Overview#
Fabric's AI and API capabilities expose governed data to conversational and application experiences. Fabric Data Agent grounds natural-language questions in selected Fabric data sources, Fabric IQ provides an ontology-oriented layer for business meaning, and API for GraphQL exposes Fabric data through application-friendly endpoints [S1] [S2] [S3].
Core concepts#
Fabric Data Agent connects a conversational experience to curated sources such as lakehouses, warehouses, KQL databases, and semantic models, so users can ask questions over governed Fabric assets rather than over arbitrary scraped context [S1]. Fabric IQ focuses on business entities, concepts, and relationships that help data consumers interpret enterprise data in a shared vocabulary [S2]. API for GraphQL provides a programmatic API surface for selected Fabric data, letting applications request exactly the fields they need through GraphQL queries [S3].
These capabilities sit above the same governed estate. They are most useful when the underlying sources are curated, endorsed, and secured before the AI or API layer is exposed [S1] [S2] [S3].
How it works and best practices#
Treat AI and API endpoints as publishing surfaces, not data-cleaning tools. Select curated sources, keep semantic definitions consistent, and ensure permissions match the intended audience before exposing a data agent or GraphQL endpoint [S1] [S3]. Use Fabric IQ concepts where users need shared business meaning across data products rather than only table and column names [S2].
For data agents, evaluate answers against known questions and expected source coverage before broad rollout [S4]. For GraphQL APIs, design queries around application use cases and avoid exposing broad, unbounded data access where a narrower endpoint would be safer [S3].
Implementation example#
AI and API surfaces should be designed as controlled publication layers over curated Fabric assets. A data agent should be grounded in selected Fabric sources, while GraphQL should expose a deliberately shaped application contract rather than every raw table [S1] [S3] [S4].
# Illustrative app query over a curated GraphQL contract.
query ProductAvailability($sku: String!) {
product(sku: $sku) {
sku
name
currentInventory
replenishmentStatus
}
}
Inference: the query shape is illustrative. The sourced claim is that Fabric API for GraphQL exposes selected Fabric data through a GraphQL endpoint [S3].
data_agent_release_check:
grounded_sources:
- certified_semantic_model
- curated_lakehouse_tables
evaluation:
- expected_questions
- source_coverage_review
- permission_review
- answer_quality_review
The release checklist follows from the Data Agent concept and evaluation coverage: grounded sources and evaluation should be explicit before broad publication [S1] [S4].
What goes wrong#
A data agent over poorly governed data can produce confident but unhelpful answers because retrieval can only be as trustworthy as the sources it is allowed to use [S1] [S4]. A GraphQL endpoint that mirrors raw storage too broadly can become an accidental public contract for data that was not designed as an application API [S3].
Internals#
Architecture & design#
The AI/API layer depends on source selection, permission boundaries, metadata, and model or schema design rather than on a new data store [S1] [S2] [S3]. Data Agent, Fabric IQ, and GraphQL therefore extend the governed Fabric estate instead of replacing it [S1] [S2] [S3].
How it works internally#
Fabric Data Agent uses configured Fabric sources as its grounding context and should be evaluated against expected answers and source behavior [S1] [S4]. GraphQL endpoints expose selected data through a schema and resolver surface that applications query directly [S3].
Performance characteristics#
The current knowledge base has stronger conceptual and practitioner coverage than internals coverage for this topic. Performance-sensitive deployments should validate prompt latency, query latency, source size, and API response patterns in their own tenant before publishing broad SLAs [S1] [S3] [S4].
Worked example#
A support application can use API for GraphQL to retrieve customer and product data for a specific case, while a Fabric Data Agent answers analyst questions over a certified semantic model and curated lakehouse tables [S1] [S3]. Fabric IQ can provide shared business vocabulary so users ask about business concepts rather than memorizing physical table names [S2].