Uno-Orchestra: Parsimonious Agent Routing via Selective Delegation

TL;DR

Uno-Orchestra is a unified RL-based agent routing system that optimizes task decomposition and worker assignment, significantly improving accuracy and reducing costs in multi-agent LLM systems.

Contribution

It introduces a novel learned orchestration policy that jointly optimizes task decomposition and worker selection, outperforming baselines across diverse benchmarks.

Findings

Achieves 77.0% macro pass@1, 16% above the best baseline.

Reduces per-query cost by roughly an order of magnitude.

Demonstrates effectiveness across math, code, knowledge, and tool-use tasks.

Abstract

Large language model (LLM) multi-agent systems typically rely on rigid orchestration, committing either to flat per-query routing or to hand-engineered task decomposition, so decomposition depth, worker choice, and inference budget are not jointly optimized under one objective. We introduce Uno-Orchestra, a unified orchestration policy that selectively decomposes a task and dispatches each subtask to an admissible (model, primitive) pair, with both decisions learned together from curated RL trajectories grounded in real worker interactions. Against 22 baselines on a 13-benchmark suite spanning math, code, knowledge, long-context, and agentic tool-use, Uno-Orchestra reaches 77.0% macro pass@1, roughly 16% above the strongest workflow baseline, at roughly an order of magnitude lower per-query cost, advancing the accuracy-efficiency frontier of selective delegation.