How is cross-source SQL different from ETL?

ETL copies and transforms data into a central database or warehouse before you query it, so results can be stale and the pipeline needs maintenance. Cross-source SQL queries the sources directly and live, with no copy step — useful for ad-hoc analysis and joining data that never shared a database.

Is cross-source SQL the same as a federated query?

It's the same idea. Federated query engines like Trino or Presto run one query across many sources, but they are server systems that need infrastructure to deploy. Varan brings the same capability to a local-first desktop app built on DuckDB — you point it at your sources and query, with no cluster to run.

What tools can run cross-source SQL?

Server-side engines like Trino, Presto and Apache Drill, cloud query federation features, and local engines such as DuckDB. Varan is a desktop app that uses a local DuckDB engine to join PostgreSQL, MySQL, DuckDB and spreadsheet files in one query.

Varan Join the beta →

Varan › What is cross-source SQL

What is cross-source SQL?

Cross-source SQL is running one SQL query — including JOINs — across more than one data source at once: a PostgreSQL database, a MySQL database, a DuckDB file and your CSV or Excel files, all in the same statement.

Cross-source SQL: one query joining a PostgreSQL orders table to two CSV files and returning revenue by customer tier — Cross-source SQL in one line of work: a PostgreSQL table joined to two CSV files, live.

Why it exists

Real data is scattered. Orders live in PostgreSQL, the product catalog in MySQL, a segment list in a CSV someone emailed you. A normal SQL client connects to one database at a time, so a query in one connection can't see a table in another — let alone a file. Cross-source SQL removes that wall: point one engine at every source and write a normal query.

How it works

A cross-source engine registers each source as a table, reads only the rows a query needs, and computes the join in one place. With a local-first engine like DuckDB, that "one place" is your own machine, using read-only access — nothing is copied into a central database and nothing is written back.

SELECT c.name, s.tier, SUM(o.total) AS revenue
FROM   customers  c                     -- CSV
JOIN   pg__orders o ON o.customer_id = c.id   -- PostgreSQL
JOIN   segments   s ON s.customer_id = c.id   -- CSV
GROUP BY c.name, s.tier;

Cross-source SQL vs ETL

ETL copies and reshapes data into a warehouse before you can query it. It's the right tool for durable, governed reporting, but it's heavy to build, and the copy is stale the moment a source changes. Cross-source SQL queries the sources live with no copy step — ideal for exploration, one-off questions, and joining data that never shared a database.

Cross-source SQL vs federated queries

They're the same concept. Engines like Trino, Presto and Apache Drill run federated queries across many sources, but they're server systems you have to deploy and operate. A local-first tool brings the same capability to a desktop app: no cluster, no setup — you point it at your sources and query.

Where Varan fits

Varan is a desktop SQL workspace built on a local DuckDB engine. It runs cross-source SQL over PostgreSQL, MySQL, DuckDB and spreadsheet files, and adds automatic anomaly detection, column lineage and git-style rollback so you can trust and trace the result.

Run a cross-source query yourself

Free during the beta — macOS, Windows and Linux.

Request beta access →