leefly-cn/tripstar-java

TripStar is a Java based travel planning system that uses multiple AI agents working together to build a full trip itinerary. It is built with Spring Boot 4 and Spring AI Alibaba, and it has a separate frontend application that talks to this backend over an API. The system offers two ways to plan a trip. In the first mode, you give it a destination, dates, transport preference, and lodging preference in plain language, and a chain of agents runs in sequence: one searches Xiaohongshu, a popular Chinese social platform, for real travel notes about the destination, another reads those notes for attractions and tips, another checks those attractions against AMap, a Chinese mapping service, another checks the weather, another looks up hotels and restaurants, and a final planner agent builds a day by day itinerary with a budget, which a review agent then checks for inconsistencies. The second mode lets a user paste in links to existing Xiaohongshu travel notes, or the note text itself. It reads the note's text and images, uses a multimodal AI model to pull out day by day details, routes, hotels, and restaurants shown in the images, and keeps every attraction from the original note rather than trimming the list. It fills in missing location details through AMap and only calls in extra agents for hotels or food if the note did not mention them. If an attraction fails to make it into the final plan, the system reports the missing names and stops rather than silently dropping them. Behind the scenes, the agent workflow is controlled with a tool called StateGraph from Spring AI Alibaba, which manages the order and branching between agent stages, while a separate ReactAgent component handles tool calls within each stage. Prompts, tool calls, and raw model output are written to trace files for inspection, and progress updates are pushed to the frontend in real time over a WebSocket connection. To run it you need JDK 21, Maven, Node.js for the frontend, MySQL, Redis, an API key for Alibaba's DashScope AI service, and a key for AMap's web service. The search based planning mode also needs a valid login cookie for Xiaohongshu. The project avoids returning fake placeholder data: if a required key, cookie, or tool call is missing or fails, it reports an error instead. The project is licensed under the GNU General Public License version 2.