Korean Multi-Turn AI Agent Trajectory Dataset

This dataset captures complete trajectories in which users and AI agents carry a single task through to completion over multi-turn Korean dialogue. Following the flow from the user's request to task completion, each trajectory records the reasoning the agent performed, the tools it called, and the execution results those tools returned as discrete steps. This makes it possible to inspect not only the final response but also the judgments and actions the agent took at each step.

Potential Use Cases

• Agentic Supervised Fine-Tuning: Complete trajectories let a model learn to generate the next reasoning step and tool call for Korean task-completion dialogue from the preceding turns.
• Process Reward Modeling: Because each step pairs a tool call with the result it returned, intermediate actions can be scored individually rather than only by whether the task finished.