Sub-processors
The third parties that process data on our behalf when you use RareMachine. We will post here before adding or changing one.
Last updated 18 September 2026
Current sub-processors
| Sub-processor | Purpose | Data processed |
|---|---|---|
| Meta Platforms | WhatsApp Business Platform — the channel the agent runs on. | Message content and phone numbers. |
| Portkey | AI gateway. Every model call transits it. | Prompts and tool results, including the CRM fields relevant to a request. |
| Amazon Bedrock | The primary model host for every tier — chat, routing and vision. | The same prompt content. |
| OpenAI | Fallback model provider, used when Bedrock is unavailable. Also the embedding model for help-center question search. | The same prompt content. For help-center search: the text of help articles, and a question typed into the help-center search box. |
| OpenRouter | Further fallback model routing, used when the above are unavailable. | The same prompt content. |
| Amazon Web Services | Hosting, database, cache and key management. | All data we store, at rest. |
| Brevo (Sendinblue SAS, France) | Transactional email — sign-in links, verification, invitations and account notices. | Email addresses and message contents. |
Notice of changes
We publish changes to this list here before a new sub-processor begins processing. More than one model provider is listed because our AI gateway can route a request to a backup provider if the primary one is unavailable — each is a genuine, active recipient of data, not a spare we hope never to use.
Questions
For a data processing agreement, or to ask what a specific sub-processor receives, write to [email protected].
What we do not send to any of these: your CRM records as a bulk export. Nothing in RareMachine copies your contacts, deals or companies into a durable store, ours or anyone else's. What reaches a model is the subset of fields a specific request needs, at the moment it needs them.