TRL and OpenEnv now let you train agents inside your own coding harness
ben_burtenshaw · x · 2026-07-24
You can now train agents in TRL and OpenEnv with a harness of your choice, such as OpenCode, pi, Codex, or Claude Code.
- The core idea is to improve the model in the same harness where it will actually be used, so it learns the right tool usage for the task.
- The workflow described is: build a task dataset, plug in a harness that drives the session, connect an OpenEnv environment plus harness adapter, then use TRL HarnessRolloutWorker to collect traces and AsyncGRPOTrainer to update weights from environment reward.
- The attached diagram shows the full cycle: code task + verifier → loop-owning agent → environment/verifier → rollout collection → policy updates.
Related event: TRL and OpenEnv Enable Custom Agent Harness Training(4 posts)→
More from coding & agent
- TRL adds Score Centering, a simple fix for RLHF training instability — lvwerra · 2026-10-27
- Precompute golden answers and match at runtime: an LLM pattern from AustinLangChain — colinmcnamara · 2026-10-27
- Precompute golden answers and match at runtime for exact wording control — colinmcnamara · 2026-10-27
- Dev argues there's still alpha in working with your coding agent interactively — alex_frantic · 2026-10-27
- Researchers launch AMA on patterns for safer AI agents used by citizen developers — Loose_Rip359 · 2026-10-27
- Gradio Adds In-Chat Human-in-the-Loop So ML Intern Pauses Before Training — Gradio · 2026-10-27