AI RAG Chatbot Guide

This is based on the starter project for the Vercel AI SDK Retrieval-Augmented Generation (RAG) guide. I have edited parts of the stack for an easier setup. Choose the script based on your OS below. MacOS and Windows are currently supported.

Stack

• Next.js 14 (App Router)
• Vercel AI SDK
• OpenAI
• Drizzle ORM
• Postgres Docker Image
• shadcn-ui and TailwindCSS

Prerequisites

Be sure to have an OpenAI API key, Git and pnpm installed.

• OpenAI API Key
• Git
• pnpm (brew install pnpm or choco install pnpm)

Clone the Repository, install dependencies and create .env file

git clone https://github.com/carloscisnerosjr/ai-rag-chatbot

Run pnpm install to install the dependencies.

Rename .env.example to .env and add your OpenAI API key. Keep the database URL as is.

Run the Setup Script

Run the setup script at the root of the repository based on your OS.

MacOS

• In Terminal, run bash setup.sh.

You may need to accept incoming networking connections on a Mac when running the script.

Windows

• Run Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass -Force on Windows to allow the script to run.
• In Powershell, run ./setup.ps1.

The server and database will start automatically after the script finishes. Navigate to http://localhost:3000 to view the chatbot and to https://local.drizzle.studio to view the database.

Chat with the Bot and verify empty knowledge base

Navigate to http://localhost:3000. Ask the chatbot it knows your favorite food. It will not know the answer because it is not trained on that data!

How can we integrate a knowledge base that the chatbot can use?

Verify database is empty

Navigate to https://local.drizzle.studio in your browser to view the database. Notice that the embeddings table is empty. The bot does not have any knowledge of your favorite food!

Add to the bot's knowledge base

Tell the bot your favorite food. Check the database again and there should now be a row with data about your favorite food! Your first embedding is stored in the database!

Ask the bot your favorite food again. It should now know the answer! Congrats, you've just added to the bot's knowledge base using embeddings and built your first RAG chatbot!

Stop the server and database studio

• Run pnpm stop to stop the server.
• Run pnpm db:studio:stop to stop the database studio.
• Run docker compose down -v --rmi all to remove the Docker container.
• Remove your API key from the .env file.