What it does
Vanna 2.0 is presented as a SQL agent that is open source, multi-database, multi-turn and access controlled. The aim is to let people across an organization query their company's systems of record in natural language without needing SQL skills. You can install it with pip, try a dummy database in Colab with no sign-in, or choose the cloud or self-host route. The site says Vanna works with any LLM provider and database. The LLM providers shown are Anthropic, OpenAI, Gemini and Ollama, and the databases shown include SQLite, PostgreSQL, MySQL, Snowflake and BigQuery, with others covered in the docs. A configuration section produces copy-paste code for the stack you pick. Optional hosted admin features cover access control, observability with metrics and tracing, agent memory in a hosted vector database, secure file storage, audit logs and long-term data retention. Pricing for these sits on a separate page.
Capabilities
- Natural-language questions over your own database
- Open-source, multi-turn SQL agent
- Works with Anthropic, OpenAI, Gemini and Ollama
- SQLite, PostgreSQL, MySQL, Snowflake and BigQuery listed
- Optional hosted access control, observability and audit logs
- Installs with pip, with cloud and self-host options
When to use it
- You want colleagues without SQL skills to query company data.
- You are building a data chat feature on the LLM provider of your choice.
- You prefer an open-source stack you can host yourself.
