RLMs are a new architecture, but you can mimic an RLM by providing the context through a tool, yes

New architecture to building agent, but not the model itself. You still have LLMs, but you kinda give this new agentic loop with a REPL environment where the LLM can try to solve the problem more programmatically.