In this paper, we address the building of ambient systems as autonomous and context-aware intelligent agents. The original contribu-tion is an algebraic language, namely AgLOTOS, used to automatically build the plan of an agent from its intentions. As plans are formally conceived as a structured set of concurrent processes, we show how to define a guidance service, helping the agent to maximize the satisfaction of its intentions. The underlying structure, called Contextual Planning System (CPS), takes the contextual information into account to predict the ability to execute the processes in plans. The last part of the paper talks about our current experiment integrating the proposed technique to assist user in a smart campus university.