Essential terms of the context layer

A single model that defines the core enterprise concepts (e.g. Customer, Order, Asset) with their properties, relationships and rules. Power BI reports, notebooks, AI agents and Copilot all read the same definition from one source. It is the executable, versioned, AI-readable modern version of what used to be called a "data dictionary" or "ERD".

Microsoft Fabric IQ Ontology, Palantir Foundry Ontology and Databricks Unity Catalog solve the same problem in different syntax: *make sure we are talking about the same thing*.

Ontology tells you *what* exists; the context graph tells you *what is happening right now* — which order is on which shipment, which sensor it passed through, whether that sensor reported a breach.

AI agents walk this live graph to answer "why is order X late?" with traceable reasoning. Gartner forecasts that 50%+ of AI agent systems will use context graphs by 2028.

A standalone LLM hallucinates — it is blind to your enterprise. Agent grounding defines which source, with which permission and which semantics an agent can access: "this agent may only query the finance ontology, read-only, and must back every answer with a source."

Microsoft Foundry IQ + Azure AI Search, Salesforce Einstein Trust Layer and Databricks Mosaic AI Gateway each implement this contract in different syntax.

Fabric IQ translates data sitting in OneLake, lakehouses, eventhouses and Power BI semantic models into a single semantic language. Core items: Ontology, Plan, Graph, Data Agent, Operations Agent.

It generalizes the Power BI semantic model concept — not just reports, but agents and apps read the same definition. Together with Foundry IQ (agent knowledge) and Work IQ (M365 Copilot context), it forms **Microsoft IQ — the unified intelligence layer**.

Work IQ remembers M365 signals (email, chat, files, meetings, calendar, work habits) and lets Copilot learn from them — building the user's personal work context.

Foundry IQ provides a centralized knowledge base for agents. A knowledge base built once is shared by multiple agents; indexing, retrieval and security controls are automated on top of Azure AI Search.

Power BI semantic model (formerly "dataset"); a model optimized for reporting with DAX measures, hierarchies and relationships. Fabric IQ Ontology generalizes the semantic model — extending semantic definition beyond analytics (agents, plans, operational decisions).

A data structure that represents knowledge through nodes (entities) and edges (relationships). Popularized by Google in 2012 for static reference use; the context graph is its procedural, agent-grounding-ready evolution.

OneLake is the single logical data lake underlying Fabric. All Fabric workloads (lakehouse, eventhouse, warehouse, semantic model) sit on the same OneLake. Cross-tenant shortcuts make data in other tenants reference-accessible.

A pattern where an LLM, before generating a response, retrieves relevant documents from an external knowledge source and adds them to the context. Agent grounding in practice is largely RAG + ontology + permission control. Foundry IQ offers the managed version on top of Azure AI Search.