Gloria is an implementation of a logic-based model of an agent, as described by Bob Kowalski in his new book (here to see Kowalski's Book in html format), (there is also a version in Spanish in here). The main component of Gloria is a Prolog program implementing the Fung and Kowalski's if and only if proof procedure, to process agent's goals and subgoals, and produce plans with actions for the agent to execute.
We are working on an extension of Gloria, to integrate a form of learning of rules, so that agents may learn to adapt its knowledge to new circunstances in their environment.
In English, an Agent is someone who does something for somebody else. This is very close to the latin root of the word (Agere: to do), except that now it does not have to be someone, but something.
In the modern history of technology, the word was recycled at the beginning of the 1990s, to emphasize the need for Artificial Intelligence systems to do actually do something. Rodney Brooks, at MIT, was one of the champions of the idea of getting rid of representations altogether, if we wanted real intelligent behaviour, which is no only clever, but also timely and opportune.
Gloria is part of an effort to respond to the first requirement (systems that actually do something, not just think), without having to embrace the second. You may find more details of this effort here.
In Kowalski's logical theory, an agent is an object, which has an internal stated, but also is permanently engaged in interactions with its environment. That agent observes the world (collecting "percepts", basically propositions describing its sourroundings and events in the neighbourhood) and acts upon it (posting actions to its environment to be executed).
Internally, an agent can be characterized by, essentially, two types of data structures:
Goals: which contains descriptions of the agent's general intentions, namely those establishing its permanent relationships with the environment, also called, maintenance goals or implications, and those, derived from the later, establishing task to be reduced to plans of actions, also called achievement goals.
Beliefs: which contains the agent's rules of behaviour.