zhongyu09/openchatbi

What it solves

OpenChatBI is a chat-based intelligent Business Intelligence (BI) tool that allows users to query, analyze, and visualize data from databases using natural language. It eliminates the need for users to write complex SQL queries manually by automating the conversion of natural language questions into executable SQL statements.

How it works

Built on the LangGraph and LangChain ecosystem, the project uses a multi-agent architecture to orchestrate data tasks. A main agent handles general interaction, while a specialized Data Analysis Agent manages complex tasks like time series forecasting and anomaly detection. The system employs a Text2SQL pipeline with schema linking and prompt engineering to generate queries, which are then executed against a data warehouse. It includes a data catalog management system that uses either vector-based or BM25-based retrieval to index database structures and business explanations.

Who it’s for

It is designed for business users and data analysts who want to interact with their data warehouses (such as Presto, PostgreSQL, or MySQL) through a conversational interface without needing deep technical expertise in SQL.

Highlights

• Natural Language to SQL: Automatically converts user questions into SQL queries with built-in error recovery and a confidence gate for quality control.
• Advanced Analytics: Includes specialized capabilities for time series forecasting, anomaly detection, and root-cause drill-down.
• Flexible Data Catalog: Supports automatic discovery of database tables and flexible storage backends for metadata.
• Integrated Visualization: Generates intuitive data visualizations using Plotly.
• Code Execution: Features a sandboxed Python execution environment for custom data analysis.
• Observability & Memory: Provides audit logs, LLM tracing via Langfuse, and a "learned SQL memory" that captures high-confidence query patterns for future use.