Privacy and trust at Familiar, in plain English

Familiar is built around one promise: nothing about your family leaves the circle of family. No ads, no selling of data, no training other people's voice models on your loved one's voice. The work below is how we keep that promise concrete instead of marketing language.

• Voices, photos, transcripts, journal entries: AES-256 encryption at rest. The voice model files (one `.mv2` per receiver) live in Google Cloud Storage; voice clones and full transcripts live in Supabase Postgres, also encrypted at rest.
• Media we text to your loved one (the photos that arrive with the daily call summary): staged on Cloudflare R2 with signed URLs that expire after delivery so they cannot be re-fetched from a copied link.
• Everything in transit: TLS 1.3 between your phone, our servers, Twilio, and the carrier.
• No third-party analytics that read your content. Aggregate site-traffic numbers only. No session replay, no message-content logging.

Only the people you invite into your circle. Up to 10 family members and designated caregivers per receiver. Each circle is per-receiver; if you set up calls for both parents, your sister can have visibility into one parent's circle without seeing the other.

Familiar staff cannot read transcripts or listen to call audio by default. When a support issue requires it (say, the call agent misbehaved and we need to debug the prompt), we ask you in writing, per issue, for narrow access. You sign per incident, not once at sign-up.

We never sell, share, or rent your data. Not to advertisers, marketing partners, insurers, data brokers, or other product teams.

Voice cloning is the most privacy-sensitive thing Familiar does. Our rules are simple and we apply them without exception.

Every Familiar Voice comes from someone who recorded themselves at sign-up. They click 'I consent' on a screen that explains what the voice will be used for (daily calls to a specific named receiver), who can hear it (the circle they've named), and how to delete it.

We do not clone voices from old recordings without per-recording consent. Voicemails, home videos, audiobooks: none of it. If a voice donor has passed away, their voice cannot be created in Familiar.

The Familiar Voice is not a true clone. It speaks in your voice with your stories, but it can also generate sentences you've never said: that's what makes it useful for Daily Calls in Family Voices. We disclose this on the voice-record screen and again before the first call goes out. You're never surprised by what your clone might say.

Settings → Delete my account. Your voice clone, narrations, journal entries, photos, transcripts, and circle memberships are purged within 30 days. The action is immediate and cannot be undone.

Receivers you set up (and the photos *they* uploaded) remain accessible to the other members of their circle, because those memories belong to that family, not just to you. Your own data leaves with you.

Familiar does not diagnose, treat, or replace clinical care. The cognitive trends the dashboard surfaces (vocabulary, repetition, name recall, time-orientation, mood) are signals to bring to a doctor: early signals that a 15-minute annual visit can't produce on its own. They don't substitute for a clinician.

We are not a HIPAA covered entity for V1: we don't bill insurance, we don't operate inside a healthcare provider. But the system is designed with HIPAA in mind: encryption at rest and in transit, least-privilege internal access, written staff-access consent per incident, and the data-deletion controls above.

• Sell or share your data with advertisers, marketing partners, insurers, or data brokers.
• Use your voice or your family's voices to train a model for any other family.
• Read your transcripts for internal staff review without your per-issue written consent.
• Use your data to underwrite, score, price, or refer any other product or service.
• Surprise you with a privacy change. If our practices change, you'll get a real notice.