SQLite RAG with Ollama

A simple demonstration of building a Retrieval Augmented Generation (RAG) system using SQLite and Ollama for local, on-device vector search. This project implements a movie recommendation system to showcase RAG capabilities without requiring complex infrastructure.

Overview

This project demonstrates how to:

• Store and query vector embeddings in SQLite
• Generate embeddings using Ollama's local models
• Perform semantic similarity search for movie recommendations

For a detailed walkthrough of this project, check out the companion blog post: Doing on-device retrieval augmented generation with Ollama and SQLite

Prerequisites

• SQLite3
• sqlite-utils
• sqlite-vec extension
• Ollama
• curl
• jq

Usage

To embed the data and build the database, run the embed.sh script.

sh embed.sh

To perform a search, run the search.sh script with a query.

sh search.sh "a movie about time travel"

License

This project is licensed under the MIT License. See the LICENSE file for details.