This paper presents a domain specific programming language dedicated to timely bounded agents. Timely bounded agents are expected to deliver the most appropriate response to a stimuli and this response has to be delivered at the right time, otherwise it is ignored. This modifies assumptions and action models used in most of current agent programming languages and frameworks. We propose a domain specific language by specifying an agent abstract machine and its operational semantics. The dynamics of the agent abstract machine is a sequence of perception-deliberation-influence cycles where an agent: (i) gets its perceptions (that remain unchanged during the cycle) (ii) evaluates its behaviors to generate a set of influences; (iii) commits all influences at once within its environment. An important property is the fact that perceptions are unchanged and side effects are not allowed until the end of a cycle. This property makes possible efficient implementation of the agent abstract machine using parallel evaluations. Simplicity and expressiveness are important features when working on operation semantics. To meet these requirements we have proposed a simple and yet expressive framework inspired from REST (Representation State Transfer) to consider agent actions as manipulation of resources within environments using only a limited set of primitives. This makes both the agent abstract machine and its operational semantics easy to comprehend and the implementation straightforward. Finally, we demonstrate how our proposal has helped building a serious game for upper limb rehabilitation.