Current enterprise data analysis faces challenges such as high barriers to using traditional business intelligence tools, complexity in unified processing of multi-source heterogeneous data, and semantic gaps between user queries and business terminology. To address this, designs a general-purpose conversational multi-source heterogeneous data analysis system based on large language models. The system employs pooling and embedding storage technologies to achieve unified management of heterogeneous data sources, builds an intelligent conversational report generation framework based on Multi-Agent architecture, optimizes retrieval-based question answering capabilities through self-correction mechanisms, and implements system reliability design based on LangGraph Directed Cyclic Graph. This effectively lowers data analysis barriers, improves analysis efficiency, and provides enterprises with convenient data analysis and visualization capabilities. The system shows broad application prospects in areas such as enterprise business decision-making, financial investment, healthcare, and educational analysis.