Cache-to-Cache: Direct Semantic Communication Between LLMs to the Raptor engine

This is what you need to be able to communicate with each other in Neuralese. It's not looking good for monitorability.