Data Architect BI Developer Business Decision Maker Platform Owner — Use this guide to choose the Fabric IQ capabilities that add trusted business context to analytics, plans, agents, and operational decisions.
Microsoft Fabric IQ
A governed business context layer over OneLake that helps people and agents reason using shared metrics, concepts, relationships, plans, and operational signals.
What is Fabric IQ?
Microsoft Fabric IQ is the business context layer of Microsoft IQ. It combines governed data in OneLake, Power BI semantic models, and operational context so people and agents can work with consistent business concepts instead of relearning raw tables and schemas for every solution.
Fabric IQ is a preview workload in Fabric. Operations Agents are generally available, and Microsoft announced general availability for Graph and Plan during Build 2026. Fabric Data Agents are documented as generally available on the standard runtime, but the Copilot/AI feature-state table still lists the Data Science row containing "Data agent" as preview, and several agent capabilities carry their own preview labels — see the full Data Agents guide. Ontology remains in preview, as does the Ontology MCP server, and the Fabric interface and Learn documentation label the grouped IQ workload as preview. Apply the status published for each item rather than one status to the entire experience.
Three Layers of Business Context
How unified data, business intelligence, and operational intelligence combine so people and agents share the same meaning.
1. Unified Data
OneLake unifies analytical, operational, and real-time data. Shortcuts, mirroring, and the OneLake catalog make governed data available without rebuilding a separate context layer for every solution.
2. Business Intelligence
Power BI semantic models provide trusted measures, dimensions, hierarchies, and relationships for reporting and agent grounding.
3. Operational Intelligence
Ontology adds entities, properties, relationships, rules, and actions. Graph, Plan, and agents use context to analyze connections, coordinate plans, monitor conditions, and support governed action.
Fabric IQ Items
The building blocks of Fabric IQ and the current release status of each one.
| Item | Use it for | Current status |
|---|---|---|
| Semantic model | Trusted KPIs, dimensions, hierarchies, and analytical relationships | Generally available |
| Ontology | Shared vocabulary, cross-domain relationships, rules, data bindings, and NL2Ontology queries | Preview |
| Graph | Nodes, edges, paths, GQL queries, and relationship-heavy analysis over OneLake data | Generally available; Data Agent graph grounding is preview |
| Plan | Budgets, forecasts, scenarios, variance analysis, and writeback to Fabric SQL | Generally available |
| Data Agent | Read-only conversational analytics across selected governed sources | Generally available |
| Operations Agent | Scheduled monitoring of Eventhouse or ontology data with rules and configured actions | Generally available |
Ontology: Shared Business Meaning
Reusable entity types, properties, and relationships bound to governed OneLake data.
An ontology defines reusable entity types such as Customer, Shipment, Product, and Sensor; their properties; and the relationships between them. Data bindings connect those definitions to OneLake data, including lakehouse tables, eventhouse data, and Power BI semantic models.
- Generate or align from semantic models: reuse established business terminology and KPIs where appropriate.
- Bind multiple domains: connect concepts to real data rather than limiting the ontology to one model.
- Explore relationships: the ontology graph is provided through Graph in Microsoft Fabric.
- Query in business language: NL2Ontology translates natural-language questions into structured queries.
Use ontology for pilots that benefit from cross-domain meaning and relationship reasoning, but account for preview change risk. The ontology graph also requires refresh before upstream data changes become visible.
Grounding Agents in Ontology
What the semantic layer adds to a Data Agent — and when that extra modeling effort is actually worth it.
Configuration layers, example queries, the security model, licensing, consumption surfaces, and the complete limitation list are documented in one place: AI & Copilot → Fabric Data Agents. This section covers only the Fabric IQ angle — grounding an agent in ontology rather than in raw tables.
Why ground an agent in ontology?
A Data Agent pointed at a lakehouse reasons over tables and columns. It has to infer that DimCust relates to FactOrd, and that your business calls the result a "customer order." An agent grounded in an ontology reasons over business concepts and relationships that you defined once — so the meaning is modeled in the platform instead of restated in every agent's instructions.
🏷️ Shared meaning, defined once
Concepts, properties, and relationships live in the ontology. Multiple agents, and other Fabric IQ experiences, inherit the same definitions instead of each drifting on its own.
🔗 Relationships are first-class
Cross-domain questions that would need multi-table joins become traversals over modeled relationships — which is exactly where table-level reasoning is weakest.
🌐 Cross-source context
An ontology can bind to a semantic model, lakehouse, or KQL database, giving the agent one coherent business view over sources that are physically separate.
What to know before you rely on it
- Ontology as an agent data source is currently a preview capability — confirm its status for your tenant before designing around it.
- Schema selection isn't available for ontology sources the way it is for SQL and Eventhouse sources — the ontology itself is the scoping mechanism, so model it deliberately.
- Example queries aren't supported for ontology sources. Your tuning levers are the ontology model and the agent/data source instructions, not few-shot pairs.
- Permissions compound. A user needs read on the ontology item and on the bound semantic model, lakehouse, or KQL database underneath it. Missing either yields authorization errors or empty results.
- Ontology counts toward the five data sources per agent limit like any other source.
A Data Agent can connect directly to supported Fabric data sources. Add ontology when shared concepts or cross-domain relationships materially improve the scenario — don't make ontology a prerequisite for every agent. A single-domain agent over one well-named warehouse rarely needs it.
Operations Agents
Scheduled rule evaluation over real-time operational data with configured, auditable actions.
Operations Agents are generally available Fabric items for monitoring real-time operational conditions. They use an Eventhouse/KQL database or ontology as context, generate an inspectable playbook, and evaluate each rule's query every five minutes.
- State conditions remain true while a value satisfies a threshold; transition conditions signal when a value crosses into or out of a state.
- Teams messages are available by default, and additional supported actions can be configured for recommendations or execution.
- Each agent receives a dedicated Microsoft Entra Agent ID for visibility and auditing.
- The agent operates in delegated mode using the creator's authorization; approved actions run on behalf of the creator.
- A paid Fabric capacity is required; trial capacities aren't supported.
Data Agents vs. Operations Agents
| Aspect | Data Agent | Operations Agent |
|---|---|---|
| Primary job | Answer an on-demand question | Monitor defined operational conditions |
| Data access | Multiple selected analytical sources | Eventhouse/KQL database or ontology |
| Behavior | Read-only query and response | Scheduled rule evaluation and configured actions |
| Identity | Requesting user's permissions | Dedicated Entra Agent ID with creator-delegated authorization |
| Status | Generally available | Generally available |
Microsoft IQ and Agent Integration
How Fabric IQ relates to Work IQ, Foundry IQ, and Web IQ, and where agents can be published.
Microsoft IQ brings together four complementary context systems:
| Capability | Context provided |
|---|---|
| Work IQ | How employees work across Microsoft 365 |
| Foundry IQ | Authoritative organizational knowledge for agents |
| Web IQ | Current public context from the web |
| Fabric IQ | Business entities, metrics, relationships, plans, and operational data |
Data Agents can be published into Microsoft 365 Copilot and integrated with Copilot Studio, Microsoft Foundry, Teams, custom applications, and multi-agent solutions. Ontology as a Foundry knowledge source and Fabric IQ as a first-party MCP tool in Microsoft Agent 365 were announced in preview. Keep those preview integrations separate from the GA status of the core Fabric IQ platform.
Getting Started with Fabric IQ
A staged path from a single business decision to validated, promoted Fabric IQ solutions.
Step 1: Choose the business decision
Start with a measurable question, planning process, relationship problem, or operational condition rather than adopting every IQ item at once.
Step 2: Curate the foundation
Confirm source permissions, data quality, semantic definitions, capacity region, and Purview controls. Reuse trusted semantic models when they fit.
Step 3: Select the smallest useful item set
Use a Data Agent for Q&A, Graph for relationship analysis, Plan for forecasting, ontology for shared cross-domain meaning, or an Operations Agent for monitored rules.
Step 4: Validate and promote
Test permissions, generated queries, answer quality, rule behavior, and supported actions. Use Git and deployment pipelines for Data Agent changes before production rollout.