Privacy Policy.

1. Who is responsible

SCHIZA is developed and operated by Mikhail Firulik in Israel, who is responsible for the personal information described here. Contact: support@schiza.ai.

This policy covers schiza.ai and the privately tested iOS app and its supporting services. The app is not publicly released. Features vary between test versions; a planned feature is not a promise that it is available. Our Terms of Use explain participation in testing.

2. Visiting the website or contacting us

You can browse without an account. The site has no signup forms, advertising, analytics scripts, tracking pixels, cookies set by its code, or browser storage. Its illustrative conversations run only in your browser. The website does not request microphone access or collect conversation audio.

Cloudflare delivers and protects the site and may process IP addresses, requested URLs, request times, device or browser information, and security logs. If you email us, we process your address, message, and any attachments to answer your request. Email routing and the destination email service also process that correspondence.

3. Information processed by the app

Providing personal information is voluntary; there is no legal requirement to tell SCHIZA about yourself. Account identifiers are needed for sign-in, and audio and conversation content are needed for a voice conversation. If you do not provide them, the corresponding feature cannot work. Personal questions can be skipped.

• Account and access: an identifier supplied by Apple or Google sign-in, account and device-session identifiers, authentication information, and preferences. SCHIZA does not receive your Apple or Google password.
• Conversations: your speech, its transcripts, AI replies, dates, language, and information you choose to share, including during the introduction. Conversation history is saved locally and synchronized to your account on the server.
• Memory and personalization: extracted facts, preferences, ideas, opinions, episodes, and boundaries, with source references and changes you make. Search representations help retrieve relevant past material. These records support future replies; they are distinct from the original conversation archive.
• Files and tools: attachments you choose to upload, generated files you request, and tool inputs needed for the task, including search queries when web search is used.
• Diagnostics: technical events such as connection failures, timing, interruptions, and playback state, used to maintain and troubleshoot the service. Diagnostic logging is designed to avoid message bodies, raw audio, and credentials.
• Support and deletion: correspondence you send us and an email address you optionally provide for a deletion-completion notice.

We use this information to provide and resume conversations, personalize replies, synchronize your account, operate requested tools, protect access, investigate errors, and handle your requests. What you share may include sensitive information about you or other people. Share only what you want processed and have a right to provide.

4. Getting to know you and remembering

The experience being developed starts with a question shown by the assistant and a spoken answer. It is designed to save information from the first spoken response, with factual memory enabled by default for new users. Some earlier test versions use a different default; check “Remember facts about me” in your app. An explicit choice to turn it off remains your choice.

Research questions help SCHIZA explore your values, decisions, habits, and reactions. Ordinary conversation also contributes information. Research preferences and factual memory are separate controls. The complete research planner and profile are still being developed.

AI extraction and interpretations can be wrong. An inference about your personality is not a verified fact or clinical diagnosis. Memory controls let you inspect source words and review, correct, exclude, or forget saved entries where available in your version. If a control is missing or a change has not completed, contact us.

Private conversations are not automatically published or used by SCHIZA to train a general model for other users. Any future product-wide learning from private content requires a separate, explicit choice. This is distinct from using your information to personalize your own conversations.

5. Voice, samples, and experimental features

While the microphone is active, audio is processed for speech recognition and conversation. Depending on the selected mode, processing may use Apple speech services or a cloud AI service. Locking your phone may not stop an active conversation: continuation while locked is enabled by default in newer versions. Mute or end the session to stop input, or change this preference in the app.

Personal-voice testing can capture and upload a limited sample of your own speech and store it with your account to prepare and check synthesis. We therefore do not promise that microphone audio is never retained. Personal voice remains experimental; successful sample collection does not guarantee a working personal voice.

The planned expanded audio bank would temporarily link speech recordings to transcripts for analysis, retain selected clips, and support personal voice improvement. Broader acoustic emotion analysis and personal fine-tuning are not described here as operational features. Before expanded collection is enabled, its purpose, retention, and controls must be explained in the app; this policy is not blanket consent to future processing.

Use only your own voice in private testing. Permission to use a microphone is not permission to impersonate someone, share a voice, or train a general model. A spoken provider declaration is not required in every synthesis mode; its absence does not give permission to use someone else’s voice.

6. Service providers and international processing

Providers receive the information needed for the relevant service, not necessarily your entire archive. The route depends on your test version and selected feature:

• Microsoft Azure: the hosted app backend, account database, file storage, and Azure AI services used for conversation, speech, memory extraction, and search representations.
• OpenAI: direct API processing in test configurations that use it, including conversation, speech, and supported file or search tools. Azure-hosted and direct OpenAI routes are different configurations.
• Apple and Google: sign-in; Apple also provides speech recognition in applicable modes. Apple speech processing may use its servers depending on device and language support.
• Runpod: GPU hosting for experimental personal-voice processing when that route is enabled. Synthesis may process a voice sample and the text to be spoken. Qwen and VoxCPM2 are model software used in this route, not a claim that every conversation is sent to their publishers.
• Cloudflare: website delivery, security, email routing, and service email delivery. Contact emails are forwarded to the developer’s Apple iCloud email service.

Processing may take place outside your country, including in the United States and outside Israel. This is a cloud-connected service, not an on-device-only service. Provider processing and retention depend on the service and configuration; deleting data in SCHIZA does not promise immediate deletion from all provider systems.

Provider information: Microsoft, Azure AI data handling, OpenAI API data controls, Apple, Google, Runpod, and Cloudflare.

We do not sell your personal information or use it for advertising. Shared Twin is not available. Future sharing would need separate permission for recipients, purposes, topics, memories, and voice; permission to share a style would not open the whole archive.

7. Storage, retention, and deletion

Account history and memory: retained to support your account and continuing conversations until you remove the relevant data or request account deletion. Turning factual memory off stops that layer’s extraction and use; it does not erase the archive. Uninstalling the app does not delete server-side account data.

Voice samples: a stored personal-voice reference can remain associated with your account for synthesis. Contact us to remove a sample if your version does not expose a separate control. Account deletion includes stored voice data. No fixed automatic expiry is promised for these samples.

Files: the current file-exchange design expires access 24 hours after upload or creation and queues physical cleanup. Outages can delay cleanup. Copies you save or send elsewhere remain outside this process. Discussion of a file can remain in conversation history until that history is deleted.

Deletion: server cleanup can run in the background and fail or require retry. Keep the deletion receipt or confirmation; a spoken AI acknowledgment is not evidence of completed deletion. Backup copies may remain until their retention period expires. The documented database configuration retains backups for seven days; this does not establish a universal seven-day limit for every provider or data category.

Other records: technical and security records, deletion receipts, and support correspondence may remain as needed to operate the service, handle a request, prevent abuse, or meet legal obligations. Retention across all provider logs and support correspondence has not yet been consolidated into a fixed schedule for public release. Contact us for details relevant to your test version.

Deletion does not remotely erase exports, recipients’ copies, or every local copy on another device. We do not promise instantaneous physical erasure across all systems.

8. Your choices and requests

• Choose what you say; skip questions, postpone topics, and manage research preferences where available.
• Mute the microphone, end a conversation, change background-conversation settings, or revoke microphone and speech permissions in iOS.
• Review and manage factual memory in “Memory and getting to know you.” A conversational correction alone may not change a stored entry; confirm it in the memory controls.
• Manage conversation history separately, or request account deletion through Settings → Account → Delete account.
• Contact support@schiza.ai to request access, correction, deletion, or restrictions on your information, or to report a missing or failed control.

We may need to verify that a request relates to your account. We assess requests under applicable law, including applicable rights to inspect and request correction of personal information. Do not email passwords, access tokens, or unnecessary identity documents. If a test build prevents you from exercising a choice, contact us through the address above.

9. Security, age, and changes

The service uses HTTPS, account access checks, protected local storage, and restricted server storage. These measures do not guarantee absolute security. SCHIZA is intended for adults aged 18 and over and is not intended for children.

We update this policy when processing changes and revise its date. Material new processing must be explained before it starts, with a separate choice where required. Publishing a policy or announcing a feature does not itself enable it or record your consent.

10. Contact

Mikhail Firulik, SCHIZA, Israel.
Email: support@schiza.ai.

For questions about using the prototype, see the Terms of Use.