duckle

Duckle moves data between systems and does it on hardware you control. Pull from Postgres, land it in Parquet, transform it with SQL or Python, push it to a warehouse: the same job people currently pay Fivetran or Airbyte per row to run. It compiles to SQL on DuckDB and uses every core on the box, so a bigger instance is a faster pipeline. The project reports 96 million rows out of Postgres to Parquet in 39.9 seconds. Dual licensed MIT or Apache 2.0, and there is no vendor cloud at all.

Every pipeline is one file in git, which is the design decision that matters most. Author it on a visual canvas, in SQL, or in Python, then ship that file to your own server. `duckle-runner serve` runs it headless on a schedule in Docker or on a plain box, with a web console, roles, and an audit trail. Around 385 components cover the usual sources and destinations, plus dbt, CDC, data quality, reverse ETL, and lineage.

Nothing is paywalled. No per-row billing, no per-connector fee, no seat count, no account, no telemetry. Solo and small teams get the whole thing. Larger teams get it too, and pay only for the compute they were going to buy anyway. Compare against Airbyte (airbytehq/airbyte) if you want a bigger connector catalog, or dlt (dlt-hub/dlt) if you would rather write pipelines as plain Python.

The catch is that it is still labelled beta and it is young. A single-file-in-git pipeline is a great story right up until you need a connector that is not in the 385, and then you are writing it yourself against a project whose API has not settled.