v3.5.0

What's new

Foundry IQ knowledge sources - 4 net-new preview integrations.

• SharePoint (remote) - live retrieval from SharePoint via the Copilot Retrieval API, with per-user ACL through on-behalf-of. See SharePoint remote howto.
• OneLake (indexed) - Fabric OneLake as a knowledge source, indexed and queried natively by Foundry IQ. See OneLake howto.
• SharePoint (indexed) - SharePoint indexed by Foundry IQ using app-only Graph auth (Sites.Selected preferred). See SharePoint indexed howto.
• Web grounding (Bing) - public web results scoped by allow/block domain lists, billed per Bing call. See Web grounding howto.

All four are OFF by default and gated by their own App Configuration flag plus knowledge-source name. Existing deployments upgrade with no config change.

Standard-mode azd provision + postProvision.ps1 now seeds every one of these keys into App Configuration with safe empty / false defaults, so provisioning is a no-op for operators who do not opt in.

Component pins

• gpt-rag-orchestrator: v3.5.0
• gpt-rag-ui: v2.3.13 (unchanged)
• gpt-rag-ingestion: v2.4.14 (unchanged)
• ailz (bicep-ptn-aiml-landing-zone): v2.3.0 (unchanged)

Validation

Standard-mode deployment validated: regression pass with all new knowledge sources OFF (no behavior change vs v3.4.3), and stub pass with each new knowledge source ON (App Configuration keys wired end to end into the orchestrator).

pytest config/search/tests on develop: 30 passed.

Follow-ups

• SharePoint Indexed Bicep module lands in a follow-up PR to Azure/bicep-ptn-aiml-landing-zone (submodule pinned at v2.3.0 today).
• aka.ms/gpt-rag/sharepoint-indexed-prereqs vanity URL registration.
• Fabric workspace RBAC for OneLake knowledge source is a manual prerequisite (documented in the howto).

Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com
Copilot-Session: 37105e3a-7d9e-4431-906c-586e1938220b