MeshHeal improves reliability of decentralized AI agent networks
MeshHeal: Two-Timescale Self-Healing for Gray Failures in Decentralized LLM Agent Networks
Artificial IntelligenceComputation and LanguageDistributed, Parallel, and Cluster Computing
Summary
Sometimes AI agents working together can still respond but do worse on their tasks without anyone noticing. The authors created MeshHeal to spot these 'gray failures' by having agents review each other's work quickly and over time to catch ongoing problems. This system helps isolate agents that are performing poorly until they recover, then lets them join back in. Their experiments showed MeshHeal works better and uses fewer resources than previous methods.
What this means in practice
- •For distributed system engineers: Build multi-agent AI systems that detect and isolate agents with degraded performance to maintain task quality without centralized control.
- •For cloud service operators: Improve reliability and efficiency of decentralized AI services by automatically managing agent health and task routing under partial failures.$Commercial implications: Enables commercial deployment of robust decentralized AI networks with automated fault management to reduce downtime and improve service quality.
Authors
Keru Chen, Sen Lin, Yingbin Liang, Nathaniel D. Bastian, Shaofeng Zou
Abstract
Decentralized LLM-based multi-agent systems coordinate through local interactions, but an agent can remain responsive while its task-solving quality persistently degrades. Such gray failures require protecting current tasks before sufficient evidence exists to alter future routing, while still allowing recovered agents to rejoin. We introduce MeshHeal, a fully decentralized self-healing framework that couples ability-matched peer review across two timescales. At the fast timescale, an adaptive hierarchy escalates uncertain or low-scoring outputs from repeated single-reviewer evaluation to committee deliberation and, when needed, correction before use. At the slow timescale, a task- and ability-conditioned peer-relative detector aggregates scores to distinguish persistent degradation from ordinary output variation, trigger mandatory committee review, and eventually exclude degraded agents from ordinary routing; recovery probes provide fresh evidence for reintegration. To faithfully evaluate routing, we introduce Model-Backed MAS Evaluation, which ties ability assignments to execution models, since prompt-based ability assignments alone can leave routing errors hidden. Across BBH, MATH, and MMLU-Pro, MeshHeal achieves 0.839 degraded-phase accuracy using 51k total model tokens per task, versus the strongest baseline Symphony's 0.807 accuracy using 115k per task. Under staggered degradation and recovery, MeshHeal isolates degraded agents, keeps them excluded from ordinary task execution until recovery, and returns them to normal routing.