Ask SAP a business question.
Get a live, governed answer — grounded in Microsoft Fabric.

A working demonstration of an AI agent that reads its meaning from a semantic map in Microsoft Fabric and fetches its facts live from SAP — nothing copied, every answer carrying its provenance.

HOW IT FITS TOGETHER

SAP · LIVE

Business Data Graph

SAP Integration Suite · API Composition on BTP federates four SAP-shaped product lines (ERP, procurement, sales, HR — all simulated by SAPSIM) into one live graph. Stores nothing — every query hits the backends at question time.

MICROSOFT FABRIC

Fabric IQ Ontology

The map, zero data: 14 business concepts, their relationships, definitions, synonyms and source provenance — generated automatically from the SAP graph by a re-runnable projector. SAP changes? Re-run, and the map stays true.

AI AGENT

Chat, with receipts

For each question: one call to the map (Fabric), one governed live call to SAP. Both calls are visible in the conversation, and every answer states its sources and the simulation disclosure.

the ontology is DERIVED from the live graph — not drawn by hand
State lives in SAP — live federation, no copies
Structure lives in Fabric — meaning, machine-readable
The agent stands on both — map + facts, merged per question
2
calls per question
— never more
1 Ground: ask the Fabric ontology what a Customer / Supplier / Stock is and how they connect
2 Fetch: one governed call to the SAP Business Data Graph for the numbers — as they are right now
The budget is enforced in the tool layer, in code. If the agent tries a second SAP call, it is refused — on screen — and answers with what it has.

THE MOMENT THAT MATTERS

S/4-STYLE ERP · supplier no.
USSU-VSF84
=
ARIBA-STYLE PROCUREMENT · supplier id
AN01000123456
ONE REAL-WORLD SUPPLIER
resolved inside the graph

One question, two SAP products, one answer. The same supplier keyed differently in two systems is merged server-side by the Business Data Graph — declared identity, not fuzzy guessing. The agent just asks the business question.

WHAT WE ASK IT — LIVE

Q1 · CROSS-PRODUCT

“Who supplies this material, and where do we stand with them?”

Supplier resolved across ERP + procurement, with both identities and the purchase-order status — in one answer.

Q2 · SEMANTICS AT WORK

“Is this customer blocked? Credit limit and exposure?”

The agent distinguishes a credit block from an order block — a distinction it learned from the ontology, not a guess.

Q3 · LIVE STATE

“Live stock of this material, by plant?”

Quantities per plant with an as-of timestamp — fetched at question time, because stock is only true now.

Map from Microsoft. Facts from SAP. Nothing copied.

Demonstration environment with simulated SAP backend data. Fabric IQ Ontology is in preview. SAP Integration Suite / API Composition, the Business Data Graph, Microsoft Fabric and the agent runtime are production services.