REQ-007 — Microsoft 365 / Teams integration

Need (what is being asked)

Integrate the AI platform with the existing Microsoft 365 estate: Entra ID as identity provider (SSO + MFA everywhere), Graph API ingestion of SharePoint/OneDrive files and selected mailboxes into RAG, Teams as a daily assistant surface, and a Teams-meeting → transcription → summarized-note flow into the knowledge base.

Context (why — what problem it solves)

M365/Teams is where employees already work; the assistant must meet them there or adoption suffers. Email and SharePoint hold much of the unstructured knowledge.

Current state (how it is done today)

M365 + Teams in use (licensing tier to validate); no integrations.

Success criteria

• SSO live at platform GA; SharePoint ingestion with per-user ACL respect; Teams access to the assistant by pilot start; meeting-summary flow piloted by ~M7.

Dependencies / related

Approach: integrations-m365-softone. Related: REQ-005 (SSO for Mission Control), REQ-008 (security), REQ-014 (pilot — first concrete mailbox-ingestion case).

History

• 2026-07-03 — captured (source: CEO transformation brief)
• 2026-07-15 — scoping finding (source: sample order-email analysis, REQ-014 pilot sample corpus): the plain M365 search connector does not expose nested message/rfc822 attachment content (“nested item content is not exposed”) — only email metadata. Mailbox ingestion for any order/quote flow (starting with the REQ-014 pilot) needs full Graph API /messages/{id}/attachments/$value access (or IMAP), scoped via an application access policy on the target mailbox — a hard requirement, not a nice-to-have, since line items live exclusively in the attachments.
• 2026-07-15 — finding refined: manually reading the nested PDFs via Outlook web (proving out what the automation must do programmatically) showed the payload is not just plain text — it’s PDF documents (quotations, purchase orders, sales-order acknowledgments, air waybills) nested inside forwarded emails. The Graph/IMAP access above must therefore recurse into nested rfc822 parts and feed extracted attachments through PDF text extraction with an OCR fallback — text-only parsing of the message body is not enough.