n8n + Ollama Tutorial: Automate Everything with Local AI (No API Keys)

Published March 24, 2026 · 15 min read · Beginner-friendly

If you've been wanting to add AI to your automation workflows but don't want to pay for OpenAI API keys or send your data to the cloud, this tutorial is for you. We'll connect n8n (the open-source workflow automation tool) with Ollama (local AI model runner) to build powerful automations that run entirely on your own hardware.

        What you'll build: An email classification and auto-response system that reads incoming emails, categorizes them (support, sales, spam), and drafts appropriate replies — all using local AI.
    

Prerequisites

A computer with at least 8 GB RAM (16 GB recommended)
Basic familiarity with n8n (we'll explain everything, but knowing the UI helps)
~15 minutes of time

Part 1: Install the Stack

1 Install Ollama

# One-line install (Linux/macOS)
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull llama3:8b

# Test it works
curl http://localhost:11434/api/generate -d '{
  "model": "llama3:8b",
  "prompt": "Say hello in JSON format",
  "stream": false
}'

You should see a JSON response with the model's reply. If you get a connection error, run ollama serve first.

2 Install n8n

# Option A: Docker (recommended)
docker run -d --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  --add-host=host.docker.internal:host-gateway \
  n8nio/n8n

# Option B: npm
npm install n8n -g && n8n start

Open http://localhost:5678 and create your admin account.

3 Verify the connection

In n8n, create a new workflow with an HTTP Request node:

URL: http://localhost:11434/api/tags (or http://host.docker.internal:11434/api/tags for Docker)
Method: GET

Execute it. You should see a list of your installed models. If this works, n8n can talk to Ollama.

Part 2: Build the Email Classifier

Our workflow will have 4 nodes:

Trigger — Manual trigger (or connect to your email via IMAP)
Classify — AI classifies the email intent
Route — Switch node routes based on classification
Draft Response — AI drafts an appropriate reply

Node 1: The Trigger

Start with a Manual Trigger for testing. Later, you can swap it for an IMAP Email trigger to process real emails automatically.

Add a Set node after the trigger to simulate an email:

{
  "from": "john@example.com",
  "subject": "Can't log in to my account",
  "body": "Hi, I've been trying to log in for the past hour but keep getting an 'invalid credentials' error. I've tried resetting my password twice. This is really frustrating. Can someone help?"
}

Node 2: AI Classification

Add an HTTP Request node to call Ollama:

URL: http://localhost:11434/api/generate
Method: POST
Body (JSON):
{
  "model": "llama3:8b",
  "prompt": "Classify this email into exactly ONE category: support, sales, partnership, feedback, spam.\n\nAlso rate the urgency: low, medium, high.\nAnd detect the sentiment: positive, neutral, negative.\n\nEmail from: {{ $json.from }}\nSubject: {{ $json.subject }}\nBody: {{ $json.body }}\n\nRespond with ONLY JSON:\n{\"category\": \"...\", \"urgency\": \"...\", \"sentiment\": \"...\", \"reason\": \"...\"}",
  "stream": false,
  "options": { "temperature": 0.1 }
}

Low temperature (0.1) ensures consistent, deterministic classification.

Node 3: Parse and Route

Add a Code node to parse the AI response:

const response = $input.first().json.response;
const match = response.match(/\{[\s\S]*\}/);
if (match) {
  const classification = JSON.parse(match[0]);
  return [{ json: {
    ...classification,
    original_email: $('Set').first().json
  }}];
}
throw new Error('Could not parse AI classification');

Then add a Switch node that routes based on {{ $json.category }}:

support → Draft support response
sales → Draft sales response
spam → Archive/delete
Default → Forward to team

Node 4: Draft Response

For each route, add another HTTP Request to Ollama:

{
  "model": "llama3:8b",
  "prompt": "Draft a helpful reply to this support email.\n\nFrom: {{ $json.original_email.from }}\nSubject: {{ $json.original_email.subject }}\nBody: {{ $json.original_email.body }}\n\nBe empathetic, professional, and solution-oriented. Include specific troubleshooting steps.\n\nWrite the reply only, no subject line needed.",
  "stream": false,
  "options": { "temperature": 0.5 }
}

Part 3: Taking It Further

This basic pattern — send text to Ollama, parse the JSON response, route based on the result — powers virtually every AI automation. Here's what you can build with it:

Content pipeline: Topic → research → outline → draft → edit → publish
Lead scoring: New signup → AI analyzes company & role → assigns score → routes to sales or nurture
Document processing: Upload document → AI summarizes → extracts key data → populates database
Meeting follow-up: Transcript → AI extracts action items → creates tasks in project management tool

Common Issues and Fixes

"Connection refused" — Ollama isn't running. Start it with ollama serve
Timeout errors — Increase the HTTP Request timeout to 120000ms or more
Bad JSON output — Add "Return ONLY valid JSON, no other text" to your prompts. Use temperature 0.1-0.3 for structured output
Docker can't reach Ollama — Use http://host.docker.internal:11434 instead of localhost
Slow generation — Use a smaller model (mistral:7b) or get a GPU. Even a modest GPU (RTX 3060) makes 5-10x difference

Want Pre-Built Workflows?

Skip the setup. We've built 11 production-ready n8n + Ollama workflows covering content generation, email automation, lead scoring, document processing, and more.

$39 one-time — import into n8n in 5 minutes

Get 11 Ready-Made Workflows →

Free Samples

Try before you buy: