In this work, we introduce a multiagent architecture called the MultiAGent Metaheuristic Architecture (MAGMA) conceived as a conceptual and practical framework for metaheuristic algorithms. Metaheuristics can be seen as the result of the interaction among different kinds of agents: The basic architecture contains three levels, each hosting one or more agents. Level-0 agents build solutions, level-1 agents improve solutions, and level-2 agents provide the high level strategy. In this framework, classical metaheuristic algorithms can be smoothly accommodated and extended. The basic three level architecture can be enhanced with the introduction of a fourth level of agents (level-3 agents) coordinating lower level agents. With this additional level, MAGMA can also describe, in a uniform way, cooperative search and, in general, any combination of metaheuristics. We describe the entire architecture, the structure of agents in each level in terms of tuples, and the structure of their coordination as a labeled transition system. We propose this perspective with the aim to achieve a better and clearer understanding of metaheuristics, obtain hybrid algorithms, suggest guidelines for a software engineering-oriented implementation and for didactic purposes. Some specializations of the general architecture will be provided in order to show that existing metaheuristics [e.g., greedy randomized adaptive procedure (GRASP), ant colony optimization (ACO), iterated local search (ILS), memetic algorithms (MAs)] can be easily described in our framework. We describe cooperative search and large neighborhood search (LNS) in the proposed framework exploiting level-3 agents. We show also that a simple hybrid algorithm, called guided restart ILS, can be easily conceived as a combination of existing components in our framework.