Fabric Ontology Knowledge Source¶
Primary manual: https://learn.microsoft.com/azure/search/agentic-knowledge-source-how-to-fabric-ontology
What It Demonstrates¶
Fabric Ontology KS lets a Knowledge Base ground responses in Fabric ontology semantics, including business definitions, entities, and relationships.
This repo uses an Airline Ops sample contract so the Fabric path has a concrete business shape even before you have production data ready:
- sample data:
samples/data/airline-ops/ - ontology contract:
samples/ontology/airline-ops/ontology-contract.yaml - setup checklist:
docs/fabric-ontology-prerequisites.md
Notebook Walkthrough¶
Use notebooks/02-fabric-ontology-ks-airline-ops.ipynb for the guided Fabric path. It validates the Airline Ops sample facts, reviews the ontology contract, builds Fabric Knowledge Source and combined Knowledge Base payloads, optionally performs delegated live retrieve, and inspects offline Fabric traces.
Key Design Points¶
- Search service and Fabric workspace must align with tenant requirements.
- The knowledge source needs Fabric workspace and ontology item IDs.
- Retrieve calls require end-user source authorization.
- Pass the raw end-user Search access token with
x-ms-query-source-authorization. - Use
includeReferenceSourceDataduring validation to inspect returned source data. - Validate data handling when Search and Fabric resources are in different regions.
- In
byo-fabric, treat the ontology as a tenant-owned prerequisite. Infull, the sample provisions a small Airline Ops ontology for greenfield validation.
Reasoning Effort¶
This repo uses low retrieval reasoning effort consistently across the Fabric and MCP walkthroughs so answer synthesis and trace inspection behave the same way. Keep the Azure OpenAI model block in the Knowledge Base definition when using low or medium reasoning effort.
Airline Ops Validation¶
After mapping the sample data into your own Fabric ontology, start with these questions:
| Query | Expected validation signal |
|---|---|
| Which airlines have the highest customer-care exposure this month? | fabricOntology activity and a ranking led by Alpine Air |
| Which routes have the most delayed flights over 15 minutes? | Route and Flight entities are joined by ontology relationships |
| Which delay categories are controllable and driving customer-care exposure? | DelayEvent category and exposure fields are used |
For offline trace inspection, run: