⚠️ Founder-reviewed interim version — posted 2026-10-29. This page has been reviewed by BNC Solutions LLC's founder for accuracy about how Mimi Care actually operates. An independent attorney review is in progress. We will update this page and note the date when that review completes.
Mimi Care Subprocessors
Last updated: 2026-10-29
Mimi Care uses the vendors below ("subprocessors") to operate the service. Each one receives only the data it needs to do its job. We do not sell data to any of them.
A mirror of this page lives at mimicare.ai/subprocessors.
Current subprocessors
| Subprocessor | Purpose | Data shared | Location | Privacy policy |
|---|---|---|---|---|
| Anthropic (Claude API) | The AI model — reads photos and writes explanations, answers questions, writes Mimi Care's side of Companion conversations and their short notes, summarizes conversation recordings, checks email sent to the Primary User's Mimi Care address for scams and finds reminders in it, and may run public web searches for some questions | Photo contents; questions and conversation text; Primary User's first name; optional full legal name; household-member names and relationships; on every turn of a conversation with Mimi Care, the day's medicines, reminders, and calendar (and, if remembering is on, the saved notes and memories); when the Primary User asks about them, saved helps and recording transcripts; help-contact names (not phone numbers) when the Primary User asks Mimi Care to call someone; recording transcripts for summaries; the text of emails sent to the Primary User's Mimi Care address; for a weather answer, the name of the town or area the forecast is for | United States | anthropic.com/legal/privacy |
| Amazon Web Services (Bedrock) | Alternative way Mimi Care can run Claude (a configuration option; Anthropic's API is the current path) | Same as Anthropic, when in use | United States (us-east-1) | aws.amazon.com/privacy |
| Cloudflare (Workers, D1, KV, R2, AI Gateway, Workers AI, Analytics Engine, Email Routing, CSAM Scanning Tool) | Core infrastructure — API, database, object storage, caching, AI request routing and logging, small automatic safety classifiers (Workers AI), request telemetry, inbound and transactional email, network-edge CSAM scanning | All data in transit, including the approximate location Cloudflare estimates from each request's network (IP) address (Privacy Policy §2). At rest in D1 and R2: account records and scan metadata; our own measurement of ads and links to the website, with no network address, browser description, or ad click ID (Privacy Policy §2); stored encrypted — helps (photos, previews, explanations, conversations, spoken-explanation audio), conversation recordings, Companion notes and memories, daily activity, medicine details with family voice clips and pill photos, reminder and calendar text, and the subject and body of email delivered to the Primary User's Mimi Care address; stored without that extra encryption (Privacy Policy §3) — fall events, dose records, wellbeing-signal notes, help contacts, scan flags, email sender addresses and display names, and quarantined email; voice-cue audio assets. AI Gateway: a log of every request Mimi Care sends to Claude and its answer — including photos and conversation text — kept in Mimi Care's own Cloudflare account (Privacy Policy §5). Analytics Engine: records of requests blocked from outside the US (country, the requester's network but not the full IP address, user agent, requested page), and app-start and speed measurements labeled with a one-way coded (hashed) seat reference. Rate-limit counters in KV. Transactional email send (magic-link sign-in, approval notifications, legal-consent receipts, dunning, billing notices — the steps to cancel or restore a subscription, and notice that renewal stopped or was turned back on, after a reported death or an account deletion — trusted-sender verification — recipient address, subject, body; no photos or scan content). | United States | cloudflare.com/privacypolicy |
| Amazon Web Services (S3 — disaster-recovery backups) | Off-Cloudflare backup storage, used only for disaster recovery | A nightly copy of our database (account records, scan metadata, and the other records stored in the database — encrypted records stay encrypted in the copy, and records stored without that extra encryption are copied as they are); weekly copies of business customers' content files and search index | United States (us-east-1) | aws.amazon.com/privacy |
| Deepgram (speech-to-text + Aura-2 text-to-speech) | Turns the Primary User's spoken words and conversation recordings into text, and reads responses aloud in a warm voice | Primary User's voice audio for transcription; conversation-recording audio; text of AI response being spoken | United States | deepgram.com/privacy |
| OpenAI (Whisper) | Backup speech-to-text, used when the primary transcription is unclear | Primary User's voice audio, or conversation-recording audio | United States | openai.com/policies/privacy-policy |
| Twilio | Text-message alerts to Mimi Care staff about safety incidents | A short description of the alert, internal reference numbers, and a link to our staff console, sent to our staff's phones — not what the Primary User said | United States | twilio.com/legal/privacy |
| Stripe | Web subscription billing, customer portal, hosted checkout | Account Holder's email, billing address, payment method last-4 (we never see card numbers), subscription records | United States | stripe.com/privacy |
| Apple App Store / StoreKit | iOS subscription billing via Apple In-App Purchase | Apple receipt identifiers, subscription status, auto-renewal state | United States (per Apple ASC terms) | apple.com/legal/privacy |
| Google Play Billing | Android subscription billing | Google receipt identifiers, subscription status, auto-renewal state | United States (per Google Play terms) | policies.google.com/privacy |
| Apple (APNs) | iOS push notification delivery — reminders and medicine-dose reminders to the Primary User; safety, SOS, scam, and missed-reminder alerts, approval requests, Inbox items, and app-update notes to family | Push token, short notification title + body, and reference IDs (for example, a help or reminder ID) | United States | apple.com/legal/privacy |
| Google (FCM) | Android push notification delivery — same notifications as Apple APNs | Push token, short notification title + body, and reference IDs (for example, a help or reminder ID) | United States | policies.google.com/privacy |
| Google Analytics | Measures visits to the public pages of mimicare.ai: which pages are read, and which ads and links bring people to Mimi Care. Google signals and ad personalization are turned off. Not used on signed-in pages or in the app. | For each visit: the page and the page the visitor came from (with any ad click ID replaced by a placeholder), browser, operating system and device type, actions such as joining the waitlist, and a random identifier kept in a browser cookie. Google estimates an approximate area from the visitor's network address. | United States and other countries where Google operates | policies.google.com/privacy |
| Didit (age verification) | The one-time, optional age check Mimi Care offers only where a law requires one — today only New York, starting January 1, 2027 (Privacy Policy §2), after the Primary User agrees on the check's screen: a face age estimate with a liveness check or, if the Primary User chooses, a photo ID check, in which Didit reads the date of birth on the ID, checks liveness, and compares the face with the photo on the ID only to confirm the ID is theirs. The pictures are used only to check age | Sent from the Primary User's phone straight to Didit, never through our servers or any AI model: a picture of the Primary User's face and, for the photo ID check, a photo of their ID. From us: a seat reference number, the place (for example, US-NY), the language, and the record of their agreement (when, and which version of the words). Didit returns only the outcome and what it needs to decide it (the age estimate, or the date of birth on the ID, which we use only to decide and never keep); we keep only the result (adult, under 18, or could not tell; how; the lowest age proved, 18 or 25). Face searching is off: the face is never used to identify or search for anyone, and no audio is recorded. | European Union (Ireland) | didit.me/terms/privacy-policy |
| Sentry | Runtime error reporting across the backend Worker, mobile app, and web app | Stack traces, error messages, app + OS version, and internal reference IDs (such as an account or seat ID) — not names, email addresses, or the content of helps | United States | sentry.io/privacy |
Public data sources
These services are not subprocessors: they do not process personal information on our behalf. Our servers ask them for public information, and the request carries nothing about who asked.
| Service | What we ask for | What the request includes | Location | Terms |
|---|---|---|---|---|
| MET Norway (Norwegian Meteorological Institute, api.met.no) | Weather forecasts, when the Primary User asks Mimi Care about the weather | Coordinates rounded to two decimal places (an area about 1 km across) — either the approximate area of the Primary User's internet connection or a U.S. place they named (Privacy Policy §2). Sent from our servers, so MET Norway sees our network address, not the Primary User's. No name, account, or device information. | Norway | api.met.no/doc/TermsOfService |
What is not shared with any subprocessor
- We do not share contacts from the phone's address book, precise (GPS) location, browsing history on other sites, or advertising identifiers with anyone — we do not collect them in the first place (when someone arrives from one of our ads, we keep only which ad platform the link came from, never its click ID). The only location we use is the approximate area of an internet connection (Privacy Policy §2): for a weather answer, rounded coordinates go to MET Norway and the town or area name goes to Anthropic so Mimi Care can say it.
- Helps, recordings, and the other records listed as encrypted in the Cloudflare row are stored in Mimi Care's database and file storage in encrypted form. The records listed there as stored without that extra encryption, and the AI Gateway log, which holds readable copies of requests to the AI and their answers (Privacy Policy §5), are not.
- No subprocessor receives the full set of data held by any other subprocessor. Each receives only what it needs to do its specific job.
Data retention at subprocessors
- Anthropic — Anthropic's Commercial Terms prohibit training models on Mimi Care content. Under Anthropic's published data-retention policy, API inputs and outputs are deleted within 30 days, and kept up to 2 years only if flagged by its trust-and-safety systems. Parts of a request may be held in Anthropic's prompt cache for 5 minutes after last use to serve follow-up questions.
- Amazon Web Services (Bedrock), when in use — per the AWS Service Terms (§75 Bedrock), Amazon does not use inputs or outputs to train any model.
- Deepgram — every request we send opts out of Deepgram's Model Improvement Program, so Deepgram does not use it to improve its models.
- OpenAI — under OpenAI's API data-usage policy, OpenAI does not use API content to train its models by default.
- Cloudflare — data in D1 and R2 follows the retention periods in Privacy Policy §8 (for example, helps 180 days unless kept, scan metadata without a stored help 90 days); audit logs retained up to 7 years; voice-cue audio in R2 retained indefinitely (static brand assets). AI Gateway logs have no fixed time limit: the gateway keeps up to a set number of requests and deletes the oldest only when that limit is reached, so an entry can remain for many months. Records of blocked requests and app-start and speed measurements in Analytics Engine expire after about 3 months. Our own measurement of ads and links, in D1, is deleted after 13 months. Weather forecasts are cached for a few hours at most, labeled only by rounded coordinates. Email Routing send logs retained per Cloudflare's standard operational window; transactional message bodies not retained at rest beyond delivery.
- AWS S3 (backups) — backup copies are kept for a limited period under our disaster-recovery plan, then deleted.
- Stripe — billing records retained per Stripe's standard lifecycle (7 years typical for tax / audit).
- Sentry — error events retained per our project-level retention setting (90 days default).
- Google Analytics — Google keeps the detailed records for up to 14 months; reports built from them (counts and totals, not single visits) can be kept longer. Its cookie lasts up to 2 years after a visitor's last visit.
- Apple / Google — push tokens and payloads retained per each provider's standard push lifecycle.
- Didit — we delete each age-check session, with its pictures, through Didit's API as soon as its result reaches us, and within about a day if it is never finished. Didit's own retention is set to its minimum (1 month) as a backstop.
Changes to subprocessors
If we add or materially change a subprocessor in a way that affects what data is shared, we will:
- Post the update to mimicare.ai/subprocessors and
docs/legal/subprocessors.md. - Announce the change in the app at least 30 days before the new subprocessor starts processing your data. When the change comes with a Privacy Policy update, the in-app request to accept that update is the announcement.
We may make non-material changes (a vendor reorganizes, a region expands) without prior notice, but we'll update the table here.
Questions
BNC Solutions LLC 418 Broadway, Ste. N Albany, NY 12207
A Spanish-language informational translation of this list is available at mimicare.ai/subprocessors?lang=es. The English version is the authoritative document.