source

A source node is where bytes on disk become a Rime table. Keep this node boring: name the file, load it, and let downstream nodes do cleanup or interpretation.

That separation makes reports easier to read. A raw CSV source can stay out of the report while the first meaningful transform gets the review attention.

Use it at the edge

Use source when the project starts from a local CSV, JSON, NDJSON, or Parquet file. It has no parents and usually sits at the top of the DAG.

If a SQL query should read a file directly with DuckDB, use a kind: sql language node instead. That path is often better for large Parquet or SQL-first ingestion.

Source contract

Field	Required	Notes
id	yes	Source binding key. Runtime source overrides and editor file bindings are keyed by this id.
kind	yes	Always source.
path	run-time required	Project-relative CSV, JSON, NDJSON, or Parquet path. The editor may hold an unfinished source without a path, but a run needs one or a runtime source override.
metadata.report	no	Often false for raw files so reports start at the first meaningful transform.

What to inspect

• The path is project-relative and can be replaced at run time with --source <id>=<file>.
• Parquet preserves types best. CSV and JSON inference are convenient, but worth checking in the editor preview.
• Set metadata.report: false for noisy raw inputs when the report should begin at a cleaned or joined table.

Small example

- id: patients
  kind: source
  path: data/patients.csv         # project-relative; under dataDir/ by convention

Related

• SQL language nodes - use DuckDB when ingestion is query-shaped
• Dataset scanning - how the editor previews loaded tables