Chatbots have a long way to go to being more useful, and memory will certainly play a critical role alongside another compounding thing of how to actually use that memory in an efficient enough way that it can scale properly. If my hard drive were my computers memory and I were conversing with it, accessing certain things can be fast or slow depending on the types of indexing done, and processing that information in a way that's useful can also vary in speed, so a great deal of problem comes from balancing an AI in it's expanded capacity alongside it's diminishing performance for low-end devices. Some central server can be used for many of these, but in some cases it must be local and for those it should also meet certain benchmarks such as compression of it's memory for file size and optimization to save battery life.