sprix-sage-router

Why we included this project

Most agent frameworks stop at discovery: they tell you which agents exist, but leave the harder question of who should actually work on a task once execution is underway. SAGE Router is a decision layer on top of the Agent2Agent protocol that answers that question at runtime, choosing between letting the current agent continue, recruiting a small set of collaborators, or handing the whole task to a specialist. The choice is computed from concrete state rather than a heuristic: completed DAG nodes, reusable artifacts, observed partial quality, remaining budget, and deadline all feed into a single auditable objective. Teams building multi-agent systems that need to recover from failures or re-plan mid-task will find the checkpoint-aware rerouting and requirement-conditioned trust tracking useful, and the reference implementation has no runtime dependencies, so it is easy to prototype against. It is a research preview rather than a turnkey product, but permission-first filtering, bounded candidate search, and state persistence make it a solid starting point for production routing logic.