Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

TL;DR

This paper introduces the ROME model and the Agentic Learning Ecosystem (ALE), a comprehensive open-source framework for developing and optimizing agentic large language models capable of complex, multi-turn real-world interactions.

Contribution

The paper presents ALE, a new end-to-end ecosystem for agentic model development, and introduces ROME, a novel agent trained with a new policy optimization algorithm for improved long-horizon performance.

Findings

ROME outperforms existing benchmarks like SWE-bench Verified.

ALE streamlines agent development with its modular components.

IPA improves training stability over long interaction sequences.

Abstract

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.