TRL Enables Finetuning Models Directly on Various Agent Traces, Standardizing Stack

Core summary

One sentence to understand this update

TRL now supports finetuning models directly on various agent traces, including Claude Code, Codex, OpenClaw, and Pi traces, paving the way for a more standardized finetuning stack.

Impact & opportunity

What this could mean

Builders developing AI agents can now more easily refine their models using real-world agent interaction data, accelerating performance improvements and model adaptation.