Recursive Language Models (RLMs) offload context by storing it in an external memory, like a file system, within the code environment