This paper introduces an Artificial Intelligence (AI) model of a virtual companion system on smartphone. The proposed AI model is composed of two modules of Probabilistic Mood Estimation (PME) and Behavior Network. The PME is designed for the purpose of automatic estimation of the mood, under uncertain and dynamic smartphone context. The model combines Support Vector Machine (SVM) and Dynamic Bayesian Networks (DBNs) to estimate the probabilistic mood state of the user. The behavior network contorts the behavior of the interactive and intelligent virtual companion, considering the detected mood and external factors. In order to make the virtual companion more believable, the system consists of an internal mood state structure. The mood of the agent, could also be inferred from another real human such as a remote partner. The fitness of the artificial companion behavior in relation to the users mood state was evaluated by user study and effectiveness of the system was confirmed.