What is a semantic layer?
Last reviewed 7 October 2026
A semantic layer is software that sits between stored data and the tools that query it, translating tables and columns into named business terms such as revenue, active customer or churn. Each term is defined once, with its calculation, dimensions and access rules. Every dashboard, spreadsheet or AI assistant that asks for the term then gets the same answer.
Also called: metrics layer, metrics store, universal semantic layer, headless BI.
Why Semantic layer matters
Ask three teams for last quarter's revenue and you can get three numbers: one excludes refunds, one counts by invoice date, one by payment date. Each was calculated inside a different BI tool, so nobody can see why they differ. A semantic layer moves the calculation out of the individual tools into one shared definition, so a change to the rule is made once and shows up everywhere.
The idea has gained weight with AI assistants. A model asked to write SQL against raw tables has to guess which column means what. Given a semantic layer, it can ask for the metric by name and receive the approved calculation. AtScale's glossary makes this point directly, describing the layer as turning complex data into consistent terms for people and AI.
How it works
Definitions vary between vendors, so it helps to look at what they share. AtScale lists five things a semantic layer centralises: metrics, dimensions such as time period or customer segment, relationships between tables, business terminology and access rules. dbt's version, powered by its MetricFlow engine, defines metrics on top of existing dbt models and handles the joins automatically.
At query time a tool asks for a metric and a set of dimensions, for example revenue by region by month. The semantic layer compiles that request into SQL for the underlying warehouse, applies the access rules for the person asking, runs it and returns the result. Some products cache or pre-aggregate frequent requests; others always query live. The tool never sees the raw join logic.
Delivery differs too. Some layers expose an API and a SQL interface that BI tools connect to; dbt also lists partner integrations and an MCP server so AI tools can query metrics. The term is used loosely: some vendors use it for a full product, some for a modelling feature inside a BI tool. Ask where the definitions live, who can change them and which tools can read them.
Worked example: one definition of active customer
A subscription business argues every month about how many active customers it has. Finance counts paid accounts, product counts accounts with a login in the last thirty days, and sales counts open contracts. The data team writes one definition in the semantic layer: an account with a paid invoice in the period and no cancellation, broken down by plan and region.
The finance dashboard, the product team's spreadsheet and the assistant that answers questions in chat all request active customers from the layer. When the business decides to exclude trial conversions in their first week, the data team changes the definition once, records why, and every consumer shows the new number the next time it queries. The other two definitions are kept under their own names, so nothing silently disappears.
How Swfte relates to it
Not a Swfte feature
Swfte does not offer a semantic layer or a metrics store, and it is not a BI tool. If you run one, a Studio workflow or an agent can call its API or its MCP server like any other tool, and the metric definitions stay where your data team keeps them.
The nearest thing Swfte builds is different in kind. The company brain holds a time-aware graph of the organisation (people, groups, reporting lines and accounts today) where every fact carries an evidence status. That answers questions about who and what, not about how revenue is calculated. The Intelligence Platform views for analysing usage and outcomes are design intent, not shipped features.
Related terms
- Knowledge graph
A knowledge graph is a store of facts in which entities, such as people, products or systems, are nodes and the named relationships between them are edges.
- Business intelligence tools
Business intelligence tools are software for collecting, modelling, analysing and presenting an organisation's data as reports, dashboards and visual analyses that inform decisions.
- Decision intelligence
Decision intelligence is a discipline that treats business decisions as things to be designed, recorded, measured and improved, using data, analytics and AI to support or automate them.
- Retrieval-augmented generation (RAG)
Retrieval-augmented generation (RAG) is a method in which a system first retrieves relevant passages from a collection of documents and then gives them to a language model as context for its answer.
- MCP server
An MCP server is a program that exposes capabilities to AI applications through the Model Context Protocol, so a model can call its tools, read its data and use its prompt templates.
Common questions
- Is a semantic layer the same as a data warehouse?
- No. The warehouse stores the data. The semantic layer stores definitions of what the data means and how to calculate business terms from it, and turns requests for those terms into queries against the warehouse. You need the warehouse either way; the layer is optional.
- Is a semantic layer the same as a metrics layer?
- Often the terms are used for the same thing. Some vendors use metrics layer for the narrower job of defining calculations, and semantic layer for the wider set that also includes dimensions, relationships, business vocabulary and access rules. Read the vendor's own documentation rather than relying on the label.
- Does a semantic layer help AI assistants?
- It can. An assistant that requests a named metric receives the approved calculation instead of guessing joins over raw tables, which removes a common source of wrong numbers. It does not help with questions the layer does not model, so the assistant still needs to say when a question falls outside it.
- When is a semantic layer not worth it?
- When one team uses one BI tool and a handful of metrics, the tool's own modelling may be enough. The layer earns its cost when several tools, teams or AI assistants need the same definitions, and when disagreements about numbers are already costing time in meetings.
Sources
Definitions on this page were read on the sources below on 7 October 2026. Where sources define the term differently, the page says so. The full glossary lists more terms.
- dbt documentation, the dbt Semantic Layer (read 2026-10-07)
- AtScale glossary, semantic layer (read 2026-10-07)