DYNO : Dynamic Neurosymbolic Orchestrator for Multi-Agent Systems

Ritvik Garimella, Chathurangi Shyalika, Renjith Prasad Kaippilly Mana, Amit Sheth

Keywords: Multi-Agent Systems, Large Language Models, Workflow Orchestration, Agents

TL;DR: We introduce DYNO (Dynamic Neurosymbolic Orchestrator), a system for managing workflows for multi-agent systems by combining symbolic planning with neural adaptability, enabling workflow refinement with semantic understanding.

Abstract: Large Language Model (LLM)-based multi-agent systems (LaMAS) represent an emerging paradigm for tackling complex, multi-step reasoning and decision-making problems. As these systems scale, orchestration, which is the ability to coordinate, manage, and evaluate the interactions among diverse agents, becomes central to their success. While recent orchestrators such as AgentFlow have demonstrated promise in managing communication and task delegation, they remain limited in their ability to understand task semantics, coordinate heterogeneous agent types (e.g., reactive vs. cognitive), and adaptively align outputs with human-defined goals. In this position paper, we introduce DYNO (Dynamic Neurosymbolic Orchestrator), a system developed as part of our broader research framework on neurosymbolic AI for robust, interpretable, and trustworthy composite intelligence. DYNO integrates interdependent components to plan, execute, evaluate, and refine workflows iteratively. Each component cooperates through shared registries of agents, data, knowledge, and evaluation metrics, allowing the system to optimize task performance continuously. We argue that such dynamic orchestration, combining symbolic decomposition with neural adaptability, is essential for achieving scalable, interpretable, and self-correcting multi-agent intelligence. The paper positions dynamic orchestration as a foundational step toward reliable, trustworthy, and human-aligned multi-agent systems.

Submission Number: 45