LakeQL is a universal data access layer that lets you query across your entire data stack through one type-safe GraphQL client and a single CLI. It combines a backend runtime with a CLI that generates schema and query files from your existing data model.
The goal is to reduce manual boilerplate while keeping your API predictable, secure, and easy to extend — no matter which data platform sits underneath.
One interface, many platforms. Connect LakeQL to the data engines your team already runs on.
| Adapter | Status | Description |
|---|---|---|
| Trino | Available | Query federated data across Trino clusters, with pushdown for joins, aggregations, and filters. |
| Microsoft Fabric | Available | Connect to Fabric warehouses and lakehouses, with Entra ID / On-Behalf-Of authentication. |
| ClickHouse | Coming soon | Sub-second analytical queries against ClickHouse clusters. |
| DuckDB | Coming soon | Query local or in-memory DuckDB files and Parquet datasets. |
- Type-safe GraphQL APIs — adapter-agnostic query generation with full TypeScript support, so you get autocomplete and compile-time checks across your data layer.
- CLI-based schema generation — pull live catalog metadata from your sources and generate typed schemas that stay in sync as upstream tables evolve.
- Filtering, sorting & pagination — built-in support for expressive filters, sorting, and pagination, rendered to each engine's SQL dialect.
- Adapter-agnostic permissions — define access rules per data source and table, enforced consistently across every connected adapter.
- OpenTelemetry tracing — optional per-data-source tracing out of the box for query latency, errors, and throughput.
- Flexible authentication — OAuth2, Basic Auth, and Bearer token auth with configurable impersonation strategies per data source.
- Node.js 24 or newer
- A package manager — npm, pnpm, or yarn
Scaffold a project with your package manager's create command:
pnpm create @lakeql/app
# or
npm create @lakeql/appThen follow the interactive setup.
The command bootstraps a project from the preconfigured template in templates/app.
Run the interactive CLI to generate your schemas.
lakeql-cli pull --env .envStart the server and query your generated types over GraphQL:
query {
orders(
sorting: [{ field: id, direction: ASC }]
paging: { page: 1, perPage: 10 }
) {
nodes {
id
status
}
pageInfo {
hasNext
currentPage
}
}
}To start local development, run:
pnpm devThe template ships with a production-ready tsdown setup.
Use pnpm build to generate the production dist output, which you can then use in your deployment setup (for example, in a Dockerfile).
Full docs, guides, and examples are available at https://lakeql.dev
Contributions are welcome — see the contribution guide to get started.
This project has its roots in the original datalake-graphql-wrapper.
LakeQL was later rebuilt from the ground up as a fully type-safe, adapter-agnostic monorepo, with a complete refactoring of the codebase and architecture.
- Igal Klebanov and the Kysely Team - LakeQL uses Kysely to generate valid SQL statements from GraphQL queries. When I started out, Igal personally helped me to make the query builder generic.
- Michael Hayes - He created Pothos GraphQL, the foundation of our GraphQL server and its code-first schema generation approach.
