ScriptsApr 8, 2026·2 min read

Vanna — AI SQL Agent That Learns Your Database

Open-source AI SQL agent that trains on your schema and queries. Vanna converts natural language to accurate SQL using RAG over your database documentation and history.

SC
Script Depot · Community
Quick Use

Use it first, then decide how deep to go

This block should tell both the user and the agent what to copy, install, and apply first.

pip install vanna
from vanna.openai import OpenAI_Chat
from vanna.chromadb 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-4o"})

# Train on your schema
vn.train(ddl="CREATE TABLE orders (id INT, customer_id INT, total DECIMAL, created_at DATETIME)")

# Ask questions in natural language
sql = vn.generate_sql("How many orders were placed last month?")
print(sql)  # SELECT COUNT(*) FROM orders WHERE created_at >= DATE_SUB(NOW(), INTERVAL 1 MONTH)

What is Vanna?

Vanna is an open-source AI SQL agent that learns your specific database. It uses RAG (Retrieval-Augmented Generation) to train on your DDL schemas, documentation, and example queries — then generates accurate SQL from natural language questions. Unlike generic text-to-SQL tools, Vanna improves over time as you add more training data specific to your database.

Answer-Ready: Vanna is an AI SQL agent using RAG to generate accurate SQL from natural language. Trains on your schema, docs, and query history. Supports PostgreSQL, MySQL, BigQuery, Snowflake, and 15+ databases. Built-in Streamlit UI for non-technical users. 13k+ GitHub stars.

Best for: Data teams wanting natural language database access. Works with: Any SQL database, OpenAI, Anthropic Claude, Ollama. Setup time: Under 5 minutes.

Core Features

1. Train on Your Data

# Train on DDL
vn.train(ddl="CREATE TABLE users (id INT, name VARCHAR, email VARCHAR)")

# Train on documentation
vn.train(documentation="The users table stores all registered users. The email field is unique.")

# Train on example queries
vn.train(sql="SELECT name, COUNT(*) as order_count FROM users JOIN orders ON users.id = orders.customer_id GROUP BY name")

2. Multi-Database Support

Database Connector
PostgreSQL psycopg2
MySQL pymysql
BigQuery google-cloud-bigquery
Snowflake snowflake-connector
DuckDB duckdb
SQLite sqlite3
Databricks databricks-sql

3. Built-in UI

from vanna.flask import VannaFlaskApp
app = VannaFlaskApp(vn)
app.run()
# Opens web UI at localhost:8084

4. Auto-Visualization

# Generate SQL + run it + create chart
vn.ask("Show monthly revenue trend as a line chart")
# Returns: SQL query, DataFrame result, and Plotly chart

FAQ

Q: How accurate is it? A: Accuracy depends on training data quality. With good schema docs and example queries, 85-95% accuracy on typical business questions.

Q: Does it support Claude? A: Yes, use vanna.anthropic.Anthropic_Chat as the LLM backend with any Anthropic model.

Q: Is my data safe? A: Vanna sends schema metadata and questions to the LLM, not raw data. Use local models via Ollama for full privacy.

🙏

Source & Thanks

Created by Vanna AI. Licensed under MIT.

vanna-ai/vanna — 13k+ stars

Discussion

Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.

Related Assets

Docusaurus — Build AI Tool Documentation Sites

Meta's documentation framework used by major AI projects. Docusaurus creates fast, SEO-friendly docs sites with MDX, versioning, i18n, and Algolia search built in.

Script Depot
◈Home🔍Search👤Me