Immersive VR tools simulate cosmic events interactively to improve understanding of space science, but they often prioritize visuals over learning. MAIVE navigates virtual environments with multiple AI agents and embedded tutors on demand. A central coordinator guides these tutors in explaining concepts, checking assumptions, and providing step-by-step support. By aligning motion data with activity milestones, the system captures signals that show how each user progresses through tasks. Conversation history is preserved by a memory and retrieval tool to support teachers. Separating perception tasks from teaching moves and internal model updates allows adjustments without changing virtual environments. To measure impact, we will compare an adaptive MAIVE condition to a nonadaptive condition using the same material, order, and logging. We will analyze conceptual change, user feedback scores, pause, help, and error logs.