Grid is a technology that implements the process of sharing resources in a flexible, secure and coordinated manner. Task management in computational grids involves planning, implementation and monitoring. The main contribution of this work consists in the development of a model with an agent-based architecture for managing computer resources each with defined operations so that the user can perform tasks efficiently and effectively and thus improve substantially the management by a gLite Grid middleware. The solution proposed provides a platform based on a collection of agents in a virtual organization. The model considers the heterogeneity of resources so that it is completely independent of any physical network architecture. This paper focus on the model, simulation and evaluation of an agent-based management of computational resources in grid environment architecture. Experimental results showed significantly the effectiveness of algorithms and planning policies to achieve load balancing, fault monitoring, and service quality. The computational complexity of the proposed model is studied and the experimental results are analyzed with respect to the use of the computing resources.