What is Vanna
Vanna is an open-source Python RAG (Retrieval-Augmented Generation) framework that helps users generate precise SQL queries for their databases based on large language models (LLMs). Vanna operates with a simple two-step process: first, train the RAG model on user data, then generate SQL queries that can be executed on the database based on questions. Vanna supports multiple LLMs, vector databases, and SQL databases, providing highly accurate queries while ensuring database content security and privacy, preventing leakage. Vanna supports custom frontend interfaces, and supports users to self-learn through feedback to improve query accuracy.
Vanna's Main Features
- SQL Query Generation: Automatically generate SQL queries based on user questions using large language models (LLMs).
- Retrieval Augmentation: Combine retrieval-augmented technology (RAG) with generative models to improve query accuracy.
- Multi-Database Support: Support connecting to multiple SQL databases, such as PostgreSQL, MySQL, etc.
- Multi-LLM Support: Compatible with multiple LLMs, including OpenAI, Anthropic, etc.
- Vector Storage Integration: Integrate with multiple vector databases, such as AzureSearch, PgVector, etc.
- User Interface: Provide multiple user interface options, including Jupyter Notebook, Streamlit, Flask, and Slack.
Vanna's Technical Principles
- Retrieval-Augmented Generation (RAG): Based on RAG technology, a method that combines retrieval (retrieving relevant information from large amounts of data) and generation (generating output based on retrieved information). It enables Vanna to understand and respond to user questions more accurately.
- Large Language Models (LLMs): Vanna relies on LLMs to understand and generate natural language. The models are pre-trained and can understand the complexity and context of language.
- Vector Databases: Vanna uses vector databases to store and retrieve information related to user questions. The databases can efficiently process and compare large amounts of data points.
- Training and Fine-Tuning: Users train Vanna based on provided DDL statements, documents, and SQL queries, enabling it to better understand and respond to business-specific queries.
- Automated SQL Execution: SQL queries generated by Vanna can be automatically executed on the user's database without manual intervention.
Vanna's Project Address
- Project Official Website：vanna.ai/docs
- GitHub Repository：https://github.com/vanna-ai/vanna
Vanna's Application Scenarios
- Data Analysts: Data analysts quickly generate complex SQL queries to analyze large amounts of data without manually writing each query.
- Business Intelligence (BI) Tools: Integrate Vanna into BI tools, supporting users to ask questions in natural language to obtain data insights, enhancing data accessibility and analysis capabilities.
- Customer Support Systems: Based on integrating Vanna, customer support systems automatically retrieve relevant information from databases, quickly responding to customer queries and issues.
- Automated Reporting: Enterprises automatically generate daily or periodic reports, reducing the workload of manual data processing and report writing.
- Data Science Projects: Data scientists quickly prototype and test new data analysis ideas, using natural language interaction to accelerate the data exploration process.