Stark Informatics
Home · Solutions · Fabric IQ

Fabric IQ Preview

The new workload for unifying business semantics across data, models, and systems. Ontologies, plans, Fabric Graph, Data Agents, and Operations Agents — Fabric IQ is the layer that makes governed AI work.

PreviewWorkload· 9 min read

What it is

Fabric IQ is the semantic ground beneath Fabric AI. It defines how your business actually works — the entities, the relationships, the metrics, the policies — and exposes that definition to agents so they can reason correctly. Without an ontology, agents treat "revenue" the way an LLM does. With one, they treat it the way your CFO does.

Ontology

An ontology in Fabric IQ is a curated definition of your business domain: entities (Account, Order, Product), relationships (Account has-many Orders), attributes, and KPIs. It can reference semantic models, lakehouses, and KQL databases as its data backing, so the ontology stays grounded in real data rather than abstracted away.

Fabric Graph

Fabric Graph is the graph database that makes the ontology queryable. It stores entity instances and relationships, supports graph traversal queries (Cypher-style), and integrates with lineage, security, and Data Agent retrieval.

Plan

A Plan is a Fabric IQ item that defines goals, constraints, and decision logic — used by Operations Agents to reason about what action to take. Think of it as the "if/then" layer expressed at a business-policy level rather than at the code level.

Agents on top

Once the ontology, graph, and plans are in place, agents become small and powerful:

  • Data Agents ground in the ontology to answer questions in business terms.
  • Operations Agents apply plans against live data to take actions.

Adoption path

  1. Pick one business domain. Finance, supply chain, or customer ops. Not all three.
  2. Map the entities. 10–25 entities is plenty for v1. Document each in the ontology.
  3. Wire to the semantic model. Each entity binds to tables, measures, or KQL queries.
  4. Build one Data Agent. Validate ontology fidelity through real questions.
  5. Add a Plan + Operations Agent. One high-value automation. Recommendation-mode first.
  6. Iterate the ontology. Real usage exposes gaps. The ontology is a living artifact.

Build an ontology that actually works

Most ontology projects fail by trying to model everything. Ours start with one domain and ship in six weeks.

See ontology starter