This article presents a reservoir-computing based associative memory system for complex dynamical attractors, including both location-addressable and content-addressable retrieval scenarios. The system uses a reservoir computer (RC) to store and retrieve complex dynamical attractors, such as periodic and chaotic ones, by leveraging the network's ability to generate diverse dynamical trajectories. The RC is trained to store multiple attractors, each associated with a specific index value or cue signal, enabling retrieval through appropriate inputs. The system demonstrates that a single RC can memorize a large number of attractors, each retrievable with a specific index or cue. The study reveals the mechanisms behind successful and failed switching between stored attractors, and identifies scaling laws between the number of stored attractors and the size of the reservoir network. For content-addressable retrieval, the system exploits multistability, where multiple attractors coexist in the high-dimensional phase space of the reservoir network. The success rate of retrieval increases with the length of the cue signal, and the system exhibits natural random itinerancy when noise is present. The study provides foundational insights into developing long-term memory devices for complex dynamical patterns. The results show that the RC can effectively store and retrieve complex attractors, with high success rates and accurate memory recall. The study also explores different coding schemes for indexing attractors, finding that one-hot coding is the most efficient. The results demonstrate the potential of reservoir computing for associative memory, with applications in neuropsychology and complex dynamical systems.This article presents a reservoir-computing based associative memory system for complex dynamical attractors, including both location-addressable and content-addressable retrieval scenarios. The system uses a reservoir computer (RC) to store and retrieve complex dynamical attractors, such as periodic and chaotic ones, by leveraging the network's ability to generate diverse dynamical trajectories. The RC is trained to store multiple attractors, each associated with a specific index value or cue signal, enabling retrieval through appropriate inputs. The system demonstrates that a single RC can memorize a large number of attractors, each retrievable with a specific index or cue. The study reveals the mechanisms behind successful and failed switching between stored attractors, and identifies scaling laws between the number of stored attractors and the size of the reservoir network. For content-addressable retrieval, the system exploits multistability, where multiple attractors coexist in the high-dimensional phase space of the reservoir network. The success rate of retrieval increases with the length of the cue signal, and the system exhibits natural random itinerancy when noise is present. The study provides foundational insights into developing long-term memory devices for complex dynamical patterns. The results show that the RC can effectively store and retrieve complex attractors, with high success rates and accurate memory recall. The study also explores different coding schemes for indexing attractors, finding that one-hot coding is the most efficient. The results demonstrate the potential of reservoir computing for associative memory, with applications in neuropsychology and complex dynamical systems.