azurelane_search_agent

This project is a multi-agent information retrieval pipeline for the Azur Lane game Wiki, built with AutoGen AgentChat. After a user enters a natural language query, the system coordinates multiple agents through a SelectorGroupChat dynamic selector: summarize_agent parses the request and generates tool call parameters, search_agent calls the external tool get_blhx_wiki_data to fetch data from the Azur Lane Wiki, extract_agent refines the fields the user cares about, and user_proxy automatically reviews the results. If the review fails, the process jumps back to extract_agent for re-refinement; if it passes, it outputs Terminated to end the task. The system has a two-layer termination safeguard: TextMentionTermination detects the normal end marker, and max_turns=3 limits the maximum number of conversation turns to prevent agent infinite loops. The project supports Function Calling integration with the Wiki crawler tool, runs fully automatically without manual intervention, collects conversation messages through the event stream, and automatically extracts the final refinement result as output. The runtime environment requires Python 3.13.15 or higher. Before use, clone the repository, create a virtual environment, install dependencies, and configure MODEL_NAME, LLM_BASE_URL, and LLM_API_KEY in the .env file. To start, enter the src directory and run python azurelane_search_agent.py. The README provides query examples, such as asking "How many pulls are guaranteed for Shinano?", and the system returns structured information including cumulative construction count in the event pool and exchangeable count.