ZenML Open-Sources Kitaru for Replay-Based Agent Evaluation

ZenML has open-sourced Kitaru (Apache 2.0, self-hostable), a replay-based evaluation framework that turns production agent traces into a continuous evaluation loop, enabling counterfactual experiments like swapping models without re-invoking live tools.

2026-08-18 ~ 2026-08-18 · 2 related posts

• ZenML's open-source Kitaru turns production agent traces into replay-based eval loops — strickvl · 2026-08-18
• ZenML open-sources Kitaru: replay production agent traces against different models or prompts — htahir1 · 2026-08-18