Magic Insight

Report Abuse

Basic Information

Magic-BI (Magic-Insight) is an open-source ChatBI product that provides conversational business intelligence over SQL databases and other mainstream data formats. It is designed to let users query business databases in natural language without building a data warehouse or heavy data governance. The project supports deployment in fully private or semi-private environments to maximize data privacy. Two runtime operating modes are provided: an Agent Mode that uses a general-purpose large model with retrieval-augmented generation (RAG) for lower setup effort, and a Fine-Tuned Model Mode that uses a domain-fine-tuned model for higher accuracy and lower inference cost. The repository includes runtime requirements, installation options (pip, Docker, source), and instructions to run a web GUI or a RESTful API for integrating the system into other applications.

Links

App Details

Features
Direct SQL database support enabling natural-language-to-SQL interactions without requiring a separate data warehouse. Two operating modes: Agent Mode using LLM+RAG for quick start and Fine-Tuned Model Mode for production accuracy via fine-tuning. Built-in tooling for generating training data and performing model fine-tuning to reduce manual labeling effort. Privacy-oriented deployment options including fully private or semi-private setups and the ability to use closed-source models to avoid internal data leakage. Multiple installation paths: pip, Docker compose deployment with GPU support recommended, and source installation. Provides a web GUI for end-user access and a RESTful API for integration. Configuration driven startup using a system.yml file and guidance on runtime environment (Ubuntu, CUDA, PyTorch).
Use Cases
Magic-BI lowers the barrier to delivering conversational analytics by automating the steps needed to turn natural-language questions into actionable SQL queries and BI insights. Its training-data generation and fine-tuning pipeline reduces the human cost of producing domain-specific question-SQL pairs, making it easier to reach production-grade accuracy. The Agent Mode lets beginners get started quickly with open-source or cloud LLMs, while Fine-Tuned Mode offers a path to lower inference costs and higher fidelity for business-critical queries. Deployment options including Docker and a RESTful API make it possible to integrate the system into existing platforms or run in isolated private environments to protect sensitive data. The web UI gives analysts and non-technical users a straightforward interface for exploratory and operational queries.

Please fill the required fields*