DeepYardDeepYard
O

OpenForgeRL

End-to-end RL training framework for stateful, multi-process agent harnesses

Open SourceFree

About

Open-source framework designed for training agents that use complex inference harnesses (Claude Code, Codex, OpenClaw) with supervised fine-tuning and reinforcement learning. Bridges the gap in traditional ML stacks by natively supporting stateful, multi-process harness inference patterns. Ideal for researchers developing autonomous coding agents and multi-step reasoning systems that require environment interaction during training.

Details

Language
Patterns

Tags

frameworkopen-sourcecoding-agentautonomousclaude-codepython