The system allows for users to either create a new agent from scratch, or load an agent previously created and archived. An archived agent records the internal state of the agent at the time of archival, including the list of articles it had retrieved for the user.
To load or create an agent, the user selects the ``New Agent'' menu-item in the ``Agent'' menubar of the Main Window. This pops up a dialog box which asks for the name of the agent and the directory in which to look for it. If an existing agent which meets this requirement is found, it is loaded, else a new agent is created. Before adding a new agent, an icon editor is presented to allow the user to create a personalized visual representation of the agent (the icon editor is a standard unix tool called bitmap). A new icon is added to the Main Window, a color code is assigned to the agent and the 3 icons for feedback and questions.
Once a new agent is added to the system, it is identical to the other agents in most respects. However, personalizing or training an agent assumes special significance in the case of new agents. This is because a new agent may not have any articles to present to the user for which it could receive feedback. The user will have to rely on other means to train the agent.
To compensate for the inability to provide direct feedback to new agents, the user can either demonstrate interesting articles to the agent, or manually edit the internal state of the agent. Programming by demonstration is particularly useful while training a new agent. By showing examples of interesting articles, the user is indirectly programming the agent to get ``good'' articles in the future. The other option to personalize new agents is to actually edit the internal state of the agent.