Overview

What is LakeQL? #

LakeQL is a universal data access layer for modern platforms. It generates a type-safe GraphQL API across your data stack. You point its CLI at a data source, generate typed schema files from the real table metadata, and serve them through a GraphQL server you compose from explicit building blocks.

LakeQL is adapter-agnostic. An adapter teaches it how to talk to one query engine. Adapters for Trino and Microsoft Fabric are available today. ClickHouse and DuckDB are on the roadmap. A single project can register multiple adapters and route queries to different engines from one API.

The Problem #

Building a GraphQL API over your data tables is repetitive. Every table needs a TypeScript interface, resolver logic, filter and sort definitions, pagination, and permission checks. Change a table — add a column, rename a field, alter a type — and you update all of those layers by hand. Schema drift and broken types pile up fast across dozens of tables and several schemas.

How LakeQL Solves It #

LakeQL inverts the workflow. You introspect the data source and generate the typed layer instead of writing resolvers by hand:

• TypeScript interfaces derived from column metadata
• GraphQL query schemas with filtering, sorting, and pagination built in
• JSON schemas that drive runtime response transformation
• Config files that register each generated endpoint

You commit the generated code to your repository, and the API server loads it at startup. When a table changes, re-run the introspection command to regenerate the typed layer.

The Shape of a Project #

A LakeQL project is an ordinary TypeScript application with three parts:

1. lakeql.config.ts declares which adapters are available and which named data sources exist. See Configuration .
2. src/index.ts loads that config, builds a runtime that owns the data source connections, and starts the GraphQL server. See Server Setup .
3. src/schemas/generated/ holds the typed files the CLI writes when you introspect a table. See the CLI commands .

Scaffold a project to get all three at once:

pnpm create @lakeql/app my-api

How It Works #

1. Introspect — the CLI connects through an adapter and discovers table structures.
2. Generate — typed files are written: interfaces, query schemas, and configs.
3. Serve — the server loads the generated schemas and exposes a GraphQL endpoint.
4. Query — clients send GraphQL queries that translate into adapter-specific SQL.

Where to Next #