RAG already gives you persistent memory outside of the context window for LLMs. An LLM can generate a list of steps to carry out a plan. You can then stuff th

RAG already gives you persistent memory outside of the context window for LLMs. An LLM can generate a list of steps to carry out a plan. You can then stuff that in a vector database.