OpenForgeRL is an open-source framework for end-to-end training of agents built around stateful inference harnesses such as Claude Code, Codex, and OpenClaw. A lightweight proxy serves model calls and records trajectories for standard RL systems such as veRL, while a Kubernetes orchestrator runs each rollout in an isolated remote container. The abstract reports strong results from hundreds to a few thousand tasks: OpenForgeClaw reaches 31.7 pass^3 and 55.9 pass@3 on ClawEval, while OpenForgeGUI scores 37.7 on OSWorld-Verified, 63.0 on Online-Mind2Web, and 72.3 on WebVoyager. Error recovery remains weak.
No heat snapshots are available in the last 24 hours.