Advancing the Science of Multi-Agent Intelligence.
AG2 is built on years of foundational research in multi-agent orchestration, tool use, and AI reasoning. Our team and collaborators publish in top conferences to advance the state of AI systems that think and work together.
From Research to Real-World Systems
AG2's technology is grounded in peer-reviewed research exploring how agents can reason, collaborate, and self-improve.
We publish actively across key areas such as orchestration frameworks, self-improving agents, tool-use optimization, document intelligence, and benchmarking.
Our open approach means every insight informs our open-source ecosystem — bridging research and production.
AutoGen: Enabling next-gen LLM applications via multi-agent conversation framework
COLM 2024
Absolute Zero: Reinforced self-play reasoning with zero data
Ranked #1 on Hugging Face (May 7, 2025)
A Survey of Self-Evolving Agents
Ranked #2 on Hugging Face (July 29, 2025)
Research by Topic
Explore AG2's research organized by topic
Collaborate on Open Research
AG2 bridges academic and open-source innovation. We welcome collaboration with universities, labs, and enterprises exploring the frontier of multi-agent intelligence. Join our research discussions, contribute evaluations, or co-author papers with the AG2 team.