Vanna AI

Vanna AI is an open-source, multi-database SQL agent framework that lets users query company data systems using natural language. It connects any LLM provider (OpenAI, Anthropic, Gemini, Ollama) to any database (PostgreSQL, MySQL, Snowflake, BigQuery, SQLite, and more), generating accurate SQL from plain English questions. Vanna learns from successful interactions via Tool Memory, improving over time without manual retraining. Optional hosted admin features add access control, observability, audit logs, and more for production deployments.

• Multi-LLM Support: Works with Anthropic, OpenAI, Gemini, and Ollama — swap providers without rewriting code.
• Multi-Database Compatibility: Connects to PostgreSQL, MySQL, Snowflake, BigQuery, SQLite, and many more databases.
• Tool Memory: Saves successful SQL queries to a vector database so the agent learns your schema and business logic over time.
• Access Control: Per-user permissions flow through the entire system, ensuring users only see data they're authorized to access.
• Observability: Real-time metrics and distributed tracing for monitoring agent behavior in production.
• Audit Logs: Comprehensive activity tracking for compliance and governance requirements.
• Self-Host or Cloud: Run Vanna fully self-hosted via pip install vanna, or use the managed Vanna Cloud with hosted admin features.
• Open Source Core: The core framework is MIT-licensed on GitHub with 23k+ stars, with no vendor lock-in.
• Multi-Turn Conversations: Supports follow-up questions and context-aware dialogue for iterative data exploration.
• Quick Start: Get running in 1 minute with a sample Chinook SQLite database in Colab, or connect your own database in ~30 minutes.