Rill launches Metrics SQL — a semantic layer that speaks plain SQL

What Rill built

Rill says it has doubled down on a simple idea: metrics — revenue, MAU, ROAS — are the core primitives of a semantic layer. Why teach a whole new language when every analyst, BI tool, and AI agent already knows SQL? It has been reported that Rill’s new offering, Metrics SQL, exposes a metrics-first semantic layer that you query like a table: pick dimensions, pick measures, filter, order — no aggregation logic to write in your query.

How Metrics SQL works

Under the hood, Rill stores metric views as SQL embedded in YAML — YAML provides the structure, SQL the expressions. A Metrics SQL request is compiled through three logical layers before hitting an OLAP engine, and the compiler expands computed dimensions and parameterizes literals to guard against injection. It has been reported that Rill has examples mapping Metrics SQL to ClickHouse, and that the system also supports DuckDB, Snowflake, and Druid SQL dialects among others.

Why this matters

Definitions drift. A revenue calculation lives in dbt, Looker, a Python notebook, and maybe an AI agent’s metadata — and then finance asks, “Why don’t the numbers match?” Cue the collective eye-roll. Metrics SQL aims to give you one source of truth, accessible with the lingua franca of data: SQL. Julian Hyde, co-creator of Apache Calcite, allegedly pressed a similar point at Data Council — define measures as first-class SQL objects — and Rill’s design follows that playbook.

Limits and the road ahead

Rill positions Metrics SQL as a pragmatic bridge between human workflows and agent-driven automation, promising simpler queries and safer parameter handling. Of course, vendor claims deserve scrutiny: adoption, interoperability with existing metric registries, and real-world scale will be the proving grounds. Still — if it works as promised — this could cut through a lot of semantic-layer noise. Who wouldn’t want fewer arguments about whose revenue is right?

Sources: rilldata.com, Hacker News