Skip to content

RAG-WORKFLOW-GUIDE.md

Latest commit

 

History

History
193 lines (140 loc) · 5.69 KB

RAG-WORKFLOW-GUIDE.md

File metadata and controls

193 lines (140 loc) · 5.69 KB
Raw

Cloudflare RAG Workflow Guide

This guide shows how to build a Retrieval Augmented Generation (RAG) workflow using Cloudflare Workers AI and Vectorize with the existing n8n-nodes-cloudflare package.

Overview

A RAG workflow consists of two phases:

  1. Ingestion: Convert documents → Generate embeddings → Store in vector database
  2. Retrieval: User query → Generate query embedding → Search vectors → Return results

Available Nodes

Node Purpose
Cloudflare Workers AI Generate text embeddings using @cf/baai/bge-base-en-v1.5
Cloudflare Vectorize Store and query vectors in Cloudflare's vector database

🔧 Ingestion Workflow

Step 1: Prepare Your Documents

Use n8n's built-in nodes to get your text data:

  • Read Binary File → Read documents from disk
  • HTTP Request → Fetch content from APIs
  • Code Node → Transform data into text chunks

Step 2: Generate Embeddings (Workers AI)

Use the Cloudflare Workers AI node:

Setting Value
Resource Inference
Operation Run Text Embedding
Model @cf/baai/bge-base-en-v1.5
Text {{ $json.text }} (your document text)

Output: An array of embedding vectors (768 dimensions for bge-base-en-v1.5)

Step 3: Store in Vectorize

Use the Cloudflare Vectorize node:

Setting Value
Resource Vector
Operation Create or Update (upsert)
Index Name Your vectorize index
Vectors See JSON format below

Vectors JSON format:

[
  {
    "id": "doc-1",
    "values": [0.1, 0.2, ...],  // From Workers AI output
    "metadata": {
      "text": "Original document text",
      "source": "my-document.pdf"
    }
  }
]

🔍 Retrieval Workflow

Step 1: Get User Query Embedding

Use Cloudflare Workers AI:

Setting Value
Resource Inference
Operation Run Text Embedding
Model @cf/baai/bge-base-en-v1.5
Text {{ $json.query }} (user's question)

Step 2: Search Vectorize

Use Cloudflare Vectorize:

Setting Value
Resource Vector
Operation Query
Query Vector {{ $json.data[0] }} (embedding from step 1)
Top K 5 (number of results)
Return Metadata All (to get the original text)

Output: Top K matching documents with similarity scores

Step 3: Generate Response (Optional)

Use Cloudflare Workers AI again:

Setting Value
Resource Inference
Operation Run Text Generation
Model @cf/meta/llama-3-8b-instruct
Prompt See template below

Prompt template:

Based on the following context, answer the user's question.

Context:
{{ $json.matches.map(m => m.metadata.text).join('\n\n') }}

Question: {{ $('Webhook').item.json.query }}

Answer:

📋 Complete Workflow JSON

Import this workflow template into n8n:

{
	"name": "Cloudflare RAG Example",
	"nodes": [
		{
			"parameters": {
				"resource": "inference",
				"operation": "textEmbedding",
				"accountId": "={{ $credentials.accountId }}",
				"modelName": "@cf/baai/bge-base-en-v1.5",
				"embeddingText": "={{ $json.text }}"
			},
			"name": "Generate Embedding",
			"type": "n8n-nodes-cloudflare.cloudflareWorkersAi",
			"position": [450, 300]
		},
		{
			"parameters": {
				"resource": "vector",
				"operation": "query",
				"accountId": "={{ $credentials.accountId }}",
				"indexName": "my-index",
				"queryVector": "={{ JSON.stringify($json.data[0]) }}",
				"queryOptions": {
					"topK": 5,
					"returnMetadata": "all"
				}
			},
			"name": "Search Vectorize",
			"type": "n8n-nodes-cloudflare.cloudflareVectorize",
			"position": [650, 300]
		}
	]
}

⚠️ Limitations

No Native LangChain Cluster Node Integration

This package provides standard n8n nodes, not LangChain "cluster nodes". This means:

  • ❌ Cannot connect directly to n8n's Document Loader nodes
  • ❌ Cannot plug into AI Agent's "tool" connector
  • ❌ Not compatible with n8n's RAG Starter Template UI

Why? n8n's cluster node architecture (NodeConnectionTypes.AiEmbedding, NodeConnectionTypes.AiVectorStore) is internal to the @n8n/nodes-langchain package and not exposed for community nodes.

Request Native Support

If you need full LangChain integration, please submit a feature request to n8n:

  1. Go to n8n Feature Requests
  2. Request: "Native Cloudflare Vectorize & Workers AI Vector Store nodes"
  3. Reference the existing Supabase/Pinecone/Qdrant implementations

🔗 Resources