A run is a single execution of an agent within a thread. You create a run by specifying an agent type and an input prompt. The run progresses through a lifecycle and produces steps and artifacts that capture the agent’s work.
Each step represents an individual unit of work within a run. Steps can be nested, reflecting how an orchestrator delegates to specialists.
Each artifact is a structured result produced by a step. Artifacts include confidence indicators and source references that trace findings back to the data that informed them.
Each source identifies where an artifact’s data originated.