Privacy policy
RareMachine helps your team capture and act on leads — through a WhatsApp agent, digital business cards, QR-code capture, events and CRM sync — on top of the CRM you already have. The single most important thing to understand about how we handle data is that we are not the system of record — and most of this page is the consequence of that.
Last updated 30 September 2026
Who we are
RareMachine is operated by Whitesense AI Technologies Private Limited, India. For anything in this policy, write to [email protected].
The short version
Your CRM stays the system of record. We store an encrypted OAuth access token for the CRM you connect — never your CRM password, which we never see — the mappings that connect a person to their CRM identity, conversation state for seven days, and an audit trail of every action taken. We do not maintain a synchronized CRM database. Conversation history can contain CRM values returned by tools, and lead captures store the information you submit for review and sync.
What we store, and for how long
| Data | Why | Retention |
|---|---|---|
| Account details | Your name, email and workspace so you can sign in and be invited by colleagues. If you sign in with Google, this comes from your Google profile — we never receive your Google password. | Until you delete the account. |
| CRM connection token | An OAuth access token for the CRM you connect — never your CRM password, which we never receive. Encrypted with a per-workspace key. | Until you disconnect or uninstall. |
| Identity mappings | The link between a WhatsApp number, a RareMachine user, and a CRM owner record — so the agent acts as the right person. | Until you delete the account. |
| Conversation messages | Web and WhatsApp conversation content, including tool results and CRM values, so the agent can follow context. | 7 days, then deleted automatically by a daily sweep. |
| Lead captures and product content | Submitted lead details, notes, photos and attachments, plus team cards, events and branding. These are product-owned data, not a synchronized CRM mirror. | Until removed by the user, workspace deletion or applicable capture cleanup. Contact support for the policy applying to your workspace. |
| Audit entries | A record of every tool call the agent made or was denied — the thing that lets you answer "who changed this, and when". | Retained for accountability; see below. |
CRM reads are fetched live rather than synchronized into a separate CRM database. Returned values can remain in conversation history until its retention sweep, and captures, notes and attachments remain product data. Disconnecting the CRM removes our stored credentials and grants; it does not by itself erase conversations or captures.
Where your data goes
Using RareMachine necessarily means data leaves your CRM — that is what an agent is. We would rather be specific about it than bury it.
| Recipient | What reaches them | Why |
|---|---|---|
| Meta (WhatsApp Business Platform) | The content of messages between your team and the agent, and the phone numbers involved. | WhatsApp is the channel. Meta processes every message in it. |
| Portkey | Prompts and tool results, which include the CRM fields relevant to a request. | Portkey is the AI gateway every model call transits. |
| The model provider behind Portkey | The same prompt content. | The model that actually answers. Named on the sub-processor page. |
| Amazon Web Services | Everything in the table above it — AWS hosts our infrastructure, including data storage and encryption key management. | Hosting and encryption key management. |
| Brevo (Sendinblue SAS, France) | Email addresses and the contents of transactional email (sign-in links, verification, invitations). | Transactional email delivery. |
The current list, with the specific model provider in use, is maintained at sub-processors. We will give notice there before adding or changing one.
What we ask your CRM for
A workspace connects to one CRM at a time (HubSpot, Frappe or Zoho CRM), and each teammate signs in to it with their own account. When you connect HubSpot, we request read and write access to contacts, companies and deals, access to contact field definitions, read access to company/deal field definitions, and read access to owners. Reading your field definitions is what lets the agent check a change against your actual setup — including any custom fields you've added — instead of assuming every account looks the same; owner access supports action attribution. V1 does not import your user roster: administrators invite teammates manually and each teammate authorizes their own HubSpot account. Every change that writes to your CRM requires a human approval before it runs.
Where your data is held
Workspaces are hosted in Mumbai, India by default. Sub-processors listed above may process data in other regions under their own terms.
Your rights
You can ask us to export or delete your personal data, and we will action it. Because we handle conversation content and captures on your workspace's behalf, requests about authoritative CRM records also need to be directed to your CRM. We can export or erase the product data we hold subject to applicable retention requirements.
- Access and export — write to us and we will provide what we hold.
- Deletion — deleting a workspace removes credentials, mappings and conversations.
- Audit entries are retained even after related data is removed, because an audit trail you can edit is not an audit trail. They record the action, not the record contents.
- Disconnecting a CRM deletes our stored credentials and per-user grants. Manage the app's external authorization separately in your CRM.
Security
Credentials are encrypted with a unique key for every workspace, and that key is itself protected by the configured key-management provider. Production deployments require independently recoverable keys; the local development stack uses a test-only key provider. Tenant-scoped database queries and row-level security provide layered isolation. These controls reduce risk; they are not a guarantee against every application bug. Credentials and other secrets must not be written to logs, error messages or API responses.
Changes
We will update the date at the top when this changes, and give notice on the sub-processor page before a new recipient starts receiving data.