Vanna AI

Vanna AI is an open-source text-to-SQL agent that lets users ask questions of company databases in plain English, built for developers who want to add natural-language data access to their own apps and internal tools. Rather than a finished BI product, it is a Python framework plus an embeddable chat component, so you control the database, the model and the permissions. The project is MIT-licensed and has more than 23,000 stars on GitHub.

What is Vanna AI used for?

Developers use Vanna to give colleagues or customers a chat box that answers data questions, without handing out raw database access.

• Turning plain-English questions into SQL and running them against your database
• Streaming results back as tables, Plotly charts and a short natural-language summary
• Passing user identity from cookies, JWTs or OAuth through to SQL filtering for row-level security
• Embedding a prebuilt chat component in React, Vue or plain HTML pages
• Adding audit logs, per-user rate limits and lifecycle hooks around queries
• Swapping between OpenAI, Anthropic, Gemini, Ollama, Bedrock and other models

Who Vanna AI is best for

Vanna fits Python developers building internal data assistants, SaaS companies adding natural-language analytics to their product, and data teams that want full control over which model and database are used. Because it can run with local models through Ollama, it also suits privacy-sensitive setups. Business users who want a ready-made BI tool with dashboards and no code should pick a finished analytics product instead.

Vanna AI pricing

As of October 2026, the open-source framework is free to self-host via pip under the MIT license. Hosted plans add admin features such as observability, agent memory, audit logs and data retention: Explorer costs $50 per month for teams of two or three with 20 questions a day and API access; Team costs $500 per month with 300 questions a day and setup and live support; Enterprise is custom with unlimited questions, on-premises support, SAML SSO and an admin API. Annual subscriptions get a 20% discount. Model and database costs are separate when you self-host.

Limitations to keep in mind

• Self-hosting needs Python skills and time to wire up auth, databases and models
• Answer quality depends on your schema, documentation and the model you choose
• Hosted plans cap the number of questions per day
• It is a framework, so dashboards and scheduled reports are up to you

Vanna AI vs alternatives

ThoughtSpot is a complete AI-powered BI platform with dashboards and governance, better for business teams that want a finished product. Julius AI is simpler for individuals analyzing spreadsheets and files. Hex suits data teams that want SQL notebooks and AI in a shared workspace. Databricks is the choice when your data already lives in a lakehouse and you want its built-in assistant. Vanna AI wins when developers want an open-source, embeddable agent with permissions they control.

Tips for getting started with Vanna AI

Install it with pip and run the quick start against a sample database before connecting real data. Wire up your user resolver early so row-level security is in place from the start. Use a read-only database role, and test common questions from your team to find gaps in schema descriptions. If you need a hosted admin layer, try Explorer before committing to a larger plan.

Vanna AI FAQ

Is Vanna AI free?

Yes, Vanna AI's framework is open source under the MIT license and free to self-host. Hosted plans start at $50 per month.

Which databases does Vanna AI support?

Vanna AI supports PostgreSQL, MySQL, Snowflake, BigQuery, Redshift, SQLite, Oracle, SQL Server, DuckDB, ClickHouse and others.

Can Vanna AI run with a local model?

Yes. Vanna AI works with Ollama for local models as well as OpenAI, Anthropic, Gemini, Azure, Bedrock and Mistral.

What is the best alternative to Vanna AI?

ThoughtSpot is a strong alternative for a finished AI analytics product, while Hex suits data teams that prefer notebooks.