edit or delete those reflections.
You can withdraw future consent for optional AI content uploads by not confirming a cloud request, by excluding personal context from a specific MiniMax reflection request, by turning memory off, by discarding a cloud dictation recording instead of uploading it, or by using manual reflection mode. RevenueCat purchase-status processing is controlled separately through purchase settings, subscription status, App Store purchase flows, and RevenueCat/Apple retention rules. Withdrawing consent does not retroactively withdraw a request already sent to MiniMax, Deepgram, Vercel, RevenueCat, or Apple. Provider-side access, deletion, and retention rights depend on the relevant provider's policy and applicable law.
Third-party protection and limitations
ShiftPearls requires third parties receiving user data to provide the same or equal protection as the privacy and security safeguards described in this policy. For the limited data and purposes described here, we rely on the providers' published commitments to security measures, restricted access, disclosed processing purposes, and applicable data-access and deletion rights. Our assessment is based on the following protections:
- MiniMax's China Open Platform privacy policy describes encrypted storage, access controls, confidentiality requirements for partners, retention limited to stated purposes with specified exceptions, and access, correction, deletion, and consent-withdrawal rights.
- Deepgram publishes technical and organizational security measures and data-rights procedures. We set
mip_opt_out=trueon each transcription request to exclude that request from its Model Improvement Program. - RevenueCat's standard Data Processing Agreement requires documented processing instructions, technical and organizational safeguards, protections for subprocessors, and assistance with applicable data-rights requests.
- Vercel's Terms of Service and Privacy Notice describe security safeguards and its processing of hosted content and technical information.
- Apple describes safeguards and privacy rights for App Store transactions in its privacy policy.
ShiftPearls further limits sharing through feature-specific AI permission, transport encryption, content minimization, and a gateway designed not to log content. These protections do not mean identical retention periods across providers or a promise of zero risk. The processing purposes and retention limitations described above remain applicable.
No automated privacy detector is perfect. Because third-party AI and transcription providers may process or retain submitted data under their own policies, ShiftPearls is not for patient records, legally required documentation, patient care, emergencies, diagnosis, treatment decisions, or any workflow requiring protected health information processing by a provider under a verified healthcare business associate agreement.
This website and hosting
The website and app gateway are hosted on Vercel. Visiting the website or making a network request transmits connection information such as your IP address, browser or client information, requested URL, and request time to the hosting infrastructure. The website does not include advertising, analytics scripts, a contact form, or a user account. Infrastructure providers may process technical logs for operation, security, and troubleshooting under their own policies. The app's private local journal is not published on this website.
Third-party services
These providers may process information outside your country. Their policies describe their own processing, security, retention, and deletion practices.
- MiniMax API privacy policy: optional AI reflection generation and transcript correction suggestions. This redirects to the MiniMax China Open Platform privacy policy.
- Deepgram privacy policy: optional cloud audio transcription.
- Deepgram Model Improvement Program documentation:
mip_opt_out=truebehavior for opted-out requests. - RevenueCat privacy policy: subscription access and restoration.
- RevenueCat Data Processing Agreement: safeguards for customer data processing.
- Apple privacy policy: App Store payments and Apple platform services.
- Vercel privacy notice: website and gateway hosting.
- Vercel Terms of Service: hosting and security commitments.
- Deepgram Terms of Service: processing and request-level training opt-out.
Policy updates
This page will be updated when these data practices change. The effective date identifies the current version.
Contact
See the Support page for help managing local information, privacy controls, and purchases. This website does not currently provide a support-message submission form.