RAG Cloudflare

A Retrieval-Augmented Generation (RAG) pipeline that ingests PDF documents into Cloudflare Vectorize and answers questions using an LLM via OpenRouter. Built with Bun.

How It Works

1. Ingest - PDFs in the data/ folder are read, split into chunks, embedded, and stored in a Cloudflare Vectorize index.
2. Query - A user question is embedded, the most relevant chunks are retrieved from Vectorize, and an LLM generates an answer grounded in those chunks.

Prerequisites

• Bun 1.0+
• An OpenRouter API key
• A Cloudflare account ID
• A Cloudflare API token with Vectorize permissions

Setup

1. Install dependencies

   bun install

2. Configure environment variables

   Copy the example file and fill in your credentials:

   cp .env.example .env

3. Edit .env

   OPENROUTER_API_BASE=https://openrouter.ai/api/v1
   OPENROUTER_API_KEY=your_openrouter_api_key
   OPENROUTER_MODEL=deepseek/deepseek-v4-pro
   OPENROUTER_CONTEXT_MODEL=gpt-4o-mini
   OPENROUTER_MAX_TOKENS=512
   OPENROUTER_EMBEDDING_MODEL=text-embedding-3-small
   CLOUDFLARE_ACCOUNT_ID=your_cloudflare_account_id
   CLOUDFLARE_API_TOKEN=your_cloudflare_api_token
   CLOUDFLARE_VECTORIZE_INDEX=belajar-rag

4. Add PDF files

   Place your PDF files in data/.

Usage

Ingest PDFs

bun ingest.js

To completely reset the index before ingesting:

bun ingest.js --delete-old

This will:

• Read all .pdf files from data/
• Split text into chunks with overlap
• (Optionally) Reset the Cloudflare Vectorize index if --delete-old is provided
• Upload fresh embeddings with stable IDs (subsequent runs without --delete-old will update existing documents rather than duplicating them)

Query

bun query.js "What is the main topic of the document?"

The output includes the answer and source chunks with relevance scores.

Project Structure

├── config.js
├── ingest.js
├── query.js
├── .env.example
├── package.json
└── data/