Privacy Policy

1. Overview & Contact

2. Information We Collect

• Directly Submitted Reservation Information: Full name, email address, optional workflow/OS notes, and preferred interface language.
• Reservation and Waitlist Records: Reservation or waitlist status, submission timestamp, and timestamp/version of Terms of Service acceptance. A reservation token and queue position are created only for a confirmed Batch 01 slot.
• Rate-Limiting & Security Telemetry: When a reservation is submitted, our backend calculates a one-way keyed cryptographic hash (HMAC-SHA256) of the client IP address to prevent automated abuse. Raw IP addresses are not stored in our reservation database table. Standard temporary server infrastructure logs may be recorded for security and operational diagnostics.
• Local Browser Storage: We store only your preferred language (lyratype_lang) and color theme mode (lyratype_theme) in your browser's localStorage. We do not use tracking cookies. If you use the conversation page, it also stores in your browser your access code, your confirmation that you are 18 or older, your microphone selection, and whether you have viewed its introductory tip or completed its survey, so that you need not provide them again; the access code can be cleared from the page menu.

3. LyraType Moon Research Preview (AI Character Conversations)

• Access requests: When you request access on the Moon page, we collect your name, email address, preferred reply language and any optional note you provide. This information is used solely to review and respond to the request and is deleted six (6) months after submission.
• Voice input: During a conversation, audio from your microphone is transmitted to our speech recognition provider for real-time transcription and is analysed on our servers to detect the end of speech. Microphone audio is not recorded or stored.
• Camera input: The camera is off by default and is used only if you enable it during a conversation. While it is enabled, a still image is transmitted to our AI provider at most once every few seconds, and only when the scene changes, to generate a short text description for the AI character. The text descriptions are retained with the conversation record; the images are not retained.
• Conversation records: We retain the text of each conversation (your spoken or typed input and the AI character's responses), its date, time and duration, and technical measurements such as response latency. We also retain the synthesised audio and avatar motion data generated for each conversation, to evaluate audio and motion synchronisation. The AI character's voice is synthetic and is never derived from your voice.
• Conversation memory: To provide continuity between conversations, the service stores brief summaries of previous conversations under your access code. Any person who uses the same access code can access conversations that draw on these summaries. Do not share your access code, and do not disclose information that you do not wish to be retained.
• Feedback: Responses to post-conversation questions are retained with the corresponding conversation record.
• Purposes: We use this information to operate the service, to improve the AI character and our models (including manual review of conversations), to maintain safety and usage limits, and to respond to you. Conversation data is not used for advertising and is not sold.
• Service providers: Modal (hosting of the conversation service and storage of conversation records, United States); DeepSeek (the language model that generates the AI character's responses; conversation text is transmitted to it, and it may process and store data in the People's Republic of China); OpenAI (a backup language model used only when the primary model is unavailable, and automated content-safety screening of conversation text, including user input and AI responses); ElevenLabs (speech synthesis); Soniox (speech recognition, and backup speech synthesis when ElevenLabs is unavailable); Supabase (access requests). Each provider processes data on our behalf under its own terms.
• Safety screening: Each conversation turn is screened automatically against our safety rules by OpenAI's content-safety service and by DeepSeek. If self-harm is mentioned, the page displays crisis resources and the conversation continues. The service is intended for adults; a conversation with a user who appears to be under eighteen (18) years of age is ended. Only aggregate counts of screening results are retained, never their content.
• Crisis protocol: Every user input and every AI response is screened for indications of suicidal ideation or self-harm, in real time by rule-based screening and OpenAI's content-safety service, and by DeepSeek against our written policy. When an indication is detected, the page displays a help card listing the 988 Suicide & Crisis Lifeline (call or text 988 in the United States) and findahelpline.com for other countries, and the AI character is instructed to respond with care: to treat the matter seriously, to encourage the user to contact a trusted person or a crisis line, and never to provide any means or method. The conversation is not terminated. An AI response that could encourage suicide or self-harm is withheld before playback. We record the number of times the help card is displayed, without any conversation content, and publish that number annually.
• Retention: Conversation records, synthesised audio, motion data, feedback and conversation memory are deleted ninety (90) days after the conversation. Aggregate usage data that contains no conversation content may be retained longer for cost management.
• Deletion: The "Your data" section of the conversation page menu permits immediate deletion of all conversations and conversation memory associated with your access code. Deletion requests may also be submitted by email.
• LyraType Studio: For studios that use lyratype.com/studio with a studio token, conversation records are retained on our service for ninety (90) days. Conversation memory and the studio's character settings are retained for that studio token until the token is deleted; a studio may request earlier deletion by email. The VTube Studio authorisation and the studio token are stored only in the studio's browser.
• Stream chat: When a studio specifies a Twitch channel, the service reads that channel's public chat so that the studio's AI character can respond to it; the service never posts to the chat. Chat messages read by the service are retained for no more than twenty-four (24) hours, to operate and screen the conversation, and are then deleted.
• Age requirement: Conversations are available only to adults aged eighteen (18) or older. Users must confirm that they are 18 or older before a conversation starts, and access codes are not knowingly issued to anyone under 18.

4. How We Use Information

• Administering provisional Batch 01 reservation slots and the waitlist.
• Sending essential transactional and operational communications, including reservation confirmations, queue status updates, allocation notifications, pre-production deposit/payment requests, order fulfillment, shipping updates, safety/recall notices, and customer support responses.
• Protecting our website against automated bots, duplicate submissions, and security threats.
• We do not send promotional marketing newsletters or sell your contact information.

5. Third-Party Service Providers

• Supabase: Database hosting, data persistence, and serverless API execution, including access requests for the LyraType Moon research preview.
• Vercel: Frontend hosting, content delivery network (CDN), and edge routing.
• Google Fonts: Web font stylesheet delivery and typography assets.
• Modal: Hosting of the talk service and storage of talk records (United States).
• DeepSeek: The language model that generates the AI character's responses; it may process and store data in the People's Republic of China. It also screens conversation text against our safety rules.
• OpenAI: A backup language model, used only when the first is unavailable, and an automatic content-safety check of the talk's text.
• ElevenLabs: Speech synthesis.
• Soniox: Speech recognition, and backup speech synthesis when ElevenLabs is unavailable.

6. Data Retention & Expiration

7. Your Rights & Reservation Cancellation

8. Security

9. California & State Privacy Disclosures

10. Do Not Track & Global Privacy Control

11. Children's Privacy

12. Policy Changes