Ask HN: Any interesting work in using multiple LLMs in concert?

I've been messing around myself trying to get multiple LLMs to work together to solve problems. I'm sure this must be an active area of research. What I've been trying: A "memory retrieval" LLM that looks at the input and decided what data to pull from long term memory A "processor" LLM that takes the input and combines it with the data from memory to produce an ouput A "memory store" LLM that looks at the input and output of the processor and decides what should be stored in long term memory. And finally a "supervisor" LLM that looks at the input and output and decides if "thinking" is complete or if everything should be fed back through the system again. Currently, I'm just messing around with ideas, but would like to see what's going on in this area.