# Command-line interface

Atom-HiFi installs a single `atom-hifi` console script
({py:func}`atom_hifi.cli.main`).

```bash
atom-hifi init [config.yaml]     # write a commented starter config
atom-hifi run  config.yaml       # run the full L/R selection workflow
```

`atom-hifi run` merges the given config under the template defaults and calls
{py:func}`atom_hifi.runner.run`, which returns a summary dict (metrics + output
paths). The tutorial wrapper (`hifi_workflow_tutorial.py`) and the tests use the
same single orchestration path.

## Entry points

```{eval-rst}
.. autofunction:: atom_hifi.cli.main
   :no-index:

.. autofunction:: atom_hifi.cli.build_parser
   :no-index:
```

## Workflow stages

The runner executes, in order: library load → descriptor compute → optional
ε_noise calibration → optional active-learning loop or piecing → final L/R scan
→ optional NSGA-II c-factor optimisation → final fit and representative-set
selection. The Python API also supports expert-only ND grid scan and patient/OOD
controls that are intentionally absent from YAML. See
{py:func}`atom_hifi.runner.run` for the full control flow.

```{eval-rst}
.. autofunction:: atom_hifi.runner.run
   :no-index:
```