Graph-Based Self-Healing Tool Routing for Cost-Efficient LLM Agents

Researchers developed Self-Healing Router, a fault-tolerant system for LLM agents that reduces control-plane LLM calls by 93% while maintaining correctness. The system uses graph-based routing with automatic recovery mechanisms, treating agent decisions as routing problems rather than reasoning tasks.

• →Self-Healing Router reduces LLM inference costs by 93% while maintaining the same correctness as traditional ReAct systems.
• →The system uses Dijkstra's algorithm for deterministic shortest-path routing through cost-weighted tool graphs.
• →Automatic failure recovery is achieved by reweighting failed tool edges without requiring LLM intervention.
• →The architecture provides binary observability with logged reroutes or explicit escalations instead of silent failures.
• →LLM usage is reserved only for cases where no feasible execution path exists in the graph.