vanna

Introduction to Vanna

Vanna stands as an innovative open-source Python framework for retrieval-augmented generation (RAG), designed specifically for generating SQL queries and related functions. Under the MIT license, Vanna offers a flexible tool for developers and data enthusiasts to seamlessly interact with their databases through simple, natural-language questions.

How Vanna Works

Using Vanna involves a straightforward two-step process that doesn't require deep technical knowledge of RAG:

1. Training a RAG Model on Your Data: The first step is to train a model using your dataset. This model will store essential metadata necessary for generating SQL queries.

2. Asking Questions: Once trained, users can pose questions regarding their data, and Vanna will return SQL queries ready to be executed against the relevant database.

The beauty of Vanna lies in its simplicity. Users do not have to understand the complex mechanics of RAG to benefit from its capabilities.

User Interfaces

Vanna supports various user interfaces, providing flexibility in how it can be integrated into different workflows:

• Jupyter Notebook: Detailed documentation is available to integrate Vanna with Jupyter for exploratory data analysis.
• Streamlit and Flask: Vanna can be set up with these popular Python web frameworks to build custom user interfaces rapidly.
• Slack Integration: For those who prefer collaborative tools, Vanna supports a Slack interface, allowing users to communicate with their databases via Slack channels.

Getting Started

Installing Vanna is straightforward using pip:

pip install vanna

Additional documentation is available to assist users in configuring Vanna with their specific needs, such as using different Large Language Models (LLMs) or vector databases.

Import and Configuration

For developers who wish to customize their setup, such as integrating with OpenAI and ChromaDB, Vanna provides an example import configuration:

from vanna.openai.openai_chat import OpenAI_Chat
from vanna.chromadb.chromadb_vector import ChromaDB_VectorStore

class MyVanna(ChromaDB_VectorStore, OpenAI_Chat):
    def __init__(self, config=None):
        ChromaDB_VectorStore.__init__(self, config=config)
        OpenAI_Chat.__init__(self, config=config)

vn = MyVanna(config={'api_key': 'sk-...', 'model': 'gpt-4-...'})

Training Vanna

Training Vanna involves several options, each suited to different needs:

• DDL Statements: Train the model using Data Definition Language statements that include details about your database tables and columns.
• Documentation: Incorporate business terminologies and definitions directly into the training for context-specific enhancements.
• Existing SQL: Provide existing SQL queries to enrich the training dataset, enabling Vanna to better understand complex query structures.

Querying Your Database

Once trained, Vanna empowers users to ask complex questions in natural language, automating the creation of SQL queries. For instance:

vn.ask("What are the top 10 customers by sales?")

This returns a SQL query, which, if connected to your database, can directly fetch the results. Additionally, Vanna can visualize the output with tools like Plotly.

Vanna's Advantages

RAG vs. Fine-Tuning

Vanna utilizes the RAG approach, offering several advantages over traditional fine-tuning methods:

• Portability: RAG models are adaptable across various LLMs.
• Cost-Effectiveness: Cheaper than fine-tuning in terms of running costs.
• Flexibility: Training data can be easily updated or removed.

Key Benefits

1. High Accuracy: Vanna excels with complex datasets due to its training-based architecture, improving accuracy with more training data.
2. Security: Your data remains private as SQL execution occurs locally, without transmitting sensitive database content over the network.
3. Self-Learning Capability: Vanna can auto-train on successfully executed queries, continuously improving its performance.

By focusing on ease of use, accuracy, and security, Vanna provides a powerful tool for developers seeking an efficient way to interact with databases through natural language, without the need for deep technical intricacies. Whether used directly or as part of a customized interface, Vanna simplifies the process of data querying and analysis.