Email data and privacy

Cabtek Brain imports the Gmail message text you select for search and source-backed summaries. Spam, trash, drafts, attachments and messages consisting entirely of automated notifications are excluded from this pilot.

Every connected mailbox, including a shared inbox such as jobs@, starts private. Its owner may share it with all staff, including people added later, and make it private again. A permissions administrator can explicitly grant colleagues access to selected mailboxes or all connected mailboxes, including future connections. The same access applies to generated summaries and Claude retrieval.

Selected text is processed on Cabtek's dedicated server. Embedding and synthesis requests use the dedicated Cabtek OpenRouter account. Claude receives results requested through your authenticated connector.

Ask the brain. Questions typed into the portal's chat surface, and the answers, are processed on Cabtek's dedicated server and sent to the same dedicated Cabtek OpenRouter account; answers are model-generated and may be incomplete. Web search is on by default: when the assistant uses it, the search query leaves Cabtek and goes to an external search engine through OpenRouter. Staff can switch web search off for a conversation, and the query the assistant sends is required to contain topic terms only — never mailbox text, page content, a customer name taken from a document, or a quoted email.

Disconnecting removes the active email index and its generated summaries. Encrypted backups are retained separately under the backup policy; older copies can remain for up to 60 days. Already retrieved results in Claude conversations are outside the brain's index.

Contact swan@whitehorseai.ai for access changes or deletion requests.