One-click mounting to a new project + global effective updates, this experience is way better than managing JSON configurations myself

LangSmith releases an evaluator template library and reusable evaluators, simplifying multi-level evaluation work for AI agents. The templates cover five major categories: safety and protection, response quality, execution trajectories, user behavior analysis, and multimodal evaluation, and include optimized judging prompts and rule-based evaluators. They are suitable for online monitoring and offline experiments. Reusable evaluators are centrally managed at the organizational level, with a new Evaluators tab, one-click mounting to new projects, and prompts that take effect globally. The library is open-sourced and released with openevals v0.2.0, adding multimodal support.