Local AI Workflows with n8n — Workshop

🛠️ Installation Guide

Follow these steps on Windows before the workshop. Only needs to be done once.

1️⃣

Install Docker Desktop

Download and install Docker Desktop for Windows. Enable WSL2 when prompted during installation.

Verify installation

docker --version
docker compose version

⚠️ Restart your computer after Docker Desktop installs. The Docker whale icon must appear in the Windows system tray before continuing.

2️⃣

Clone the n8n AI Starter Kit

Open PowerShell and run these commands to download and start the n8n AI Starter Kit.

# Clone the starter kit
git clone https://github.com/n8n-io/self-hosted-ai-starter-kit.git
cd self-hosted-ai-starter-kit

# Start all services in CPU mode
docker compose --profile cpu up -d

# Check all 4 containers are running
docker ps

✅ Expected: 4 containers running — n8n, ollama, postgres, qdrant

3️⃣

Verify Services

Open these URLs in your browser to confirm everything is running correctly.

n8n Dashboard

http://localhost:5678

Qdrant Dashboard

http://localhost:6333/dashboard

💡 Always use http://ollama:11434 inside n8n nodes — never localhost. Containers communicate using service names.

4️⃣

First Login to n8n

When you open n8n for the first time it asks you to create an account. Use these workshop credentials:

Email	`admin@n8n.local`
Password	`Admin@1234`
First Name	`Admin`
Last Name	`User`

⚠️ For lab use only.

🧪 The 7 Exercises

Each exercise builds on the previous. Download the workflow JSON, import it into n8n, and follow the lab manual.

🔍 AI Code Reviewer

Webhook · Basic LLM Chain · tinyllama

⏱ 15 min3 Nodes

Submit any code snippet via HTTP POST and receive a professional AI code review — issues found and improvements suggested. Test using the browser HTML tester page provided.

Webhook → Basic LLM Chain → Respond to Webhook ↑ Ollama Model (tinyllama)

Node	Key Parameter
Webhook	`Path: roast-my-code · Method: POST · Response Mode: Response Node`
Basic LLM Chain	`Prompt (Expression): ={{ $json.body.code }}`
Ollama Model	`http://ollama:11434 · Model: tinyllama`
Respond to Webhook	`={{ JSON.stringify({ roast: $json.text }) }}`

⬇️ Workflow JSON ⬇️ Browser Tester

🧠 Explain Like I'm 5

Webhook · Basic LLM Chain · Prompt Engineering · phi3:mini

⏱ 15 min3 Nodes

Submit any CS topic and receive a simple, jargon-free explanation anyone can understand. Teaches prompt engineering — how the system message completely changes the AI output style.

Webhook → Basic LLM Chain → Respond to Webhook ↑ Ollama Model (phi3:mini)

Node	Key Parameter
Webhook	`Path: explain-eli5 · Method: POST · Response Mode: Response Node`
Prompt	`=Explain {{ $json.body.topic }} like I am a 5 year old...`
System Message	`You are a friendly teacher. Use simple words. No jargon...`
Respond to Webhook	`={{ JSON.stringify({ explanation: $json.text }) }}`

⬇️ Workflow JSON ⬇️ Browser Tester

📂 AI File Summariser

Manual Trigger · Docker File Handling · phi3:mini

⏱ 15 min6 Nodes

Upload a text document into the n8n Docker container, let AI summarise it in exactly 3 sentences, and download the result. Teaches Docker file handling with docker cp commands.

Manual Trigger → Read File → Extract from File → Basic LLM Chain ↑ Ollama Model (phi3:mini) ↓ Convert to File → Write File

Docker file commands

# Create folder (run after every Docker restart)
docker exec -it n8n mkdir -p /home/node/.n8n-files

# Upload document to container
docker cp document.txt n8n:/home/node/.n8n-files/document.txt

# Download summary to current folder
docker cp n8n:/home/node/.n8n-files/summary.txt .

Node	Key Parameter
Read File	`/home/node/.n8n-files/document.txt`
Prompt	`={{ $json.data }} (Expression mode)`
System Message	`Summarise in exactly 3 clear sentences...`
Write File	`/home/node/.n8n-files/summary.txt`

⬇️ Workflow JSON

🌐 Simple HTTP Request

Chat Trigger · HTTP Request · Edit Fields · Public API

⏱ 5 min3 NodesNo AI Model

Fetch live data from a public API and display it in the n8n chat. Teaches HTTP GET requests and JSON response handling — the foundation of all API integrations in n8n.

Chat Trigger → HTTP Request → Edit Fields (catfact.ninja) (extracts fact)

Node	Key Parameter
HTTP Request	`Method: GET · URL: https://catfact.ninja/fact`
Edit Fields	`Name: fact · Value: ={{ $json.fact }}`
Advice API (bonus)	`https://api.adviceslip.com/advice → $json.slip.advice`

⬇️ Workflow JSON

🧮 AI Calculator

Chat Trigger · AI Agent · Calculator Tool · Simple Memory · qwen2.5:3b

⏱ 10 minAI Agent

A conversational AI financial assistant that uses the built-in Calculator Tool for exact computations and remembers the last 10 messages. Introduces the AI Agent — n8n's most powerful AI construct.

Chat Trigger → AI Agent → Response ↑ ↑ ↑ Chat Model Memory Calculator (qwen2.5:3b) (10) Tool

Node	Key Parameter
AI Agent	`System: You are a helpful financial assistant...`
Ollama Chat Model	`http://ollama:11434 · qwen2.5:3b · keepAlive: 1h · numPredict: 200`
Simple Memory	`Context Window Length: 10`
Calculator Tool	`No parameters needed`

⬇️ Workflow JSON

💾 AI Interaction Logger

Chat Trigger · AI Agent · Code Tool · Insert Row · Data Table

⏱ 15 minAI Agent + DB

An AI CS tutor that answers questions AND automatically logs every interaction to an n8n Data Table — saving the question, answer summary, word count, sentence count, and timestamp permanently.

Chat Trigger → AI Agent → Insert Row (Data Table) ↑ ↑ ↑ Chat Model Memory Code Tool

Code Tool — JavaScript

const input = $input.first().json;
const text = input.code || input.query || input._query || '';
const words = text.trim().split(/\s+/).filter(w => w).length;
const sentences = text.split(/[.!?]+/).filter(s => s.trim()).length;
return `Analysed: ${words} words, ${sentences} sentences.`;

Field	Expression
question	`={{ $('When chat message received').item.json.chatInput }}`
answer_summary	`={{ $json.output.slice(0, 200) }}`
word_count	`={{ $json.output.trim().split(/\s+/).length }}`
sentence_count	`={{ $json.output.split(/[.!?]+/).filter(s=>s.trim()).length }}`
timestamp	`={{ new Date().toISOString().replace("T"," ").slice(0,19) }}`

⬇️ Workflow JSON

🔍 RAG Pipeline

Two Workflows · Qdrant · mxbai-embed-large · qwen2.5:3b

⏱ 20 minTwo WorkflowsVector DB

Build a full RAG system — upload a PDF, store it as vectors in Qdrant, then chat with your own document. The AI answers only from your uploaded content, not from training data.

Workflow 1: n8n Form → Qdrant (insert) ← Embeddings + Text Splitter Workflow 2: Chat Trigger → AI Agent → Response ↑ ↑ ↑ Model Memory Qdrant Retriever ← Embeddings Ollama

Setting	Value
Qdrant Collection	`rag_collection · Vector Size: 1024 · Distance: Cosine`
Text Splitter	`Chunk Size: 200 · Chunk Overlap: 50`
Embedding Model	`mxbai-embed-large:latest (both workflows)`
Tool Name	`retriever`
Chat Model	`qwen2.5:3b · keepAlive: 1h · numPredict: 150`

⚠️ The Qdrant retriever needs its own separate Embeddings Ollama node — it cannot share the one from Workflow 1.

⬇️ Workflow JSON ⬇️ Sample PDF

Local AI Workflows
with n8n

Install Docker Desktop

Clone the n8n AI Starter Kit

Verify Services

First Login to n8n

Pull All AI Models

File Folder Setup — Exercise 3

Complete Lab Manual

Workshop Presentation

MCQ Assessment

Sample RAG PDF

All Workflow Files

Browser Testers

Local AI Workflowswith n8n

Install Docker Desktop

Clone the n8n AI Starter Kit

Verify Services

First Login to n8n

Pull All AI Models

File Folder Setup — Exercise 3

Complete Lab Manual

Workshop Presentation

MCQ Assessment

Sample RAG PDF

All Workflow Files

Browser Testers

Local AI Workflows
with n8n