Decentralised computing offers more flexible solutions for application management compared to centralised approaches, particularly in environments affected by network bottlenecks, latency, or real-time processing demands. In decentralised systems, application management is delegated to local orchestrators, each responsible for a subset of resources. In this paper, we address orchestration in the Edge–Cloud Continuum by dynamically selecting optimal groups of physical resources to host application components. These groups, which are formed via clustering based on the physical and logical proximity of the computing nodes, are managed by dedicated orchestrators. This design mitigates the limitations of single-orchestrator architectures, which can become performance bottlenecks in large-scale systems. To enable dynamic, proximity-aware cluster formation, we propose two nature-inspired clustering algorithms within a simulation framework built on the DISSECT-CF-Fog simulator. The framework supports both cluster formation and decentralised application management, including deployment, scaling, scheduling, and monitoring. We evaluated our approach using a European-wide case study. The results show that decentralised clustering reduces average execution time and communication overhead across multiple applications compared to baseline strategies, primarily by considering network-aware node grouping. However, this improvement comes at the cost of an increased coordination effort, as reflected in a higher number of control messages.