Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

OpenForgeRL: Train Harness-native Agents in Any Environment

Source

arXiv

arxiv.org/abs/2607.21557v1

Read the source

Source

Plain-Language Brief

OpenForgeRL is an open-source framework for training AI agents using various inference harnesses like Claude Code, Codex, OpenClaw, etc., in diverse environments. The framework allows end-to-end training with standard RL codebases, enabling research and improvement of agents directly in the real environments they are deployed in. The paper presents results from testing OpenForgeRL across different harnesses and environments, showing improved performance compared to open baselines on most benchmarks.

What the Document Says

  • OpenForgeRL is an open-source framework for training AI agents using various inference harnesses.
  • The framework allows end-to-end training with standard RL codebases and a Kubernetes orchestrator.
  • It can train agents on any harness in any environment at scale.
  • Results from testing OpenForgeRL are presented across different harnesses and environments.
  • Improved performance compared to open baselines on most benchmarks is shown.

Key Concepts Mentioned

  • OpenForgeRL
  • AI agents
  • inference harnesses
  • standard RL codebase
  • Kubernetes
  • end-to-end training
  • performance improvement

Claims Made by the Document

  • OpenForgeRL allows end-to-end training of AI agents with diverse harnesses and environments.
  • The framework improves agent performance compared to open baselines on most benchmarks.

Evidence Used

  • Modern AI agents rely on elaborate inference harnesses...
  • OpenForgeRL achieves this with a lightweight proxy...
  • We validate our framework across diverse, complex harnesses and environments...
  • Using only hundreds to a few thousand tasks, OpenForgeClaw reaches 31.7 pass^3 and 55.9 pass@3 on ClawEval and 33.7 on QwenClawBench...

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Modern AI agents rely on elaborate inference harnesses such as Claude Code, Codex, and OpenClaw to drive multi-turn reasoning, tool use, and access to external systems. While powerful, these complex harnesses also make agents hard to train end-to-end with open infrastructure, whose SFT/RL stacks cannot natively express stateful, multi-process harness inference. To address this, we present OpenForgeRL, an open-source framework for training harness-based agents end-to-end in diverse environments. OpenForgeRL achieves this with a lightweight proxy that serves the harness's model calls while recording them as training data for a standard RL codebase (e.g., veRL), and a Kubernetes orchestrator that runs each rollout in its own remote container, together enabling training on any harness in any environment at scale. By decoupling training and inference, OpenForgeRL allows researchers to easily train, study, and improve agents directly in the real harnesses and environments they are deployed with. We validate our framework across diverse, complex harnesses and environments, spanning tool/claw-based agents and multimodal GUI browser- and computer-use agents. Using only hundreds to a few thousand tasks, OpenForgeClaw reaches 31.7 pass^3 and 55.9 pass@3 on ClawEval and 33.7 on QwenClawBench. OpenForgeGUI reaches 37.7 on OSWorld-Verified, 63.0 on Online-Mind2Web, and 72.3 on WebVoyager. Both outperform open baselines of similar size on nearly all benchmarks, and in the GUI setting match or surpass models several times larger. Beyond benchmarks, we analyze how harness choice (e.g., ZeroClaw, OpenClaw, Codex) and RL shape agent behavior. We find that some harnesses are substantially harder to learn than others, and that RL improves agentic reliability, such as self-verification, tool coverage, and completing multi-step plans, though critical abilities such as error recovery remain weak.