Who we are
Multi Inference is an iOS app made in San Francisco. It gives you a Multi, an AI friend who gets to know you through your photos, and turns moments from them into a Daily and a weekly video. Because the app works with something as personal as your camera roll, we want to be plain about what we collect and why. If anything here is unclear, email us at hello@multi-inference.com.
What we collect
- Account info. The app creates an anonymous account when you first use it. If you choose Sign in with Apple, we also receive an Apple account identifier and your email if you choose to share it.
- Names. Your first name and the name you give your Multi.
- Photos you choose to share. These are uploaded securely to our servers so the app can create your Daily, your weekly video, your Multi's letters and posts, and your Multi's understanding of you.
- Messages. Your chats with your Multi and messages you send to other people and their Multis.
- Calls with a Multi. With microphone permission, your voice is sent to OpenAI to produce the Multi's spoken replies. We save a text transcript, call summaries and call details such as participants, time, duration and status. These let you revisit calls and help your Multi remember the conversation. We do not save an audio recording on our servers, and we disable OpenAI's optional session-recording storage.
- Calls with other people. LiveKit carries the live audio between participants. We save call details such as participants, time, duration and status, but do not record or transcribe these person-to-person calls. We also store a device push token to deliver incoming calls.
- Purchase records. If you subscribe, Apple handles the payment and tells us what you bought. We never see your card details.
- Diagnostics. Basic diagnostic information, if you send us a bug report.
- On this website. Your email if you request access, plus cookieless, privacy-respecting analytics — no cookies, no cross-site tracking, no ad identifiers.
How we use it
We use your photos and chats to create your Daily and weekly content and help your Multi understand you. Provider processing is described below. Call data lets us connect calls, show call history and let your Multi follow up on conversations or tasks you ask it to do. There are no ads, and we never sell your personal information.
Photo screening on your device
Before camera-roll photos are uploaded, a safety filter checks them on your phone for sensitive content, including nudity, financial documents, identity documents and wallet recovery phrases. Photos flagged by the filter are skipped. Automated screening may miss some sensitive images.
Faces, selfies and likeness references
Photos you provide may show your face and other people's faces. You may also take an optional onboarding selfie. We store resized copies of these images, selected reference images and crops, and AI-generated appearance descriptions and reference selections, such as which photos show you clearly. We use them to understand your photos, choose and check your profile image, and create personalized pictures and videos that resemble you. When you change your profile image, we may compare it with your selfie to check that it shows the same person. This comparison is not used to sign you in or authenticate your account.
Multi does not collect Face ID data, TrueDepth depth maps, ARKit face meshes or facial landmark templates, or create its own biometric identification database. AI providers process face images for the purposes described in this policy. We do not sell face data or use it for advertising. Only provide photos you have the right to use, including permission from other people where required.
Your selfie can become your profile picture and be visible to other people in Multi. Personalized content that you share can also show your likeness. Your uploaded camera-roll library is not made publicly browseable just because you connect it. You can skip the selfie, change your profile picture, and request deletion using the options below.
AI processing and providers
Anthropic and OpenAI process photos, including selfies and face references, messages and relevant context to provide Multi's conversations, photo understanding, likeness selection, image checks and generated images. Their commercial API terms do not use customer content for model training by default. OpenAI processes audio and relevant conversation context for calls with a Multi; Anthropic and OpenAI may process call transcripts and context to support memory, tasks and follow-up reports. LiveKit provides the audio connection for calls between people.
For new welcome and weekly video renders, Higgsfield receives selected photo or likeness references, generated images and instructions, and may send these to the model services it uses, including ByteDance's Seedance. Previously started renders may finish on Hedra; references previously sent to Hedra remain subject to its retention and data-use terms. Hedra also processes narration and music text for audio generation using ElevenLabs models.
Higgsfield's API training opt-out has been enabled. Under its terms, it can take up to 10 business days to take effect, applies to Higgsfield-controlled processing, and does not automatically cover upstream providers or excluded models. Before it takes effect, Higgsfield's standard terms permit training on submitted content. Hedra's published terms also permit training on submitted content. These terms differ from Anthropic and OpenAI's no-training defaults described above.
More information: Hedra API terms, Hedra privacy policy, Higgsfield terms, Higgsfield API terms, and Higgsfield privacy policy.
Where we store face data
Our backend runs on Render. Uploaded photo and selfie derivatives, profile pictures, likeness-reference crops and generated media are stored in Cloudflare R2. Account-linked photo records, appearance descriptions, reference selections and processing results are stored in Supabase Postgres. The app also keeps local caches; a captured selfie is staged in the app's protected local storage before upload. Data sent for AI processing is also handled on the providers' infrastructure under their respective retention policies. These services may process data outside your country of residence.
Sharing
We share data only with the service providers above, and with authorities if the law requires it. Pieces you choose to share with friends are visible to the people you share them with. That part is up to you.
Messages are shared with their recipients, and live call audio is heard by the participants. If you send your Multi to call someone, the person it calls and you can see the call transcript, and your Multi reports back to you. If another person's Multi calls you, that person can likewise see the transcript and receive a report. Information you ask your Multi to pass on is shared with the intended recipient.
Retention and deletion
We retain your uploaded photos and selfies, likeness references, appearance descriptions and generated content while your account exists so that Multi can continue to understand your context and create content for you. This retention also applies to inactive accounts. To delete your account and withdraw consent to future processing, open Settings and choose Delete account. This removes account-linked database records and queues deletion of the account's media from our Cloudflare storage. Media deletion runs asynchronously. You can also email hello@multi-inference.com to request account deletion or deletion of a particular photo or selfie.
Anthropic's standard API retention is up to 30 days, with exceptions such as safety enforcement or legal obligations. OpenAI may retain API content in abuse-monitoring logs for up to 30 days by default; some features and safety or legal exceptions have different retention. Disabling optional audio recording does not disable provider-side safety retention.
Higgsfield retains API request parameters and output metadata while our provider account is active. Its API terms specify at least seven days of availability for generated media without a fixed deletion deadline. Hedra retains personal information for its stated purposes and legal obligations; deleted information may remain in its database backups for up to 30 days. Deleting a Multi account does not automatically delete files held in our accounts with these providers. Contact us to request removal of provider-held copies; we must coordinate those requests with the applicable provider. Higgsfield states that content deletion stops future training use; it does not reverse training completed before deletion.
Copies may remain in backups or caches until those are cleared, or where retention is required by law. Limited operational records and bug reports may be retained separately; do not include private photos in a bug report unless needed. Deleting a source photo or replacing a selfie does not automatically erase earlier generated content or information already incorporated into your Multi's context. Account deletion removes that account's stored context. Copies other people have received or saved remain under their control.
Security
Your data is encrypted in transit, and access to it is limited to what's needed to run the service.
Eligibility
Multi Inference is for adults 18 and older in the United States and Canada. The app is not directed to children, and we don't knowingly collect data from anyone under 18.
Your California rights
If you live in California, you have the right to access, delete, and correct your personal information. We do not sell or share personal information as defined by the CCPA/CPRA. To exercise your rights, email hello@multi-inference.com.
Changes
If this policy changes, we'll update this page and change the date at the top.
Contact
hello@multi-inference.com — a human reads it.