Inference
The process of running a trained model on new input to produce an output, as distinct from training the model in the first place.
In more detail
Inference is the runtime phase of a model's life. Training happens once and is expensive; inference happens every time the model is used and is where ongoing cost and latency accumulate. Every classification, draft, or decision a workflow asks a model to make is an inference call.
Because inference cost scales with usage, the economics of an automated process depend heavily on how many model calls it makes and how large each one is. Reducing unnecessary calls, keeping inputs tight, and using smaller models where they suffice are the main levers for keeping automation affordable at volume.
Where this shows up at Ceven
Ceven manages inference at the step level so a workflow only spends model time where it adds value. Deterministic steps call connectors, not models; judgment steps call a model sized for the task. The result is that a workflow you describe in plain language runs at a sensible cost even when it processes high volume.