# ShannonBase

A MySQL-compatible database that adds vector search, ML model training, local LLM inference and RAG pipelines as native SQL calls, collapsing a typical AI data stack into one engine.

ShannonBase extends MySQL with vector search, machine learning training, embedded ONNX Runtime LLM inference, retrieval-augmented generation and agent execution, all callable from ordinary SQL, aiming to replace a stack of MySQL plus an ETL pipeline plus a separate vector database, feature store and inference server with one wire-protocol-compatible engine that existing MySQL drivers, ORMs and tools such as Navicat, DBeaver and SQLAlchemy work with unchanged.

A `sys.ML_GENERATE` SQL call runs text generation against a locally loaded ONNX model such as Qwen2.5-0.5B-Instruct, `sys.ML_RAG` runs a full retrieval-augmented generation pipeline in one call, `sys.ML_TRAIN` trains a model on live data without exporting it to Python, and stored procedures can be written in JavaScript through an embedded JerryScript engine. It runs as a Docker image or built from source. C++.

Website: https://www.shannonbase.org