OpenForgeRL enables end-to-end training for harness-based agents
July 22, 2026
OpenForgeRL provides an open-source framework for training agents that utilize complex inference harnesses like Claude Code. It uses a lightweight proxy to record harness model calls as training data and a Kubernetes orchestrator to manage remote container rollouts.
HOW THIS AFFECTS YOU
●
builderYou can now train agents in stateful, multi-process environments using standard RL stacks like veRL.
●
researcherThis allows for end-to-end evaluation and training of agents within realistic tool-use ecosystems.