We present time-constrained automata (TCA), a model for hard real-time computation in which agents behaviors are modeled by automata and constrained by time intervals. TCA actions can have multiple start time and deadlines, can be aperiodic, and are selected dynamically following a graph, the time-constrained automaton. This allows expressing much more precise time constraints than classical periodic or sporadic model, while preserving the ease of scheduling and analysis. We provide some properties of this model as well as their scheduling semantics. We show that TCA can be automatically derived from source-code, and optimally scheduled on single processors using a variant of EDF. We explain how time constraints can be used to guarantee communication determinism by construction, and to study when possible agent interactions happen.