Command Line Interface

Command-line helpers for validating configs, scaffolding templates, and inspecting a Datarax installation. Pipelines themselves are built in Python (see the DAG Construction Guide).

Commands

Command	Purpose	Example
datarax validate	Validate a pipeline config	datarax validate -c pipeline.toml
datarax create	Scaffold a config template	datarax create -o pipeline.toml -t basic
datarax list	List available components	datarax list --type sources
datarax benchmark	Show a calibrax store summary	datarax benchmark --dataset synthetic
datarax version	Print the installed version	datarax version

!!! note "Key points"

- Config files are TOML and are validated, not executed, by the CLI
- `datarax benchmark` delegates to [calibrax](https://github.com/avitai/calibrax)
  and prints a store summary; for comparative benchmarks use
  `uv run python -m benchmarks.cli run`
- Build and run pipelines with the Python API

Quick Start

# Scaffold a template, then validate it
datarax create --output pipeline.toml --template basic
datarax validate --config-path pipeline.toml

# List the registered source components
datarax list --type sources

# Print the installed version
datarax version

Modules

• main - Main CLI entry point and commands
• benchmark - python -m datarax.cli.benchmark timing tool

Config File Format

datarax create writes a TOML template describing the pipeline as a list of nodes and edges:

[pipeline]
name = "my_pipeline"

[[nodes]]
id = "source"
type = "DataSource"
class = "MemorySource"

[nodes.params]
num_samples = 1000
sample_shape = [28, 28, 1]

[[nodes]]
id = "batch"
type = "BatchNode"

[nodes.params]
batch_size = 32

[[edges]]
from = "source"
to = "batch"

datarax validate accepts this [[nodes]] layout as well as configs that declare a [dag] or [sources] section.

See Also

• Config - Configuration system
• Benchmarking - Programmatic benchmarking
• Installation - Installing the CLI