Data governance
8 min

Semantic Layer Tools: Metrics Layers, Native Views and Meaning Layers Compared

Key takeaways

  • Semantic layer tools fall into four categories: BI-native metrics layers such as Looker and Power BI, warehouse-native views such as Snowflake semantic views and Databricks metric views, standalone semantic layers such as dbt, Cube, and AtScale, and ontology-based meaning layers such as TQ Data Foundation.
  • The first three categories hold the calculation. In each, a customer is a table, a logical table, or a join key, as the vendors' own documentation describes. The fourth holds the concept, with a definition, a rule, a scope, and a named owner, and attaches the calculation to it.
  • The five competing lists we read name the same seven products, define the category as metrics, dimensions, joins, and access rules, and never use the word ontology. Three are written by catalog vendors placing their own product above the list.
  • A metrics layer is the right purchase for the dashboard problem. It stops being enough when an AI agent has to resolve what a customer is before it composes a query, or when a second domain needs a different definition of the same word.
  • Add two tests to any shortlist: ask the product what a customer is, and ask what leaves if you leave. The answers tell you which category you are buying, whatever the box says.
In this article
Share

A platform lead at a bank has a shortlist to build by Friday. The five listicles on semantic layer tools name the same seven products in a different order, graded on APIs and governance signals. None of them has a row for the question the bank's first AI agent asked on Monday: which of the forty customer tables is the customer? This article sorts the semantic layer tools market into four categories by the question each one answers, and places the named products in them from each vendor's own documentation.

Three of the four categories are metrics layers. They hold the calculation. The fourth is a meaning layer. It holds what the thing being counted is, with an owner on it.

BI-native models define metrics for one tool. Warehouse-native views define them for one platform. Standalone semantic layers define them once for every tool. An ontology-based meaning layer defines the concept the metric is about, with a named owner, and attaches the calculation to it.

What are semantic layer tools?

Semantic layer tools are software that holds business definitions once and serves them to whatever queries the data. In our post What Is a Semantic Layer? we set out the four things a layer holds: concepts, metrics, mappings, and entitlements. Most products on the market hold the second and third, and the listicles say so. Coupler's list (2026-10-28) says a semantic layer "holds the joins, metric formulas, dimensions and access rules" (Coupler). OvalEdge's list (2026-10-05) says the tools "define business metrics, dimensions, and access rules in one governed place" (OvalEdge). Cube's list (updated 2026-11-18) adds "the entities they belong to" (Cube).

What the five lists share

  • The same seven products. dbt, Cube, AtScale, Snowflake semantic views, Databricks metric views, Looker, and Power BI appear on every list we read.
  • The same definition. Metrics, dimensions, joins, and access rules. The word "ontology" appears on none of the five pages.
  • The same silence. No page asks who approved a definition, when it expires, or in what form it leaves the product.

What are the four categories of semantic layer tools?

Sort by the question the product is built to answer, and by what "customer" is inside it. Our post The Ontology Market: A Guide to Five Very Different Things already files dbt, Cube, and AtScale under one heading, the metrics semantic layer. This map splits that heading by where the metric lives, and adds the category above it.

The category map

CategoryWhat "customer" is in itProducts named in the lists we readThe question it answers
BI-native metrics layerA table in the tool's modelLooker (LookML), Power BI semantic models, Tableau, ThoughtSpot, MicroStrategyWhat does this dashboard's number mean?
Warehouse-native viewA logical table over physical tablesSnowflake semantic views, Databricks metric viewsWhat does this number mean, on this platform?
Standalone semantic layerAn entity that is a join keydbt Semantic Layer, Cube, AtScale, Honeydew, DremioWhat does this number mean, in every tool?
Ontology-based meaning layerA class with a definition, a rule, a scope, and a named ownerTQ Data FoundationWhat is a customer, who agreed that, and how much did they spend?

Catalogs appear on three of the five lists, each time as the list author's own product in a category above the semantic layer.

What do BI-native metrics layers ship?

The category most buyers already own. Google's LookML documentation (read 2026-11-22) says "LookML, short for Looker Modeling Language, is the language that is used in Looker to create semantic data models. You can use LookML to describe dimensions, aggregates, calculations, and data relationships in your SQL database" (Looker documentation). Microsoft's documentation (dated 2025-10-01, updated 2026-04-28) says "Power BI semantic models represent a source of data that's ready for reporting and visualization" (Microsoft Learn). Both describe a model one tool reads.

Two things to know before you lean on one

  • The definition is a description. In both, the sentence that says what a measure means is a text field on the tool's object.
  • The model leaves in the tool's language. LookML files in a Git repository, or a semantic model owned inside the Power BI service.

What do warehouse-native views ship?

Both large platforms now hold the metric definition as an object in the warehouse. Snowflake's overview of semantic views, read 2026-11-22, says "Within a semantic view, you define logical tables that typically correspond to business entities, such as customers, orders, or suppliers", and that "Semantic views are considered metadata". On the agent side: "Cortex Agents reads the information captured in the semantic view definition and generates the SQL against the physical tables directly" (Snowflake documentation). Databricks' metric views page (updated 2026-10-29) describes them as "separating measure definitions from the fields (also called dimensions) used to group, filter, and aggregate them" (Databricks documentation).

For an estate on one platform they are the cheapest metrics layer available. A logical table named "customers" is a pointer to a table. Our post How to Build a Semantic Layer: Platform-Neutral Steps and What Each Native Layer Leaves Out says what each native layer leaves for you to build.

What do standalone semantic layers ship?

The category the listicles rank first. Three products, each from its own pages.

Three standalone layers, in their own words

ProductWhat it defines, as documentedLicence and price, as publishedSource
dbt Semantic Layer"It operates through YAML files, where a semantic graph links language to data."MetricFlow "is distributed under the Apache 2.0 license"; "The dbt Semantic Layer is proprietary to dbt, however some components of it are open-source." Starter tier $100 per user per month, with 5,000 queried metrics.MetricFlow, 2026-11-19; FAQ, pricing, read 2026-11-22
Cube"Cube is the agentic analytics platform built on a semantic layer." "It centralizes metric definitions, joins, access rules, and caching upstream of every BI tool."Cube Core is "what we maintain as open-source". No price on the page read.Cube documentation, read 2026-11-22
AtScaleGovernance "applies your existing warehouse permissions in one place" and "Versions and reviews metric definitions like code". Named integrations include Power BI, Tableau, Excel, Looker, and Model Context Protocol."AtScale is priced on consumption."AtScale product page, read 2026-11-22

dbt's entities page (updated 2026-07-20) is the clearest statement of what a concept is in this category. "Entities are real-world concepts in a business, such as customers, transactions, and ad campaigns." The next sentence: "In our Semantic Layer models, these entities serve as a join key across semantic models" (dbt documentation). That is the category in one line. A metrics layer can define a calculation. It cannot define a concept. Our post Metrics Layer vs Semantic Layer: The Difference Is a Noun takes one metric apart field by field.

One thing this category does well is portability. Coupler's list names five of the seven products as supporters of Apache Ossie, "formerly known as Open Semantic Interchange (OSI)", whose core specification is marked "DRAFT version" (Apache Ossie, read 2026-11-22). The two specification files we read carry datasets, metrics, and concepts. Neither carries a field for who approved a definition.

Where do catalogs fit on a semantic layer list?

Three of the five lists come from catalog vendors, and each places its own product in a category above the semantic layer. Atlan's list (published 2026-05-20, updated 2026-09-18) says "A semantic layer defines what metrics mean" and describes its own entry as "the layer that sits on top of whichever semantic engine you pick" (Atlan). Its documentation, read 2026-11-22, describes the concept side as glossary terms: "A glossary term is the lowest unit that can exist independently inside a glossary" (Atlan documentation). Alation's comparison post (2026-05-06) defines an ontology as "the semantic layer, plus operational logic" (Alation).

Keep the catalog. It is where people find the data, and where the approved definition is displayed against each asset. Our post Metadata Management vs Data Catalog: Keep the Catalog, Govern the Meaning says which product should hold what. On a semantic layer shortlist the catalog is a consumer of the definition.

What does an ontology-based meaning layer add?

The fourth category holds the concept as its unit of record. An ontology is the formal model of what a business's concepts are and how they relate, written in OWL and SKOS so any tool can read it. Our post What Is an Ontology? defines it. The meaning layer maps each class to the warehouse and defines the metrics on the class. It puts a named owner, a version, and a review date on the record. It tests the data against the definition with SHACL, the W3C Shapes Constraint Language.

Customer, in each category

  • BI-native. A table in the tool's model, with a description.
  • Warehouse-native. A logical table. Snowflake's documentation says logical tables "typically correspond to business entities".
  • Standalone. An entity. dbt's documentation says entities "serve as a join key across semantic models".
  • Meaning layer. A class: a party that holds at least one policy in force, owned by a named underwriter, approved on a date, at version 3, with the metric "active customers" defined on it.

Our post Semantic Layer vs Ontology: Which One Says What a Customer Is argues why the concept comes first. Our post Ground Truth: SME-Approved vs Inferred Metadata covers what the approval adds.

When is a metrics layer enough?

Often. The dashboard problem is real, three categories solve it well, and the right one depends on who is asking.

Pick by the consumer

  • One reporting tool, people asking. The BI-native model you already own.
  • One platform, people and its native agent asking. The warehouse-native view, which the platform's agent reads directly.
  • Many tools, one definition per metric. A standalone semantic layer, served over an API and, increasingly, MCP.
  • An agent that has to say what a customer is, and who agreed. A meaning layer, with the three above as its consumers.

Our post Semantic Layer and MCP: What the Agent Actually Receives covers what comes back over MCP from each kind of layer.

What two tests should you add to the shortlist?

Two questions settle the category faster than any scorecard on the five lists.

Two tests

  • Ask it what a customer is. If the answer is a key, a logical table, or a description field, you are buying a metrics layer. If the answer carries an owner, a date, and a rule the data is tested against, you are buying a meaning layer.
  • Ask what leaves if you leave. The answer will be LookML files, a YAML repository, a CSV of glossary terms, an Ossie document, or an RDF file with its SHACL shapes. The format decides whether the next product reads the definition or a person retypes it.

Where does TopQuadrant fit?

In the fourth category, above the other three. TopQuadrant does not sell the dashboard, the warehouse, or the metrics file. TQ Data Foundation is the governed meaning layer those things read from. Every metric carries its business definition, its owner, its calculation, and its mapping to physical columns, on one governed asset. Queries resolve to Snowflake, Databricks, or Postgres without moving the data. Entitlements attach to the concept and are evaluated per request, so an agent sees exactly what its caller may see. Agents receive concepts, permitted relationships, and metric logic over API, SPARQL, and MCP. AI proposes the models and the mappings; your experts confirm them. Every definition is validated with SHACL, and TopQuadrant helped author SHACL and RDF, so the record exports whole. We maintain 300 reusable standards-based models. Your catalog stays, and it syncs. Our semantic layer use case page shows one governed metric record.

Read the listicles for the metrics layer. Then add the row they leave out. A metrics layer tells you how much. TQ Data Foundation tells you what you counted, and keeps the owner's name on the answer.

Give your AI the context it's been missing

See how the TQ Data Foundation turns your enterprise knowledge into trusted, Al-ready context.