TripChat: A Long-term Planning Agent Hierarchy for Travel

In this work, we introduce a zero-shot, unsupervised trip and travel planner. Planning a trip automatically has several use cases - it can save people time and money, help people create better itineraries, and provide inspiration. Our framework uses a hierarchical multi-agent approach to coordinate, gather real-world data using APIs, and plan out a travel plan. Existing AI-based approaches for travel planning are often too simplistic and inflexible, rely on outdated data, or are of poor quality and detail. Our results show that our output plans generated this way are comparable to the best human made plans, approaching similar levels of detail, experiences, and personalization.