Semantic layer

A semantic layer is a single place where your metrics and dimensions are defined, sitting between the raw database tables and whoever is asking a question. It is what makes revenue mean the same thing in a dashboard, a report, and an AI agent’s answer.

Without it, every tool and every person rebuilds the same logic from scratch. One report counts trial users as customers, another does not, and the numbers stop matching. The semantic layer fixes the definition once so everything downstream agrees.

Why it matters for agentic analytics

An analytics agent is only as trustworthy as its definitions. If the agent has to infer what active user means, it will sometimes get it wrong. A semantic layer hands the agent the real definition, so it computes the metric the way your company actually does. This is one of the core pieces of context engineering.

How it relates to text to SQL

Text to SQL writes a query from a sentence. A semantic layer tells that query what the words mean. The two work together. The semantic layer supplies the definitions, and the query generation uses them, which keeps the agent from inventing its own version of a metric.