This report proposes Harness-Native agentic routing, moving model selection from single-turn serving to step-level decisions conditioned on the full agent harness state. The router can select one best-fit model for cost-efficient execution or combine complementary models for higher accuracy. In OpenSquilla, the authors instantiate a four-layer routing stack, an open LightGBM cold-start ranker, and a staged router-model path that turns arena logs into stronger routing policies. The report evaluates singleton and multi-model routing on DRACO and PinchBench, framing execution traces as a feedback loop for both routing and agent-native model training.
No heat snapshots are available in the last 24 hours.