The increasing adoption of Agentic AI systems is reshaping how complex tasks are addressed across a wide range of domains. In many emerging applications—such as collaborative information retrieval, distributed data analysis, simulation-based policy modeling, and automated scientific discovery—multiple agents interact, exchange intermediate outputs, and iteratively refine collective decisions over extended interaction cycles. At the same time, deploying these systems in practice requires distributed learning and coordination across heterogeneous sites, edge devices, and organizational boundaries.
While multi-agent systems demonstrate remarkable capabilities, their widespread deployment introduces two deeply intertwined research challenges. On the one hand, fundamental questions arise concerning how knowledge is acquired, shared, validated, and updated through interaction. On the other hand, real-world environments rarely permit centralized training or global data sharing: learning must occur across distributed sites under privacy, regulatory, and communication constraints. At the intersection of these two dimensions, a central and still underexplored question emerges: how can multi-agent systems discover and remain grounded in reliable knowledge while learning and coordinating in a fully distributed manner?
AGENSYS jointly addresses knowledge-grounded multi-agent learning and distributed intelligence, treating them as complementary facets of the same overarching problem. Agents rely on external knowledge sources such as structured databases, knowledge graphs, retrieval systems, scientific corpora, or streaming data while collaborating or competing in dynamic environments. Simultaneously, the underlying training and adaptation processes must be decentralized, communication-efficient, and resilient to data heterogeneity.
