Active constraint-based clustering iteratively incorporates user knowledge into the clustering process through pairwise constraints, specifying whether two elements should be linked (Must-Link) or separated (Cannot-Link). Existing active approaches often overlook the underlying data distribution, particularly across regions with varying scales, which can lead to uninformed exploration of the data space. We introduce a novel approach that employs a locally adaptive kernel, automatically adjusting its bandwidth based on local data density. This local adaptation identifies regions that deviate from uniform or Gaussian assumptions, prioritizing them for constraint queries and thus minimizing user effort while improving clustering quality. Unlike global kernels, which require extensive hyperparameter tuning incompatible with interactive learning, our method dynamically selects the appropriate local bandwidth, avoiding manual configuration. Experiments on 29 real-world datasets demonstrate that the proposed approach consistently outperforms state-of-the-art methods and global kernels in terms of query efficiency, clustering accuracy.