Abstract
Large language models (LLM) agents are capable of taking actions and performing useful tasks. Complex tasks need collaboration between multiple such agents to accomplish a goal. Existing multi-agent communication frameworks are deficient in supporting such collaboration since these frameworks fail to incorporate critical aspects that would enable effective interactions between agents. This disclosure describes an inter-agent communication framework to improve communication, decision making, and collaboration between LLM agents. In the framework, new embedding paradigms centered around communicative facets can be utilized. The embeddings can enhance communication efficacy and also serve as a foundation for training multi-agent systems to accomplish their objectives effectively. A communication router is utilized that has awareness of inter-agent communication structure and strategies. The router can append a communication embedding to incoming tokens from an LLM and determine outgoing actions for one or more other LLMs that are part of the task workflow.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
"Communication Embeddings and Router to Support LLM-based Multi-agent Systems", Technical Disclosure Commons, (October 03, 2024)
https://www.tdcommons.org/dpubs_series/7395