Researchers propose a shared memory framework for multi-agent vision-language model systems to facilitate collaboration and efficient information exchange across agents.