AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation Framework
Abstract
This technical report presents AutoGen, a new framework that enables development of LLM applications using multiple agents that can converse with each other to solve tasks. AutoGen agents are customizable, conversable, and seamlessly allow human participation. They can operate in various modes that employ combinations of LLMs, human inputs, and tools. AutoGen's design offers multiple advantages: a) it gracefully navigates the strong but imperfect generation and reasoning abilities of these LLMs; b) it leverages human understanding and intelligence, while providing valuable automation through conversations between agents; c) it simplifies and unifies the implementation of complex LLM workflows as automated agent chats. We provide many diverse examples of how developers can easily use AutoGen to effectively solve tasks or build applications, ranging from coding, mathematics, operations research, entertainment, online decision-making, question answering, etc.
Community
This paper and framework seems like it has the potential to be useful in a plethora of use cases - shocked so few people are talking about it. :)
code: https://github.com/microsoft/FLAML
docs: https://microsoft.github.io/FLAML/docs/Use-Cases/Autogen
Introduces AutoGen: a framework for building multi-agent LLM applications (agents can converse and take multiple modes); conversational programming: declare agents that have specific capabilities and roles, and program interaction behavior between agents. Agent capabilities of LLM: wraps capabilities (role playing, feedback, coding, etc.) with prompting techniques such as result caching, error handling, etc. to create one agent; for Humans: for solicitation from humans; tool-backed agents for code or function execution (and group chat manager). Hard-coded conversation control flow (agents can execute python code); two types of conversation programming: automatic flows through auto-replies, and programmed fusion (language guided, programmed, or transition based). Applied to math problem solving with expert LLM or human in loop (base AutoGen LLMs better than GPT-4 and ChatGPT with Wolfram Alpha plugin), retrieval augmented chat/generation, decision making in text-world environments (ALFWorld - align text and embodied environments), multi-agent coding on OptiGuide, dynamic group chat, and conversational chess. Appendix has related work (AutoGPT and LangChain single-agent systems, BabyAGI, MataGPT, CAMEL multi-agent systems); usage guidelines and default message template; more details on the applications (workflows, methodology, results); more example outputs. From Microsoft, University of Washington.
Links: website, blog (research post), arxiv, GitHub
Is it worth using mutli-agents for RAG?
Models citing this paper 1
Datasets citing this paper 0
No dataset linking this paper