Skip to content
Rime

What is Rime?

Rime is a local workflow system for reproducible, polyglot data analysis. You declare a DAG of typed nodes, Rime runs the graph, caches every output, and can render the run as an HTML report.

The Short Version

Section titled “The Short Version”
IdeaMeaning
DAGA pipeline.dag.yaml file describes the graph and node configuration.
NodeA pure step over tabular data: read a source, transform rows, run SQL, call Python/R/JS, or compute a statistic.
RuntimeRime materializes inputs, executes the node, captures logs and plots, and persists outputs.
CacheA node reruns only when its definition, source code, inputs, or runtime version changes.
Reportrime build turns the run into outputs/run_report.html.

Most projects mix two node styles:

  • Core nodes for ordinary table work: filter, derive, aggregate, join, pivot, select, sort, concat, and built-in statistical tests.
  • Language nodes for custom work in SQL, Python, R, or JavaScript. You write the function or query; Rime owns dataframe handoff and artifact writing.

A Tiny Pipeline

Section titled “A Tiny Pipeline”
specification_version: "2.1"
nodes:
  - id: penguins
    kind: source
    path: data/penguins.csv


  - id: adelie_only
    kind: filter
    inputs: [penguins]
    expr: '[species] == "Adelie"'


  - id: by_island
    kind: aggregate
    inputs: [adelie_only]
    groupBy: ["[island]"]
    metrics:
      - "[mean_bill_length] = [bill_length_mm].mean()"
      - "[n] = [bill_length_mm].count()"

Run it with:

Terminal window
rime run pipeline.dag.yaml
rime build pipeline.dag.yaml

The first command writes Parquet/JSON artifacts under outputs/. The second also writes an HTML report.

What Rime is not

Section titled “What Rime is not”
  • Not an orchestrator. There is no scheduler, cloud runner, queue, or retry policy. Use Airflow, Prefect, or Dagster for that layer.
  • Not a notebook with hidden state. Nodes are pure functions over declared inputs. Downstream work depends on named outputs, not ambient cell state.
  • Not a warehouse transformation framework. SQL is first-class, but Rime is built for local analysis graphs that may cross several languages.
  • Not yet 1.0. APIs and file formats can still change.

Who it’s for

Section titled “Who it’s for”

Researchers, data journalists, analysts, and small data teams who want to:

  1. Mix SQL, Python, R, and JavaScript without writing file handoff glue.
  2. Re-run an analysis end to end with one command.
  3. Keep intermediate data, logs, figures, stats, and reports tied to the same DAG run.

Compared to

Section titled “Compared to”
  • dbt: SQL-only and warehouse-oriented. Rime borrows the “write the query, let the tool materialize it” model and extends it beyond SQL.
  • Quarto / R Markdown: document-first. Rime is pipeline-first; reports come from the executed graph.
  • Snakemake / Nextflow: file-oriented workflow engines. Rime is dataframe- and report-oriented, with built-in table/stat nodes.
  • Airflow / Prefect / Dagster: orchestration systems. Rime does not schedule or deploy jobs; it makes a local analytical graph reproducible.

Next

Section titled “Next”