Agent Lightning v1.0: Toward Harnessed Agentic RL

Summary

Modern agents operate within agent harnesses that manage tools, context, and control flow. The original Agent Lightning proposed a disaggregated architecture connecting arbitrary agents to RL training via an LLM endpoint proxy—a paradigm termed harnessed agentic RL. v1.0 is a lightweight framework of about 3,500 lines of code, supporting arbitrary harnesses and offering a reproducible pipeline for coding-agent RL. With only 6K training examples and modest compute, it improves Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4%, a 14.6-point absolute gain.