sprix-sage-router: Routing to Reconfigure Running Agent Teams in the A2A Network

Agent discovery only tells you who exists; it does not determine who to work with during execution. Sprix SAGE Router is a decision layer built on top of the Agent2Agent (A2A) protocol to fill this gap. It makes real-time decisions on whether the current agent should continue alone, summon complementary peers, or hand off tasks based on expertise.

It compares three paths—SELF, COLLABORATE, and HANDOFF—within a single utility function. Instead of simply gathering agents with high reputations, it prioritizes how well they complement missing requirements (Complementarity). It also learns per-agent and per-requirement confidence levels to eliminate the bias that an agent strong in coding is also strong in research.

Tasks are decomposed into a DAG (Directed Acyclic Graph), and an optimal executor is assigned to each requirement. Dependency edges are converted into communication topologies, and critical path latency is estimated. Agents that violate constraints such as permissions, budget, or deadlines are excluded from ranking regardless of their predicted quality.

In a benchmark of 2,500 synthetic tasks, online SAGE achieved a quality of 0.634 and utility of 0.487 compared to existing strategies. Although cost and latency increased, it transparently reveals the trade-offs hidden behind a single score. A2A clients can receive these routing decisions and proceed with actual execution through message sending, streaming, task polling, and more.