Sprix AI releases SAGE Router prototype for stateful multi-agent task routing

Sprix AI has publicly released SAGE Router, a research prototype designed to decide whether an AI agent should continue a task alone, collaborate with others, or hand the task off entirely. Unlike conventional agent-discovery layers, SAGE treats routing as a scheduling problem that accounts for work already completed, context-transfer costs, and unresolved task dependencies. The system sits on top of the existing Agent2Agent protocol and returns routing decisions without executing task transmissions itself. SAGE evaluates candidates using a utility function that weighs success probability against penalties for latency, cost, coordination overhead, and context loss. A key feature is progress-aware replanning, which allows the router to distinguish tasks that are still cheap to reassign from those where the current agent has built up critical accumulated context.