Checkpointing is a commonly used approach to provide fault-tolerance and improve system dependability. However, using a constant and pre-configured checkpointing frequency may compromise other QoS criteria when there are resource and timing constraints. The frequency of checkpointing should be based on the importance of corresponding entities? behaviors and should also be runtime adaptable. The paper presents a behavior-based coordination framework. In this framework, behavior-based adaptive checkpointing is treated as a coordination concern; different coordination constraints are applied based on the behaviors that the entities are currently engaged in. Our prototyping and empirical experiments have shown that we are able to achieve behavior-based coordination in a more modular way with limited performance overhead. The experiments also demonstrate that the framework scales well when the number of entities involved in the system increases.