[matchagaucho]

As context window sizes increase and token prices go down, it makes more sense to inject dynamic memories into context (and use RAG/vector stores for knowledge retrieval).