[ model graph ]

Model Graph

The smallest primitives for representing a model-powered system as a graph of transforms.

A model-powered system is rarely one model. It is a chain of steps: read an input, call a model, apply a rule, check a result, act. Model Graph treats that chain as a graph of small transforms, each with an input and an output you can inspect.

The layer stays deliberately minimal. A transform runs an input to an output. A graph composes transforms. A trace records what happened. Signals, evaluations, and feedback annotate the run. Nothing domain-specific lives here.

primitivetransform graph

input  ->  transform  ->  transform  ->  output
              |             |
            trace         signal

The point is generality. Anything built on top, interpretability, reliability boundaries, an agent, should be expressible from these primitives without the core needing to know about any of them.

Composable Model Graph builds the product layer on top of this.