Sub-processors
Who processes data on Alfaaz's behalf.
Alfaaz (Pulkit Mendiratta, eenmanszaak, Amsterdam, NL) uses the third-party providers below to operate the service. Each receives only the data it needs for its function. We do not sell or rent personal data, and we do not permit these providers to use your Content to train AI models. Not every provider processes every message — some are used only for specific features.
01
AI providers
| Provider | What it does | Data it receives | Location | Retention / training |
|---|---|---|---|---|
| Anthropic (Claude) | Generates conversation questions; extracts people/places/events and story summaries; writes life chapters and the elder profile | Transcripts, elder & family names, topic, previously extracted entities | USA | API inputs retained briefly for safety/abuse monitoring then deleted (per Anthropic’s API policy); not used to train models. |
| ElevenLabs | Speech-to-text (transcribing the elder’s voice) and text-to-speech (generating Alfaaz’s voice) | Elder voice-note audio; response text | USA | Processed to produce text/speech. Retention follows ElevenLabs’ policy and our tier; we are working to enable zero-retention. Until then, treat audio as retained on ElevenLabs’ default tier. |
| Google (Gemini API) | Scores conversations internally (e.g., emotional depth and shareability) to surface the right stories to the family | Transcript text + story summary for the scored conversation | USA / global | Processed for scoring; not used to train models (per the API terms we rely on). |
| OpenAI | Part of our model-routing layer; may process text for specific tasks/fallbacks | Transcript-derived text for the specific task | USA | API data not used to train models (per OpenAI API policy). Used only where routed. |
| Sarvam AI | Part of our model-routing layer for Indic-language text tasks | Transcript-derived text for the specific task | India | Processed for the task. |
02
Messaging, infrastructure, and operations
| Provider | What it does | Data it receives | Location | Notes |
|---|---|---|---|---|
| WasenderAPI | Delivers and receives messages over WhatsApp (third-party messaging provider) | Elder phone number; voice-note audio; message metadata | Provider infrastructure (varies) | Transient — messages pass through for delivery. WhatsApp itself is operated by Meta under its own policy. |
| Vercel | Hosts the web application; stores audio files (Vercel Blob) | Application traffic; voice-note and generated audio files | USA + global edge | Audio retained while the elder is active; deleted on elder deletion. |
| MongoDB Atlas | Primary database | All account, conversation, extracted-entity, and profile data | Cloud region (Mumbai option under evaluation) | Encrypted at rest and in transit. |
| Clerk | Authentication and session management | Email, name, login sessions | USA | Manages credentials; we do not store passwords. |
| Resend | Transactional email (session digests, invitations, reminders) | Recipient email, elder name, session-summary highlights, links | USA | Used only for service emails, not marketing. |
| Stripe | Payment processing | Card details, billing info (entered with Stripe) | USA / global | We never store card numbers. One-time, manual charges only. |
| PostHog | Product analytics | Usage events; identifiers hashed before leaving our servers where possible | EU (eu.i.posthog.com) | No advertising trackers; not used for behavioural advertising. |
03
Notes
- This list reflects providers that may process personal data. We keep Data Processing Agreements with providers acting as our processors and rely on SCCs / EU-US Data Privacy Framework participation for transfers out of the EEA where applicable.
- Questions or objections about a sub-processor: privacy@alfaaz.me.
Trust & care