Symphony-Coord: Emergent Coordination in Decentralized Agent Systems

Multi-agent large language model systems can tackle complex multi-step tasks by decomposing work and coordinating specialized behaviors. However, current coordination mechanisms typically rely on statically assigned roles and centralized controllers. As agent pools and task distributions evolve, these design choices lead to inefficient routing, poor adaptability, and fragile fault recovery capabilities. We introduce Symphony-Coord, a decentralized multi-agent framework that transforms agent selection into an online multi-armed bandit problem, enabling roles to emerge organically through interaction. The framework employs a two-stage dynamic beacon protocol: (i) a lightweight candidate screening mechanism to limit communication and computational overhead; (ii) an adaptive LinUCB selector that routes subtasks based on context features derived from task requirements and agent states, continuously optimized through delayed end-to-end feedback. Under standard linear realizability assumptions, we provide sublinear regret bounds, indicating the system converges toward near-optimal allocation schemes. Validation through simulation experiments and real-world large language model benchmarks demonstrates that Symphony-Coord not only enhances task routing efficiency but also exhibits robust self-healing capabilities in scenarios involving distribution shifts and agent failures, achieving a scalable coordination mechanism without predefined roles.

💡 Research Summary

Symphony‑Coord tackles a fundamental bottleneck in large‑language‑model (LLM) based multi‑agent systems: the reliance on static role assignments or a central controller for routing sub‑tasks. Both approaches suffer from scalability issues, poor adaptability to shifting task distributions, and fragile fault‑tolerance. The authors reconceptualize agent selection as an online contextual bandit problem, where each available executor is treated as an arm and the observable context combines task semantics with dynamic agent state (load, latency, reliability, availability, cost).

The framework introduces a two‑stage “dynamic beacon” routing protocol. In Stage 1, a lightweight scoring function aggregates (i) a similarity score between the sub‑task description and the agent’s declared capabilities (computed via embedding cosine similarity or lexical overlap), (ii) a smoothed prior success rate, and (iii) a recent reliability metric. The top‑L agents according to this composite score form a candidate set Cₜ, dramatically reducing communication and inference overhead.

Stage 2 applies a LinUCB algorithm within Cₜ. For each candidate j, a context vector x_{j,t} =