Vanna.ai is an MIT-licensed open-source text-to-SQL framework that pioneered the trainable RAG (Retrieval-Augmented Generation) approach. Unlike static tools, Vanna learns from your specific DDL statements, documentation, and example SQL queries to improve accuracy over time. Self-hosted deployment offers complete data sovereignty with no external API calls required. Supports any LLM backend (OpenAI, Anthropic Claude, Google Gemini, local models via Ollama) and any vector database (ChromaDB, Pinecone, pgvector, Weaviate, Qdrant). Vanna 2.0 enterprise features include row-level security, user-aware agents, lifecycle hooks for custom workflows, and automatic schema change detection. The Python library integrates with Jupyter notebooks, Streamlit, Flask, and Slack. Page should cover: installation and setup guide, training methodology explanation, LLM and vector database configuration options, self-hosted vs cloud deployment comparison, enterprise features deep-dive, accuracy improvement strategies, and comparison with Wren AI.