This seminar presents a general overview of multi-agent, derivative-free metaheuristic methods for global nonconvex optimization, with a focus on stochastic interacting particle systems. We briefly recall swarm-based approaches such as Particle Swarm Optimization (PSO), where a population of agents evolves through simple interaction rules inspired by collective behavior, and we then focus on Consensus-Based Optimization (CBO), a more recent and analytically tractable framework in which particles interact through a global consensus point defined via a Laplace-weighted average of their positions. We review recent theoretical results on CBO, including strong convergence of time-discrete schemes to global minimizers, with explicit rates depending on the number of particles and the time step. The analysis is based on a stochastic differential equation interpretation of the algorithm. A key aspect is the role of structure in the objective function. In particular, for additively separable problems, anisotropic CBO can automatically exploit the decomposition into low-dimensional components, leading to convergence results and computational complexity that depend on the intrinsic dimension rather than the full ambient space, thus mitigating the curse of dimensionality. Finally, we present an alternative macroscopic viewpoint on CBO, where instead of focusing on individual particle trajectories, one studies the evolution of statistical quantities such as the mean and variance of the particle distribution.