Magic-BI is an AI-based fully automated ChatBI product that currently supports SQL databases. Magic-BI is open-source and can be deployed in a fully private or semi-private manner, which maximizes user privacy protection while lowering the usage threshold.
| Magic-BI | Other Chat BI Products | |
|---|---|---|
| Data Security | Magic-BI can operate entirely using a closed-source model, preventing internal data or metadata leakage. | Some competitors use closed-source |
| large models, which can lead to data and metadata leakage from business databases, introducing commercial risks. | ||
| Build Cost | Magic-BI can fine-tune the model with zero or minimal user input training data, significantly reducing usage costs. | Users or vendors need to build training |
| data (such as question-SQL pairs) based on the business domain and database structure, typically in the hundreds or thousands, requiring substantial human and time resources. | ||
| Ease of Use | Magic-BI provides complete functionality for training data generation, model fine-tuning, and deployment, allowing users to perform these tasks with simple page operations, significantly lowering the product usage | |
| threshold. | Professional technical personnel are required to build training data, select base models, fine-tune models, and deploy them, which is time-consuming, costly, and uncertain. |
|
| Data Governance Required | Magic-BI works directly on business databases without requiring data governance or data warehouse construction, offering the following advantages:
1. No restrictions on question types;
2. Avoids the
need for data warehouse construction;
3. Suitable for production deployment. | Some competitors use closed-source large models to improve accuracy. However, in real business systems, the accuracy of general-purpose large models is
often insufficient for production use. |
Magic-BI currently supports two operating modes: Fine-Tuned Model Mode and Agent Mode.
This mode is implemented using a general-purpose large model + RAG (Retrieval-Augmented Generation) + agent approach. The advantage is low usage threshold, but the disadvantage is lower accuracy and higher inference cost. For beginners, it is recommended to use open-source large models of 70B+ or other large model cloud services.
This mode is implemented using a fine-tuned large model + agent approach. The advantage is high accuracy and low inference cost, but the disadvantage is that it requires generating training data and fine-tuning the model, which slightly increases the usage threshold.
Magic-BI can run on Ubuntu 22.04, RTX 4090, CUDA 12+, and PyTorch 2.0+. Other environments have not been rigorously tested, but similar environments should also work. If you encounter any issues while using Magic-BI, you can contact us through the following contact information.
Magic-BI supports three installation methods: pip installation, Docker deployment, and source code compilation. We recommend using the Docker deployment method.
Install the PostgreSQL development library:
sudo apt-get update
sudo apt-get install libpq-dev
sudo yum install postgresql-devel
brew install postgresql
Execute the following command to complete the installation of Magic-BI:
pip install magic-bi
To deploy using Docker, you need to install Docker on your system and enable GPU support for Docker. The specific operations are not covered here. Execute the following two commands to start Magic-BI and its related dependencies:
cd $Magic-BI;docker compose -f deployment/docker-compose.yml up -d.
Enter the Magic-BI directory and execute the command pip3 install -e .
Execute the following two commands to start the dependencies for Magic-BI:
cd $Magic-BI;docker compose -f deployment/docker-compose.yml up -d.docker compose -f deployment/docker-compose-component.yml up -d.
Magic-BI supports two types of clients: WEB GUI and Restful API. If you want to use Magic-BI directly, access it through the web interface. If you want to integrate Magic-BI into your system or use specific functionalities, access it via the Restful API.
Execute the command python3 -m magic_bi.main --config config/system.yml to start the system. Enter http://127.0.0.1:6688 in a web browser (currently supported browsers are Chrome and Firefox) to access the system.
Execute the command python3 -m magic_bi.main --config config/system.yml to start the system. Use an API tool or another system to call Magic-BI.