The AI Agent App Privacy Policy

Effective date: May 22, 2026
Last updated: October 20, 2026
Applies to: The AI Agent App, an iOS personal AI assistant published by PrentusAI LLC.

Versions of the app before 1.5.0 were published under a different name. This policy covers every version.

This is the app's privacy policy. It is separate from the general PrentusAI website privacy policy. The app is its own product with its own data flows.

If you have questions about this policy or want to make a data request, email team@prentusai.com.

1. The short version

• The app runs on your iPhone. Your chats, your memory, your files, and the API keys you provide all live on your device, not on our servers.
• There is no PrentusAI server in the path between you and the AI models. The only PrentusAI service the app can talk to is an optional wake-up service for scheduled briefs, and it never sees what you ask (section 3.1). Version 1.5.0 does not have that option, so it never talks to the service.
• When you ask your agent something, your message goes directly from your phone to whichever AI service you have configured (Anthropic, OpenAI, Google, etc.). The AI service answers, your phone receives the reply, and the conversation stays on your phone. Voice mode runs on-device by default. When you speak to your agent, the transcription happens on your iPhone using Apple's built-in speech recognition, and the audio does not leave your device unless you explicitly opt into cloud transcription in Settings → Audio. Voice synthesis is fully on-device by default, so your agent's spoken replies are generated on your iPhone using an on-device neural voice model, downloaded once on first voice use and then cached. That audio never leaves your device unless you opt into a cloud voice provider (OpenAI or ElevenLabs) in Settings → Audio.
• We do not track you across other apps. We do not show ads. We do not include analytics, telemetry, or crash-reporting SDKs.
• You can delete every chat, memory, and cached file from inside the app at any time (Settings → Privacy & Memory → "Delete all data"). You can also remove every saved API key and disconnect every cloud account with one button (Settings → Privacy & Memory → "Sign out and remove API keys"). Uninstalling the app removes the rest.

2. What information the app handles

2.1. Information you create inside the app

• Chat messages and voice transcripts. What you type or speak to the assistant.
• Voice recordings. Captured only while voice mode is active. By default, transcription happens on-device via Apple's built-in speech recognition, so the audio never leaves your iPhone. If you have opted into cloud transcription in Settings → Audio, the audio is sent to your chosen provider (OpenAI Whisper, Groq Whisper, or Cloudflare Workers AI) and then deleted from your device immediately.
• Memory facts. Durable items you ask your agent to remember (preferences, profile facts, things you explicitly told it to recall).
• Attached files and photos. Documents or images you attach to a chat.
• In-progress drafts. Message text you type but haven't sent yet, auto-saved so you can come back to it.
• Action receipts. Local records of confirmed actions (calendar events created, reminders set, drafts prepared) so your agent can avoid duplicating work and so you can audit what was done.
• Learning signals. Thumbs-up / thumbs-down reactions and message retries that improve future replies.

All of the above is stored only on your device, in the app's private SQLite database and secure local storage.

2.2. Information the app reads from your phone (only when you grant permission)

• Contacts. Used to resolve a name like "text Sarah" into a contact, only when the conversation references a person.
• Calendar. Used to read your day or write new events, only when you ask.
• Reminders and local notifications. Used to schedule timers and reminders you create through chat.
• Location. Coarse (city-level) location, used only when you ask something location-aware (weather, "near me", etc.).
• Photos. Only when you explicitly attach a photo to a chat.
• Microphone. Only while voice mode is actively recording. Released when you end the recording or leave the voice screen. (Voice mode is iOS-only in v1. Android voice will ship after launch.)

You can grant or revoke any of these permissions from iOS Settings → AI Agent.

2.3. Information the app exchanges with your connected accounts (only when you connect)

Removed August 10, 2026: earlier versions of the app could connect Google Gmail (draft preparation, plus an optional read scope) and Google Drive (read-only search and import). Those connections were removed from the app on August 10, 2026, and nothing in the current app can connect a Google account. If you connected one in an earlier version, Sign out and remove API keys (section 5.1) deletes any Google sign-in tokens still stored on your phone. While a connection existed, data flowed only when you used it, under the OAuth scopes you explicitly granted.

2.4. Information that goes to AI model services

When you send a message, the message and the minimum context your agent needs to answer it go directly from your phone to the AI model service you have selected. Before answering, the app also makes one short routing call to that same provider, using its cheapest model and a PII-masked copy of your message, to work out which tools the turn needs. On the on-device path that routing step runs locally and makes no network call. Voice transcripts (the text of what you spoke) are sent in the same flow. The audio itself does NOT leave the device unless you opted into cloud transcription. Photos and attached documents are sent only when you include them in a turn that requires the AI to analyze them.

Voice synthesis output (the audio of your agent's spoken reply) also stays on your device by default. The on-device neural voice model generates audio locally and plays it through the speaker. The default voice (Piper "amy", en_US-amy-medium) is downloaded once on first use from Hugging Face (the public model repository at huggingface.co/diffusionstudio/piper-voices) and cached on your phone, and subsequent voice replies are produced fully offline. The text of your reply is not sent anywhere on this default path, only the model files themselves are fetched, one time. Audio is sent to a third-party voice provider only if you have explicitly opted into a cloud voice (OpenAI TTS or ElevenLabs) in Settings → Audio. In that case, the app sends the text of the reply to that provider, the provider returns audio, and the audio is then played on your device.

The list of AI model services the app can talk to (only those you configure) is maintained at https://prentusai.com/ai-agent-app/data-providers, along with a link to each provider's own privacy policy. The provider's policy governs how the provider handles your message after it leaves your phone.

3. Information the app does NOT collect

• No analytics, telemetry, or usage tracking. We do not include Sentry, Mixpanel, Amplitude, Segment, Datadog, PostHog, Firebase Analytics, Crashlytics, or any similar SDK.
• No advertising identifiers. We do not collect IDFA. We do not show ads. Apple's App Tracking Transparency prompt is intentionally not shown because the app does not track.
• No PrentusAI server in your conversations. No PrentusAI backend sits in the path between you and the AI services. Your messages, memory, files, and keys do not flow through us. The one PrentusAI service the app can talk to is the optional wake-up service in 3.1, which never receives what you ask. Version 1.5.0 does not have that option.
• No background data collection. When the app runs in the background (for example, to clean up old data or update the home-screen widget), it collects nothing about you and sends nothing to us, with one exception you control. If you turn on brief wake-ups, the app keeps the PrentusAI wake-up service up to date with a random ID, a push token, and the times of your briefs. Version 1.5.0 does not have that switch, so it sends us nothing. A few network calls can happen without you starting them, and each is written out in 3.1 below. One of them, a scheduled brief you created, does send your AI provider the context it needs, which can include your saved memory.
• No password or API key export. Your AI service keys are stored in the iOS Keychain and are read only by the relevant provider call.

3.1. Network calls you did not start

Nearly every network call the app makes is one you started. You send a message, run a tool, or open something that fetches. A few calls are not started by you, so they are described here instead of being covered by the word "only".

• App update checks. The app checks for updates at launch, when you come back to the app after switching away (throttled, so flicking in and out does not mean a request each time), and whenever you tap "Check for updates" in Settings. The check asks Expo's update service (u.expo.dev) whether a newer version of the app's code is available, and downloads it if there is one. The request carries the app version, the platform, and a random install ID that the update library creates on your iPhone, which Expo uses to count how many installs received an update. That ID is not tied to your name, your Apple account, or your phone's hardware, and like any web request the check reaches Expo with your IP address. It never contains your messages, your memories, or your keys. The app does not restart itself to apply an update. You either tap Restart, or the update applies the next time you open the app fresh.
• Scheduled briefs you created. If you have scheduled a brief (a morning summary, for example), The app may prepare that answer up to 30 minutes before it is due, while the app is in the background, by sending that brief's question to the AI service you selected. It is the same call your tap would have made, run a little early, and like that call it carries the context your agent normally uses to answer you, which includes your saved memory. It happens only for briefs you created and uses only tools that read. Delete the brief and the call stops.
• Brief wake-ups, only if you turn them on. No version released so far, including 1.5.0, has this switch, so nothing is sent for it. In a version that has it, this is how it works. If you turn on "Get briefs ready early" (Settings → When your agent reaches out), the app sends a small PrentusAI service, run on Cloudflare, three things: a random ID made on your iPhone, your iPhone's push notification address (an Expo push token), and the times of your scheduled briefs. About 20 minutes before each brief, the service asks Expo's push service, which passes the request to Apple, to wake the app so the answer can be written early with the AI service you connected. The service never receives the brief's question, title, or answer, or anything from your chats, memory, or keys. For a brief that repeats, the app sends the times for the coming week and keeps them up to date, so the service keeps this for as long as the brief repeats. It deletes what it holds after the last wake-up, when you turn the switch off, or when you use Delete all data. If the app stops sending times, for example because it was deleted, the service deletes what it holds within 7 days. Turning the switch off stops it.

4. Storage on your device

Uninstalling the app removes all of the above from your device.

5. Your rights and controls

5.1. In-app controls

• Delete all data (Settings → Privacy & Memory → Delete all data). Wipes every conversation, memory fact, document, action receipt, and cached file. If brief wake-ups are on, it also tells the wake-up service to forget this iPhone. Your saved API keys stay so the app keeps working after.
• Sign out and remove API keys (Settings → Privacy & Memory → Sign out and remove API keys). Removes every saved AI service key, and deletes any Google sign-in tokens left from versions before August 10, 2026. Your chats, files, and memory stay.
• Granular permissions (iOS Settings → AI Agent). Toggle any specific permission at the operating-system level.
• Per-message deletion (long-press a chat message). Removes a single message.

5.2. By request

If you cannot or prefer not to use the in-app controls, you can email team@prentusai.com with your request:

• Data access / export request. We will provide a JSON copy of data we can retrieve about your local installation within 30 days of receiving your request. Note: because the app's data lives only on your device, we generally do not have a server-side copy to give you. The in-app data is itself the export.
• Data deletion request. We will assist you in performing the in-app deletion above, or confirm that no PrentusAI-side data exists for your installation (the only PrentusAI-side data the app can create is the wake-up record in 3.1, which deletes itself within 7 days of the app's last update to it), within 30 days.

Please include enough detail for us to identify the request (a description of when and how you used the app is usually sufficient, as we do not require account identifiers because the app does not require an account).

5.3. Children

The AI Agent App is rated 17+. The app is not intended for children under 17. We do not knowingly collect information from children.

6. Third parties

Each AI model service you configure is a third-party processor. We do not control or supervise what those providers do with your message after it leaves your phone. Each provider publishes its own privacy policy:

• The complete current list of providers the app can be configured to use, with links to each provider's policy, lives at https://prentusai.com/ai-agent-app/data-providers.
• If you do not configure a provider, no data flows to it.

If you turn on brief wake-ups (3.1), Cloudflare hosts the PrentusAI wake-up service, and Expo and Apple deliver the wake-up to your iPhone. They handle the random ID, the push token, and the times only. Expo's privacy policy is at https://expo.dev/privacy.

Google Gmail and Google Drive connections were removed from the app on August 10, 2026. While they existed, they were governed by Google's privacy policy and the OAuth scopes you explicitly granted.

We do not sell your personal information. We do not share your personal information with third parties for their marketing purposes. We do not use your data to train AI models.

7. Security

• API keys and OAuth tokens are stored in the iOS Keychain, encrypted by iOS on your device.
• All network calls to AI providers and cloud connectors use HTTPS.
• We do not include custom encryption beyond the standard iOS keychain and TLS. We have declared ITSAppUsesNonExemptEncryption = false to Apple, which is accurate given that posture.
• We cannot recover your local data if your device is lost, your iPhone is reset, or the app is uninstalled. There is no cloud backup of your app data on our side.

8. International data transfer

Because data flows directly from your device to the AI model service of your choice, the geographic location where your messages are processed depends on the provider's infrastructure. Each provider publishes the regions in which they operate. Apart from the optional wake-up service in 3.1, which runs on Cloudflare's global network and holds only a random ID, a push token, and times, PrentusAI does not receive your data, so it does not transfer your data across borders.

9. Changes to this policy

If we change this policy, we will update the "Last updated" date at the top. Material changes will be noted in the app's release notes when the new version ships.

This is a hosted document at https://prentusai.com/ai-agent-app/privacy. The version you accepted when you installed the app is captured in your App Store receipt; if a future version of the app requires acceptance of a new policy, the app will prompt you.

10. Contact

PrentusAI LLC
30 N Gould St, Ste R
Sheridan, WY 82801
United States

team@prentusai.com

This policy was written for the app's v1, the iOS App Store launch build, by PrentusAI LLC. It is not legal advice. If you have legal questions about this policy as it applies to your jurisdiction, please consult counsel.