Reliability redundancy optimization is intended to design a near-optimal reliable system with optimal active redundancy for a complex system with non-repairable components and minimal costs. This study introduces a surrogate-assisted metaheuristic optimization framework for addressing reliability redundancy allocation problems in complex systems, leveraging a Levy Flight-based Grey Wolf Optimizer integrated with an adaptive kernel selection mechanism for co-Kriging surrogate modelling. The proposed algorithm enhances the exploration capabilities of the Grey Wolf Optimizer by incorporating Levy flight dynamics, enabling efficient navigation through high-dimensional search spaces and mitigating premature convergence. A dynamic adaptive kernel selection mechanism is also introduced to identify the most suitable kernel functions for co-Kriging, thereby improving the accuracy and efficiency of fitness landscape approximations. An advanced, penalty-based constraint handling mechanism manages trade-offs between reliability, redundancy, cost and other constraints. Comparative evaluations reveal that the proposed algorithm achieves the best reliability of 0.9742 in series and 0.9992 in series-parallel configurations, outperforming others, respectively. It also demonstrates the highest mean reliability, 0.9703 in series, 0.9987 in series-parallel, with the lowest standard deviations, indicating robust convergence. Although it incurs higher computational costs, its robust performance makes it ideal for applications prioritizing reliability.