A conversational component is a part of the conversation where an agent (virtual/real) takes over the conversation in order to accomplish one or more goals. Upon accomplishing the goal, the agent (virtual/real) relinquishes control of the conversation back to the calling agent. Each component is defined by the goal or goals it is attempting to solve.
Conversational components are architected to be modular
A component may be called at any point in the conversation so it needs to be ready to assume control of an existing conversation or, alternately, initiate a conversation. Every component needs to decide how to handle the conversation history it inherits - if any.
Terminating a component
Each component can signal the calling agent to stop calling it at any point. Usually it will send the signal along with information about the component's stated goal: 1. component_done: The Component accomplished the stated goal 2. component_failed: The Component didn't accomplish the stated goal
Going out of context
Sometimes when a user is talking to a component, he/she will ask questions which are outside the domain of the component (Out of Context) There are several options for this scenarion: one way to handle it is to relinquish control back to the calling agent for one turn (to answer the non-contextual question) and then resume control immediately after providing the relevant response. In CoCo this signal is called "out_of_context"