Malloy (Précis)

Malloy is an open-source semantic modeling and query language that compiles to SQL. Built in TypeScript as a ~93K-line monorepo, it supports BigQuery, Snowflake, DuckDB, PostgreSQL, MySQL, Trino, Presto, and Databricks as execution engines. The core innovation is a two-phase compilation architecture: Malloy source → Translator → serializable IR → Compiler → dialect-specific SQL. The IR is a JSON-serializable representation of the entire semantic model, enabling caching, transmission, and multi-dialect compilation from a single parse.

Key technical achievements: symmetric aggregates that correctly handle joined data without double-counting (path-prefixed expressions like line_items.amount.sum() specify the aggregation grain); a pipeline query model where every query transforms a source into a new source, enabling composable analytics; a trait-based dialect system with ~40 boolean flags per database adapter; a Solid.js + Vega rendering layer with a plugin system for custom visualizations; and a tag/annotation system that serves as both user-facing renderer hints and a compiler metadata side-channel.

The project was originally developed at Google by a team that included the creator of Looker (Lloyd Tabb) and represents a language-first alternative to SaaS semantic layers.