RSIHub: A Composable Framework for Agent-Evolution Methods
Zimu Wang, Siyu Ye, Xiaodong Gu, Zilong Zheng and colleagues (ByteDance, Shanghai Jiao Tong University, BIGAI, UC Berkeley) release RSIHub, an open-source framework that runs agent-evolution methods on shared infrastructure so they can be compared under fixed conditions.
Ask this paper
Design. Recipes compose reusable operators and declare which resources may change; the framework supplies execution, evaluation, recovery and lineage, and keeps the evaluator outside the mutable surface.
Coverage. AHE, Hyperagents, A-Evolve and GEPA are implemented without modifying the framework core.
Results. With MiniSWE and Codex on Terminal Bench 2 and tau-3 Banking, the largest observed pass@1 gain is 13.5 points, but method rankings change across settings.
Transfer. Optimization gains do not always carry over to held-out tasks, which is the main reason the authors argue for a shared protocol with sealed splits.
Abstract
Agent-evolution methods use execution feedback to improve prompts, skills, tools, and orchestration, but their implementations often rebuild the same experimental machinery. We present RSIHub, an open-source framework for building, comparing, and extending these methods on shared infrastructure. Recipes compose reusable operators and declare mutable resources; RSIHub supplies execution, evaluation, recovery, and lineage. Researchers can focus on evolution policy while evaluation stays outside the mutable surface and results remain tied to candidates and evaluation conditions. We instantiate AHE, Hyperagents, A-Evolve, and GEPA without modifying the framework core, keeping method-specific logic in recipes and operators. We evaluate them with MiniSWE and Codex on Terminal Bench 2 and $τ^3$ Banking under fixed model and evaluator configurations. The largest observed overall pass@1 gain is 13.5 percentage points, while method rankings vary across settings and optimization gains do not always transfer to held-out tasks. RSIHub provides a common protocol for comparing evolution policies, inspectable candidate histories, and reusable components for developing new methods.