The main ideas for building a multi-agent model of intelligent modeling for diagnosis are offered. The peculiarities of actions in a certain system of medical environment are indicated. The function of a medical agent is defined, which determines the actions that can be used by a medical agent in response to any sequence of acts of perception of medical data and knowledge. The problem of development of the specification of the problem medical agent is considered. The direction of development of the program of the medical agent is shown. Consider the requirements for the creation of reflex medical agents that respond directly to acts of perception. The requirements for the design of target-based agents are also considered.