Mochi Talk Privacy Policy

Last updated: 2026-10-15

Kenya Yasumura ("we", "the operator") provides the mobile app "Mochi Talk" (the "App"). This policy explains how we handle your information.

1. Our core principle — your voice and API keys never reach our servers

• Your conversation audio and content never pass through our servers. They are sent directly from your device to the AI provider you choose (e.g., Google).
• Your API keys are stored only on your device (in the OS secure storage) and are never transmitted to us.
• Conversation history, scenes (prompts), edit history, usage statistics, and emotion records and ratings are stored only on your device. We cannot access or collect them.

2. Information we process on our servers

3. Data that stays on your device only

AI provider API keys (OS secure storage), conversation history and its summaries, the emotions the AI reports for its expressions (its own, and its estimate of yours) with their summaries, your in-call and post-call ratings and reasons (without the reply text; this is used only on your device and is never sent to us or any third party), scenes and their edit history, partner settings (name/avatar/voice/personality), app settings, and usage statistics. You can erase these by deleting them in the app or uninstalling it.

Replies that used live web search (section 4.1) are stored only as a marker, without their text.

4. Third parties

4.1 Sent directly from your device (not via our servers)

• Your chosen AI provider: the following is sent directly from your device to the provider you selected. Before anything is sent for the first time, the in-app "How your data is handled" screen names the provider, lists these items, and asks for your permission (recorded per provider; nothing is sent if you decline).
  • Your microphone audio during calls (in real time)
  • The conversation transcript, and, when you continue an earlier conversation, a summary of it and the most recent exchanges (the summary is also generated by the same provider)
  • Conversation settings: the scene text, the partner personality text, your and your partner's names, and the conversation language
  • Text you enter in scene creation, the personality interview, and "Adjust your partner", together with the scene or personality text being created or edited
  • Searches for scenes with "Search the web for current information" turned on (see below)
  • A fixed sample sentence used to generate voice previews (only if OpenAI or xAI is selected)
  • Your API key (to authenticate with your own account and to verify the key)
  • Google (Gemini): https://policies.google.com/privacy
  • OpenAI (if selected): https://openai.com/privacy
  • xAI (if selected): https://x.ai/legal/privacy-policy
  • Each provider's data handling (including training use) is governed by your agreement and settings with that provider.
• Speech recognition (only if you turn on "Allow network speech recognition" in Settings): when you dictate into a text field while creating a scene or in the personality interview, the audio is sent to Apple's or Google's speech recognition service. By default, recognition runs on your device only and no audio leaves it.
• Live web search (only when "Search the web for current information" is on for a scene): the AI provider you selected performs web searches based on the conversation. The search and the use of its results happen on the provider's side; the operator is not involved.
  • Google (Gemini): Google retains the prompts and outputs used for search for up to 30 days. Replies that include search results are shown together with Google-provided search suggestions.
  • xAI (if selected): governed by xAI's privacy policy.
  • This feature is off by default. Search usage is billed under your own API key agreement with the provider. The text of replies that used search is not saved in this app's conversation history.

4.2 Services we use

• Supabase (US): anonymous accounts and subscription status — https://supabase.com/privacy
• RevenueCat (US): subscription management — https://www.revenuecat.com/privacy
• Apple / Google: in-app purchase processing
• PostHog (US): usage analytics, and receipt of reports (when you report a response, the text of that AI response, your reason, and any note you entered are sent; never audio) — https://posthog.com/privacy
• Sentry (US): crash reporting — https://sentry.io/privacy/
• Cloudflare (US): delivering the availability of AI models (end of service and successor models). The app fetches a file with this information at most once a day, for example at launch. No user data or API keys are sent, but Cloudflare may keep connection records such as your IP address — https://www.cloudflare.com/privacypolicy/

These services host data primarily in the United States. We rely on their contractual safeguards.

5. Purposes

We use the above information only to (1) provide and maintain the App, (2) manage paid features, (3) improve quality and fix bugs, (4) prevent abuse, and (5) respond to inquiries. We never sell your information.

6. Retention and deletion

• "Delete account" in the App removes your anonymous account and associated server data. (Cancel your subscription separately in your App Store / Google Play settings.)
• Long-inactive anonymous accounts are periodically deleted.
• On-device data can be erased in the App or by uninstalling.

7. Security

Communications are encrypted (TLS). Server data is protected by access controls (row-level security). API keys are stored in the OS secure storage (iOS Keychain / Android Keystore).

8. Age

The App is intended for users aged 18 and older.

9. Your rights

You may request disclosure, correction, suspension of use, or deletion of your information under applicable law. Most actions (such as deletion) are available directly in the App; for anything else, contact us below.

10. Changes

We will announce policy changes in the App, with clear notice for material changes.

11. Contact

Operator: Kenya Yasumura — support@mochitalk.app
(Please include the "Support ID" shown in Settings → Account.)