Vanna AI
: How to use it, features, and the business problems it solves

How to Use

1. Install

   Add Vanna to your Python environment via pip. It integrates with Jupyter, Flask (FastAPI), Streamlit, Slack, and more depending on your use case.

2. Choose an LLM and vector store

   Configure the LLM you want to use — OpenAI, Anthropic, Google Gemini, or Ollama (local models) — along with a vector store for storing training data.

3. Train the model

   Feed it your database DDL (table definitions), business documentation, and question-SQL pairs. This training data becomes the retrieval corpus for RAG, improving SQL generation accuracy.

4. Query in natural language

   Ask questions like "What were the top 10 products by revenue last month?" Vanna performs Agentic Retrieval to find relevant context and generates the corresponding SQL.

5. Execute and visualize

   Run the generated SQL against your database and retrieve results as tables, charts, or summaries. An embedded web chat component is available for interactive use.

Features

• Natural language to SQL (Text-to-SQL)

  Generates accurate SQL for your target database from natural language questions. RAG (Retrieval-Augmented Generation) grounds generation in your trained schema and business knowledge.

• RAG training (DDL, documents, SQL examples)

  Train on table definitions (DDL), business documentation, and question-SQL pairs to optimize generation for your own data.

• Multi-database support

  Works with PostgreSQL, MySQL, Snowflake, BigQuery, Redshift, SQLite, Oracle, SQL Server, DuckDB, ClickHouse, and other major databases.

• Multi-LLM and vector store support

  Swap in OpenAI, Anthropic, Google Gemini, Azure, AWS Bedrock, Mistral, or Ollama (local), and choose from a range of vector store backends.

• Multi-turn conversation and visualization

  Supports multi-turn dialogue and returns results as tables, charts, or summaries. An embedded web chat component is included.

• Access control and auditing (higher tiers)

  Row-level security (per-user queries), audit logs, and rate limiting for production deployments (Cloud/Enterprise).

Pricing

Pricing as of June 2026. The core framework is MIT-licensed open source and free to self-host (LLM API costs apply separately). Cloud/Enterprise pricing varies — check the official site for the latest.

※ The OSS framework itself is free, but LLM API costs (e.g., OpenAI) apply separately. Note that the official GitHub repository was archived (read-only) on March 29, 2026. Check the official site for the current availability before use.

Pros & Cons

Pros

• Query internal data in natural language — accessible even to team members unfamiliar with SQL
• MIT-licensed OSS; self-hosting is free and integrates into your own environment
• Freely combine any major DB, LLM, or vector store
• Training on DDL and business knowledge optimizes generation for your specific data

Cons

• Setup and operation require basic knowledge of Python and RAG
• SQL accuracy depends on training data quality; validation and guardrails are essential
• UI and documentation are primarily in English
• The official repository is archived; check the official site for future update plans

Reviews & Reputation

• Engineers and data teams praise it: "Once trained on DDL and business knowledge, it generates accurate SQL tailored to our data."
• "The flexibility to self-host on OSS and combine with your preferred LLM and DB is a real advantage."
• Common caveats: "English-centric," "always validate generated SQL," and "the quality of training data makes or breaks it."

FAQ

Q. Is Vanna AI free to use?

The core framework is MIT-licensed open source and free to self-host. However, LLM API costs (e.g., OpenAI) apply separately. A Cloud/Enterprise version with operational features is also available.

Q. Which databases are supported?

PostgreSQL, MySQL, Snowflake, BigQuery, Redshift, SQLite, Oracle, SQL Server, DuckDB, ClickHouse, and other major databases.

Q. Can I use it without knowing SQL?

Yes. You ask questions in natural language and get back SQL plus results (tables, charts, summaries), so team members unfamiliar with SQL can still analyze internal data. Validating generated SQL is recommended for critical analyses.

Q. Is Japanese supported?

The UI and documentation are primarily in English. A Japanese UI is not provided.

Vanna AI vs. Other Development & Data Tools

For more detailed comparisons, see the pages for related services.