This page is the same version of the Privacy Policy shown in the app on first launch (2026-10-19).

Privacy Policy

Version: 2026-10-19

Lyrune (月晤, "the app") is developed and provided by Limburgite Inc. ("we"). This policy explains what information the app handles, why, where it is stored, and what choices you have.

The most important points

• Your chats, AI friends, personas, memories, images, and voice recordings are stored encrypted on your device. We have no server that collects this content, and we cannot see it.
• By default the app uses AI models, speech recognition, speech synthesis, and image recognition that run on your device, so content does not leave it.
• Conversation content is sent to an online provider only after you separately agree and turn on an online model, and only to the provider you chose. You can disconnect at any time.
• We do not sell your personal information, train any model on your content, or use it to show or tailor ads. The app contains no ads and no third-party analytics.

Information the app handles

• Birth year and month: used only to confirm you are 18 or older; not saved if you are under 18. Stored on your device.
• Country or region: chosen by you, used to show local help lines. The app does not read your location.
• Chat content: chat history, AI friends, personas, style settings, saved lines, memory notes, conversation summaries, and letters, used to provide chat and companionship features. Stored encrypted on your device.
• Content about emotions or mental state: may be part of your chats; used for companionship and to detect risk and show help information. Stored encrypted on your device.
• Images and voice: original images and recordings you send, and avatars and chat backgrounds you set, stored encrypted on your device; recognition happens on the device.
• AI voice replies (optional, off by default): after you turn them on in Settings → Voice → Voice replies, the app turns an AI reply into speech on your device and plays it only when you tap play. Synthesis uses an open-source speech model installed with the app and does not use the network; the audio is kept only in memory and never written to a file; the most recent clip stays in memory for replay or saving until you leave the chat or turn off Voice replies. An audio file is created only when you choose "Save voice"; you pick where it goes, the file is not encrypted, and it is marked as AI-generated. Each clip starts by announcing "The following speech is generated by artificial intelligence." The app does not offer voice cloning and does not collect or imitate anyone's voice.
• Consent records: which terms you agreed to and when, kept as a record of your choices.
• Emergency contact (optional): name and phone number, stored only on your device and used only to place a call after you confirm it yourself.
• Safety events: only the type and time, without chat content, used to check that safety features work.
• Settings and preferences: appearance, selected model, playlist, unsent drafts, and similar, stored on your device.
• Device memory size: read on your device during first use, only to suggest a suitable model. It is not saved or uploaded.
• Usage time: the app counts, on your device, how long it has been open in the foreground today and shows it on the home screen. Only today's total is stored, and it is not uploaded.
• API key for online models (optional): provided by you, stored in the system's secure storage, and used only to connect to the provider you chose.

Device permissions

• Microphone: requested only when you use voice input; audio is turned into text on your device.
• Photos and files: through the system photo or file picker, only the images, recordings and documents you select are read.
• Notifications: used for status such as "replying" and playback controls; notifications contain no chat content.
• Network: used only for online features you choose to use (online models, web check, online media, model downloads).
• The app does not request location, contacts, SMS, or call log permissions.

When data leaves your device

• Online models: after you separately agree and turn one on, the app sends the current conversation, your question, persona settings, memory notes, and conversation summaries to the provider you chose so it can generate replies, assess risk, and review replies. In long chats, earlier messages that are not sent directly are also sent to the same provider to produce a summary. When you tap to create a letter recap, that day's conversation is sent. If the model you choose can read images, images you send are also sent for image recognition (on by default; you can turn off “Let the provider read images” on the provider page). The confirmation shown before you turn it on names the provider and the regions where data may be processed (for example China, the United States, Singapore, or the EU). The provider's own privacy policy governs how it handles that data. Speech recognition and speech synthesis always happen on your device and are never sent to the provider.
• Web check: if you turn it on with a supported provider, your question or keywords may be used by that provider for web searches.
• Fetching model lists and checking the connection: only when you tap, the app uses your saved key to send the provider a request that contains no chat content.
• Online media: when you open a web address in the player, your device connects directly to that address.
• Model downloads (optional): only after you enter a repository in Settings → Local models → Custom local models and tap, the app connects to Hugging Face to look up and download files. The requests contain only repository and file information, never your conversations. Like any website, Hugging Face sees connection details such as your network address, and its own privacy policy applies. Downloaded or imported models are stored in the app's private folder and are deleted when you delete all data.
• When you share or export: data goes to the place you choose (including AI voice files you save).
• Apart from these cases and where the law requires, we do not provide your information to anyone.

Retention and deletion

• Data stays on your device until you delete it.
• Settings → Delete all data removes all content and the key to the encrypted database. This cannot be undone.
• Uninstalling the app also removes all of its data.
• You can view, edit, delete, or pause memories at any time in My space → What AI remembers.
• AI voice audio is kept only in memory, and the most recent clip stays there for replay or saving; it is cleared when you turn off Voice replies or leave the chat.

Security

The on-device database is fully encrypted, and its key is kept in the system's secure storage. Please use a screen lock on your device and keep exported files safe.

Your rights

In the app you can export all data, delete all data, withdraw each consent, change or delete your emergency contact, and manage memories. Depending on where you live (for example under the EU GDPR, Canadian privacy laws, or the California CCPA), you may have additional rights. Because your data is on your own device, you can exercise most of these rights directly in the app; for questions, contact us as described below.

Information about emotions or mental state

Many laws treat this kind of information as sensitive. Before you start, the app asks for your separate, explicit consent. You can withdraw it in Settings → Privacy and consent; after that, chat is no longer available but the Help page still works.

People under 18

The app is only for people 18 and older, and we do not knowingly handle information about minors.

Changes and contact

• If this policy changes in an important way, the app will show the new version and ask you to confirm again.
• Contact: hello@limburgite.com (developer responsible for privacy: Limburgite Inc.).