Sub-processors
WIP — work in progress, not a final document. This page is published so the text can be read and reviewed while it is still being worked on. It is not in force, binds nobody, and its wording will change.
Version 0.2 — draft. Page versioning starts at 1.0 on publication.
Foyer relies on a short list of companies to operate its service. Each clinic's data processing agreement entitles it to advance notice before a new sub-processor is added or an existing one replaced; this page and its change log are part of that mechanism.
Standing statement: conversation content between patients and clinics is never used to train AI models — neither by Foyer nor by any model provider listed below (model providers are bound by commercial terms excluding training on customer content).
What crosses a deployment's perimeter depends on the deployment level purchased; each level is sold explicitly and documented in our data processing agreement:
- Cloud tiers (shared multi-tenant or dedicated instance): conversation content is processed by the cloud model provider listed below.
- On-premise with a cloud model (BYOK): content still goes to the model provider — on the clinic's own key — and does leave the clinic's perimeter.
- On-premise with a local model: none of the cloud entries below receive conversation content. The only outbound traffic is an hourly technical report — software versions and usage counters, never message text.
| Sub-processor | Role | Data categories | Country | Status |
|---|---|---|---|---|
| Anthropic PBC | Cloud AI model: extracting facts from clinic websites, drafting FAQ answers, generating dialogue responses for cloud-tier clinics | Clinic website content; conversation content of clinics using cloud tiers | USA | Active |
| Cloudflare, Inc. | Website hosting, DNS, network protection for foyerbot.com | Visitor network metadata (IP, user agent, timestamps) | Global network (US-headquartered) | Active |
| Google LLC | Email correspondence (Google Workspace) | Email content and metadata of our correspondence | Varies by data centre | Active |
| To be announced | Application and database infrastructure for the platform | All platform data for cloud deployments | Not yet selected | Pending selection — no clinic data is processed yet |
| Meta Platforms Ireland Ltd. / Meta Platforms, Inc. | WhatsApp Business Platform — message delivery on behalf of clinics | Message content in transit; phone numbers; connection events | Ireland / USA | Planned (with first connected clinic) |
| Stripe Payments Europe, Ltd. / Stripe, Inc. | Subscription billing for clinics | Billing contact details, payment card data handled by Stripe directly | Ireland / USA | Planned (M4; not engaged) |
| Telegram Messenger Inc. | Internal relay of email sent to our published contact addresses into a private bot chat, so correspondence is seen and answered within our stated response times | Subject, sender and body of email people send us — including any personal data the sender chooses to include | Varies (see Telegram's published policy) | Active |
Change log
| Date | Version | Change |
|---|---|---|
| August 27, 2026 | 0.2 | Initial published list |
New sub-processors are announced here at least 30 days before they begin processing clinic personal data, except where a shorter period is required for a security patch — in which case the change is announced retrospectively within the same day.