Concepts

Steerability structures steering methods, termed controls in the toolkit, into four categories: input, structural, state, and output. To learn more about these categories, and what determines the category of a control, please see the conceptual guide on controls.

The toolkit additionally allows controls from different categories to be composed into a single operation on the model. These composed controls are referred to as steering pipelines. For a conceptual outline of what steering pipelines are, please see the guide on pipelines.

Alongside steering, the toolkit reads model internals for detection through calibrated probes, which drive conditional steering and routed decoding. For the conceptual overview of detection, please see the guide on probes.

Steered pipelines can also be written down as portable .spipe bundles that contain both the configuration and the products of an expensive steer step. For the conceptual overview of saving, sharing, and loading pipelines, please see the guide on sharing pipelines.