A platform lead sits through a vendor deck. Slide four is the semantic layer. Slide five is the context layer, drawn above it. The next vendor's deck draws the same two boxes the other way round. That is the semantic layer vs context layer question as most people meet it: two names, one diagram, and no agreement on which box holds what.
A semantic layer is a thing you can build and point at. It is the governed record of what the data means, served to every tool that asks. "Context layer" is a position in a diagram, and the three vendors we read for this piece draw it in three different places. We retired the term in our own copy. The job it names we call context management. The meaning that job serves we call authoritative context. This post says what each term means and why we stopped using the second.
A semantic layer is the asset: the governed record of concepts, metrics, mappings, and entitlements. "Context layer" is the market's name for the job of serving that record to an AI agent. We call the job context management, and the meaning it serves authoritative context.
Our post What Is a Semantic Layer? is the full definition. A semantic layer sits between the warehouse and everything that queries it. It holds the agreed definition of each concept and each metric, the mapping from each to the physical columns, and the rules about who may see what. The market ships two versions of it.
Our post Metrics Layer vs Semantic Layer: The Difference Is a Noun takes the two apart. Every "context layer" definition below is a reaction to the first version.
It depends on who is selling. We read the three pages that hold the term for this query. The definitions agree on the reader. They do not agree on where the layer sits.
| Who | Their definition, verbatim | Where it sits | What it describes |
|---|---|---|---|
| One catalog vendor's explainer on this query (2026-01-30, updated 2026-06-30) | "A context layer wraps that semantic layer with lineage, policy, decision precedents, and sensitivity rules, then exposes it to AI agents at runtime" | Around the semantic layer, as "a superset" that "consumes and enriches semantic layer definitions" | A governance wrapper over whatever metrics layer you already run |
| A second catalog vendor's blog (2026-05-06) | "The Enterprise Context Layer is the ontology, plus the broader enterprise context that agents need to navigate real-world decisions." | Above the ontology, which that page defines as "the semantic layer, plus operational logic" | Three nested boxes, each one a bigger product than the last |
| SurrealDB, a database vendor (Ignacio Paz, 2026-06-05) | "The layer that gathers, filters, and packages data as context for an LLM or AI model. It answers: what does the model need to know right now?" | After the semantic layer, at request time; "inherently dynamic" | Retrieval and prompt assembly |
The first vendor's homepage now describes its catalog as "the context layer for AI", read 2026-11-22. Across three sites it is a wrapper, a superset, and a retrieval step. The one thing the three share is the consumer. Each definition exists because an AI agent is now reading the definitions. The term names the audience. It does not name a layer.
No. A semantic layer is an asset. "Context layer" is a pitch about what that asset should carry once an agent reads it. Take the pitch apart and ask where each promised item already lives. Most of them live in a semantic layer built on an ontology. Few of them live in a metrics layer, which is why the pitch exists.
| Promised by the "context layer" pitch | In a metrics layer | In a semantic layer built on an ontology |
|---|---|---|
| Lineage | The model file, if someone reads it | The mapping from concept to column, held as data you can query |
| Policy and sensitivity | Row and column rules per model | Entitlements on the concept, evaluated per request |
| Decision precedents, or who decided | Git history | A named owner, a version, and a review date on each definition |
| Entity meaning, or what a customer is | A join key | A concept with properties, relationships, and a validation rule in SHACL, the W3C Shapes Constraint Language |
| Exposure to agents at runtime | A SQL endpoint | The record served over API, SPARQL, and MCP (the Model Context Protocol), with the owner and version attached |
The first vendor's page on metrics layers (published 2026-09-10, updated 2026-09-18) names the missing piece. What a context layer adds, it says, is "the entity meaning, active ontology, and ownership tracking a metrics layer was never built to carry". We agree on the gap. We disagree on the fix. Entity meaning is the first row of a semantic layer, and it belongs on the same record as the metric that measures it. Our post Semantic Layer vs Ontology: Which One Says What a Customer Is argues that from the warehouse side.
Three words, each for a different thing, and one retired word that covered all three at once.
We retired "context layer" for three reasons. It names a position in a diagram, and three vendors put that position in three places. It implies a second product above the one you have, and the job is to govern the one you have. And it hides the discipline behind a box. A box ships. A discipline has to be run. The objection that all of this is a rename of older ideas is a fair one. Our post Semantic Layer, Data Fabric, Data Mesh, Context Layer: Is Any of This New? answers it: the substrate is old and the obligation is new.
The query has a twin, and we answered it already. Our post Knowledge Graph vs Context Layer: How They Relate sets out what the job adds to a graph: authority, scope, and resolution at query time. Those three apply unchanged to a semantic layer, so this post does not restate them. What is specific to the semantic-layer half is the metric. A semantic layer holds calculations over concepts, and an agent runs them. That raises four questions first.
Our post A Semantic Layer for AI Agents: What the Agent Needs Before It Asks for Revenue covers the agent's side. Our post Semantic Layer and MCP: What the Agent Actually Receives shows what arrives over MCP when the record is governed.
Put four questions to any vendor using the term, including us. The first catalog vendor's page offers one of its own: "Does it enforce policy at runtime, or only describe policy in metadata?" We would ask it back, and add three.
A vendor that answers all four has a semantic layer run as context management, whatever the slide calls it. A vendor that answers one has a metrics layer with a new name on the box.
TopQuadrant does not sell a context layer, and we do not ask you to buy a second product above your semantic layer. TQ Data Foundation is the semantic layer, run as context management. Every metric carries its business definition, its owner, its calculation, and its mapping to physical columns, on one governed asset. Concepts are modelled on open standards: RDF, OWL, SHACL, and SPARQL. TopQuadrant helped author SHACL and RDF, so the record exports in full. AI proposes definitions and mappings; your experts confirm them. Every definition is validated with SHACL. Entitlements attach to the governed concept and are evaluated per request, so an agent sees exactly what its caller may see. Your catalog stays, and it syncs. Our agent-ready semantic layer page shows the agent side of this.
Search for a context layer if that is the word you have. What you are looking for is a semantic layer with the concept row filled in, run by people who sign what they approve. That is what TopQuadrant built TQ Data Foundation to be. We call the discipline context management, and the meaning it serves authoritative context. The agent gets the definition with a name on it.