Vanna AI

Vanna AISince 2023-08

Vanna AI is an open-source Python-based RAG (Retrieval-Augmented Generation) framework engineered to generate high-accuracy SQL queries from plain English questions. Unlike general-purpose chatbots that frequently hallucinate non-existent database columns and tables, Vanna trains specifically on your database schema, table DDL, documentation, and historical query logs. When a business user or developer asks a natural language question, Vanna retrieves relevant schema definitions and validated SQL examples to construct an accurate, executable SQL query for PostgreSQL, Snowflake, BigQuery, MySQL, SQLite, or SQL Server. With over 12,000 GitHub stars, Vanna allows organizations to deploy self-hosted text-to-SQL agents inside Slack, Streamlit dashboards, or internal REST APIs without exposing private database records to third parties.

Vanna operates on a simple two-phase architecture: training and inference. In the training phase, you feed Vanna your database DDL statements, business documentation, and validated SQL query pairs, which Vanna stores in a vector database (such as ChromaDB, Qdrant, or Pinecone). During inference, Vanna uses semantic vector search to find the closest matching schema definitions and verified SQL patterns, injecting them into the LLM context (supporting OpenAI, Anthropic, Mistral, and local Ollama models) to synthesize the target SQL query. Vanna can automatically execute the query against your database connection, generate interactive Plotly charts, and explain the SQL logic in plain English.

MarimoSince 2023-11

Marimo is a modern, open-source reactive Python notebook that solves the reproducibility and out-of-order execution nightmares of traditional Jupyter notebooks. Designed for data scientists, ML engineers, and researchers, Marimo guarantees that notebook state is always consistent by treating code cells as a Directed Acyclic Graph (DAG). When you modify a variable or code cell in Marimo, all dependent downstream cells automatically re-run instantly, eliminating hidden state bugs. Furthermore, Marimo notebooks are stored as pure, version-control-friendly Python scripts rather than messy JSON files. With built-in AI code generation, interactive UI sliders, instant conversion into shareable web apps, and native SQL query cells, Marimo represents the state-of-the-art computational notebook for 2026.

Marimo's reactive execution engine analyzes variable references and mutations across cells to construct a dynamic dependency graph. This ensures deterministic execution whether running interactively in the browser, testing via Git CI/CD pipelines, or deploying as a standalone dashboard. The notebook environment includes built-in interactive UI elements (sliders, dropdowns, tables, file uploaders) that bind directly to Python variables without external JavaScript frameworks. Marimo also features native DuckDB SQL integration, allowing users to query Pandas and Polars DataFrames using standard SQL syntax. Its integrated AI assistant allows developers to generate data visualization scripts, explain complex algorithms, and refactor data pipelines directly inside the notebook interface with support for local Ollama models and cloud LLM providers.