Vanna — Chat with Your SQL Database Using AI

What it is

Vanna is an AI-powered text-to-SQL tool. You ask a question in plain English, and Vanna generates the correct SQL query, runs it against your database, and returns the results with optional visualizations. It learns your schema, table relationships, and query patterns to improve accuracy over time.

Vanna targets data analysts, business users, and developers who need quick answers from databases without writing SQL manually.

How it saves time or tokens

Vanna's training step lets it memorize your schema, column meanings, and example queries. Once trained, it generates accurate SQL on the first attempt rather than requiring iterative prompt engineering. This reduces both human time (no SQL writing) and token cost (fewer retries).

The visualization layer means you get charts directly from natural language without exporting to a BI tool.

How to use

1. Install Vanna: pip install vanna
2. Connect to your database (PostgreSQL, MySQL, SQLite, BigQuery, Snowflake)
3. Train Vanna on your schema with DDL statements and example queries
4. Ask questions in natural language and get SQL + results

Example

import vanna
from vanna.remote import VannaDefault

vn = VannaDefault(model='my-model', api_key='...')
vn.connect_to_postgres(
    host='localhost',
    dbname='analytics',
    user='reader',
    password='secret'
)

# Train on your schema
vn.train(ddl='CREATE TABLE orders (id INT, customer_id INT, total DECIMAL, created_at TIMESTAMP)')
vn.train(question='Total revenue last month', sql='SELECT SUM(total) FROM orders WHERE created_at >= DATE_TRUNC(\'month\', NOW()) - INTERVAL \'1 month\'')

# Ask questions
sql = vn.generate_sql('What are the top 10 customers by revenue?')
df = vn.run_sql(sql)
vn.generate_plotly_chart(df)

Related on TokRepo

• Database tools -- Database management and query tools
• RAG tools -- Retrieval-augmented generation pipelines

Common pitfalls

• Vanna accuracy depends heavily on training quality; provide representative DDL and at least 10-20 example question-SQL pairs
• Complex joins across many tables may produce incorrect queries; verify generated SQL before running on production data
• The default model uses Vanna's cloud API; for privacy, configure a local LLM backend