DuckLess

Run DuckDB on a big VM in your own GCP project, only for the time of the job. No cluster, no HMAC keys, nothing left running.

uv tool install duckless
duckless init --project my-project --region europe-west1
duckless run daily_revenue.sql --machine n2-highmem-32 --spot

Your project, your data

Jobs run on Compute Engine through Cloud Batch, in your project and region. Data never leaves it, IAM stays yours.

GCS through ADC

The runner reads and writes gs:// with the VM’s service account. No HMAC keys, no JSON key files.

Big machines, short lives

Pick an n2-highmem-32 for five minutes, with local SSD for spill and Spot pricing. The VM is deleted when the job ends.

SQL or your own code

Run a .sql file, a Python script with a ready-to-use DuckDB connection, or your own image and command (dbt, …).

Files, BigQuery, tables

Read Parquet, CSV, JSON and Excel on GCS, read and write BigQuery tables. Keep tables that change in DuckLake, with snapshots and time travel.

Runnable examples