Sub-processor List
Controller: Artem Pecheriukin (NIF Z0774077V), Madrid, Spain. Postal address: see the Legal Notice. Last updated: 2026-07-07 Keeping this list current: we update this list when our sub-processors change; the date above shows the current version. Any AI provider we use is barred from training on your content; retention is zero or limited to a short abuse-monitoring window, as shown per provider below.
Active sub-processors
| Sub-processor | Purpose | Personal data shared | Location / jurisdiction | Transfer mechanism (for EU/UK) |
|---|---|---|---|---|
| OpenRouter | Gateway for speech-to-text and image/document text-extraction only (your dictated audio and uploaded files); routes only to no-retention providers (data_collection: "deny" set on every request). Not used for Google user data: content imported from Google APIs never routes through OpenRouter, and chat/embeddings do not use it |
Dictated/uploaded audio; uploaded images and scanned documents | US gateway, routes to US/EEA inference providers | Zero data retention + no training enforced per request; SCCs (their ToS, EU law) + TIA |
| Novita | Managed chat-model inference for all tiers, including deep reflection modes and nightly memory consolidation (open-weight LLM; specific model may change); reached directly, no gateway. Chat processor for memory content, including imported Google Calendar content | Chat messages + memory context (only structured IDs masked; may include imported calendar events) | US-incorporated (San Francisco); distributed multi-cloud processing | Zero data retention + no training in their Terms of Use (binding); SCCs (their Privacy Policy) + TIA |
| OpenAI (direct) | Text embeddings for memory indexing (may include imported calendar content) + failover chat inference when the primary chat provider is unavailable | Memory text for embeddings; chat messages + memory context during failover | US | No training on API content (API default); requests retained at most 30 days for abuse monitoring, then deleted; DPF or SCCs + TIA |
| Google — Gemini (via OpenRouter) | Image and scanned-document OCR | User-uploaded images and documents (may contain arbitrary PII) | US | DPF or SCCs + TIA |
| Cloudflare (R2 + reverse proxy) | Media object storage; CDN/proxy fronting the API | Uploaded audio/images; IP/traffic metadata | EU (R2 bucket in EU region) | Data at rest in EU; SCCs + TIA for US-parent access |
| RevenueCat | Subscription entitlement source of truth | User UUID, transaction/product IDs, entitlement status (no card data) | US | DPF or SCCs + TIA |
| Google — Firebase Cloud Messaging | Push notifications | Device push tokens + notification payloads | US | DPF or SCCs + TIA |
| Brevo (Sendinblue) | Transactional email (verification, reset, security notices) | Email address + message body | EU (France) | EEA — no transfer |
| Railway | Application, database, and cache hosting | All runtime data in transit and at rest | EU | Data at rest in EU; SCCs + TIA for US-parent access |
| Apple / Google Sign-In | Identity (ID-token sign-in) | ID token → email/name | US | DPF or SCCs |
| SearXNG (self-hosted) | Web search during chat tool-calling; federates to upstream search engines | Search queries derived from chat | self-hosted → upstream engines | per upstream engine |
| Sentry | Crash and error monitoring / app stability | Crash reports, error events, device/OS/app version, technical context (may include IP + user ID) | US / EU data region | DPF or SCCs + TIA |
| Aptabase | Privacy-preserving product analytics | Anonymous, aggregated usage events (no PII, no persistent device IDs, no cross-app tracking) | EU region | EEA — no/limited transfer |
Integration sub-processors (only if you connect them)
| Sub-processor | Purpose | Data shared |
|---|---|---|
| Google (Calendar) | Import calendar events into your memory | Event subject, location, attendees, times |
| Notion | Import Notion pages into your memory | Page titles and block content |
Note on imported data: when you connect an integration, Memini may process personal data about third parties (calendar attendees, Notion collaborators). For that data Memini acts as an independent controller under a documented legitimate-interest assessment.
Bring-your-own-key (BYOK) providers — not Memini sub-processors
These providers are available for bring-your-own-key inference. When you supply your own API key, they process your data under your account and terms, not as Memini sub-processors. Review each provider’s terms before use.
| Provider | Training default | Jurisdiction |
|---|---|---|
| OpenAI | No training by default (API); verify your account settings | US |
| Anthropic | No training by default (API); verify your account settings | US |
| Google (AI Studio) | Free tier may use data for product improvement; paid / Vertex: no training | Global |
| Groq | States no training; formal retention terms weaker | Undisclosed |
Managed vs BYOK: for Memini-managed AI processing (our key), the sub-processor table above applies, with the no-training and retention terms shown per provider. BYOK is a separate path: your key, your provider, your terms, your risk.