chatgpt

Connecting Cloud n8n with Local Ollama – Complete Tutorial

by
an abstract image of a sphere with dots and lines

Introduction

This tutorial will guide you through connecting your cloud-hosted n8n instance with Ollama running locally on your PC. We’ll cover multiple methods, from quick testing solutions to more permanent setups.

What is Ollama?

Ollama is a free, open-source application that lets you run large language models (LLMs) locally on your own computer. Think of it as having ChatGPT running on your PC instead of in the cloud.

Prerequisites

  • n8n instance hosted in the cloud (e.g., n8n.cloud, self-hosted VPS)
  • Ollama installed and running on your local PC
  • Basic command line knowledge
  • Administrator access to your PC

Understanding the Problem

Your cloud n8n instance cannot directly access your local Ollama because:

  • Your local PC is behind a router/firewall
  • It doesn’t have a public IP address accessible from the internet
  • Cloud services can only connect to publicly accessible endpoints

Solution: Create a secure tunnel that exposes your local Ollama to the internet.

Method 1: Using ngrok (Best for Testing)

Why ngrok?

  • Quick setup (5 minutes)
  • Free tier available
  • Automatic HTTPS
  • Perfect for development and testing

Step 1: Install ngrok

  1. Visit ngrok.com and create a free account
  2. Download ngrok for your operating system
  3. Extract the executable to a folder (e.g., C:\ngrok)

Step 2: Configure ngrok

  1. Copy your authtoken from the ngrok dashboard
  2. Open Command Prompt or PowerShell and run: ngrok config add-authtoken YOUR_AUTH_TOKEN

Step 3: Start Ollama

  1. Make sure Ollama is running on your PC
  2. By default, Ollama runs on port 11434
  3. Test it locally by opening: http://localhost:11434

Step 4: Create the Tunnel

  1. Open a new terminal window
  2. Run: ngrok http 11434
  3. You’ll see output like: Forwarding https://abc123-456.ngrok.io -> http://localhost:11434
  4. Copy this URL – this is your public Ollama endpoint!

Step 5: Configure n8n

  1. Log into your n8n instance
  2. Create a new workflow
  3. Add an HTTP Request node
  4. Configure it:
    • Method: POST
    • URL: https://abc123-456.ngrok.io/api/generate
    • Authentication: None
    • Body: JSON
    • JSON Body: { "model": "llama2", "prompt": "Why is the sky blue?", "stream": false}
  5. Execute the node to test

Step 6: Parse the Response

  1. Add a Set node after the HTTP Request
  2. Extract the response:
    • Name: ollama_response
    • Value: {{ $json.response }}

Important Notes for ngrok

  • Free tier: Your URL changes every time you restart ngrok
  • Keep terminal open: The tunnel stays active only while ngrok is running
  • Paid tier: Get a permanent URL that doesn’t change

Method 2: Using Cloudflare Tunnel (Best for Permanent Setup)

Why Cloudflare Tunnel?

  • Free forever
  • More reliable than ngrok free tier
  • Can create named tunnels
  • Better for long-term projects

Step 1: Install Cloudflare Tunnel

Windows (using winget):

winget install --id Cloudflare.cloudflared

Windows (manual):

  1. Download from Cloudflare’s website
  2. Extract to a folder like C:\cloudflared
  3. Add to PATH or use full path in commands

Linux/Mac:

# Linux
wget https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb
sudo dpkg -i cloudflared-linux-amd64.deb

# Mac
brew install cloudflare/cloudflare/cloudflared

Step 2: Authenticate Cloudflare

  1. Run: cloudflared tunnel login
  2. Your browser will open – log in with your Cloudflare account
  3. Select a domain (or create one free at Cloudflare)
  4. Authorize the tunnel

Step 3: Quick Tunnel (Temporary)

For quick testing:

cloudflared tunnel --url http://localhost:11434

This gives you a temporary URL like: https://random-words.trycloudflare.com

Step 4: Named Tunnel (Permanent)

For a permanent setup:

  1. Create a tunnel: cloudflared tunnel create ollama-tunnel Note the tunnel ID that appears
  2. Create a config file (~/.cloudflared/config.yml or C:\Users\YourName\.cloudflared\config.yml): tunnel: TUNNEL_ID_HERE credentials-file: C:\Users\YourName\.cloudflared\TUNNEL_ID.json ingress: - hostname: ollama.yourdomain.com service: http://localhost:11434 - service: http_status:404
  3. Create DNS record: cloudflared tunnel route dns ollama-tunnel ollama.yourdomain.com
  4. Run the tunnel: cloudflared tunnel run ollama-tunnel
  5. Run as a service (optional, so it starts automatically): # Windows (PowerShell as Admin) cloudflared service install # Linux sudo cloudflared service install

Step 5: Use in n8n

Use your tunnel URL: https://ollama.yourdomain.com/api/generate

Method 3: Router Port Forwarding (Advanced)

Why Port Forwarding?

  • No third-party service needed
  • Full control
  • Best performance

When NOT to use this:

  • You don’t have a static IP
  • Your ISP blocks incoming connections
  • You’re not comfortable with network security

Steps:

  1. Find your local IP:
    • Windows: Run ipconfig in CMD
    • Look for “IPv4 Address” (e.g., 192.168.1.100)
  2. Access your router:
    • Usually 192.168.1.1 or 192.168.0.1
    • Log in with admin credentials
  3. Create port forwarding rule:
    • External Port: 11434
    • Internal IP: Your PC’s IP (e.g., 192.168.1.100)
    • Internal Port: 11434
    • Protocol: TCP
  4. Configure Windows Firewall: # Run as Administrator New-NetFirewallRule -DisplayName "Ollama" -Direction Inbound -LocalPort 11434 -Protocol TCP -Action Allow
  5. Set up Dynamic DNS (if no static IP):
    • Use services like DuckDNS, No-IP, or Dynu
    • Install their client on your PC
    • Get a free domain like yourname.duckdns.org
  6. Use in n8n:
    • URL: http://YOUR_PUBLIC_IP:11434/api/generate
    • Or: http://yourname.duckdns.org:11434/api/generate

Building a Complete n8n Workflow

Here’s a practical example workflow that uses your local Ollama:

Workflow: AI-Powered Email Responder

  1. Trigger Node: Manual Trigger or Webhook
  2. HTTP Request Node (Call Ollama):
    • Method: POST
    • URL: YOUR_TUNNEL_URL/api/generate
    • Body: { "model": "llama2", "prompt": "Write a professional email response to: {{ $json.email_content }}", "stream": false}
  3. Code Node (Parse Response): const response = $input.item.json.response; return [{ json: { ai_response: response } }];
  4. Send Email Node or Slack Node:
    • Use {{ $json.ai_response }} as the message

Security Best Practices

1. Add Basic Authentication

Wrap your Ollama endpoint with a reverse proxy (nginx, Caddy) that requires authentication.

2. Use HTTPS Only

  • ngrok: Provides HTTPS automatically
  • Cloudflare: HTTPS by default
  • Port forwarding: Set up Let’s Encrypt

3. IP Whitelisting

If your n8n instance has a static IP, configure your firewall to only accept connections from that IP.

4. Monitor Access

  • Check ngrok/Cloudflare dashboards for unusual traffic
  • Set up alerts for excessive requests

5. Use Environment Variables in n8n

Store your tunnel URL as an environment variable instead of hardcoding it:

  • URL: {{ $env.OLLAMA_URL }}/api/generate

Troubleshooting

“Connection Refused” Error

Check:

  1. Is Ollama running? Test: curl http://localhost:11434
  2. Is your tunnel active? Check the terminal window
  3. Is the URL correct in n8n?

“Timeout” Error

Solutions:

  1. Increase timeout in n8n HTTP Request settings
  2. Use smaller models or shorter prompts
  3. Check your internet connection speed

ngrok URL Changed

Solution:

  • Upgrade to ngrok paid plan for static URLs
  • Or switch to Cloudflare Tunnel (free permanent URLs)

Ollama Not Responding

Check:

  1. Restart Ollama service
  2. Check if port 11434 is already in use: # Windowsnetstat -ano | findstr :11434
  3. Check Ollama logs for errors

Firewall Blocking Connection

Windows:

# Allow Ollama through firewall
New-NetFirewallRule -DisplayName "Ollama" -Direction Inbound -Program "C:\Program Files\Ollama\ollama.exe" -Action Allow

Performance Tips

  1. Use "stream": false in your requests for easier n8n handling
  2. Keep models small for faster responses (e.g., llama2:7b instead of llama2:70b)
  3. Cache responses in n8n if you’re asking the same questions repeatedly
  4. Set appropriate timeouts – AI generation can take 30-60 seconds for longer responses

Cost Comparison

MethodFree TierPaid TierBest For
ngrokURL changes on restart$8/month for static URLTesting
CloudflareFree foreverFree (or paid plans available)Production
Port ForwardFreeFreeAdvanced users

Next Steps

  1. Test the connection with simple prompts
  2. Build a workflow that uses your local LLM
  3. Secure your setup using the security practices above
  4. Scale up by running multiple Ollama models
  5. Automate the tunnel startup using system services

Additional Resources

Conclusion

You now have multiple ways to connect your cloud n8n with local Ollama:

  • Quick testing: Use ngrok
  • Long-term projects: Use Cloudflare Tunnel
  • Maximum control: Use port forwarding

Start with ngrok to test everything works, then move to Cloudflare Tunnel for a permanent setup. Happy automating with local LLMs!

My AI experiments

by

Of late, I have been experimenting a lot with AI tools. These experiments cement my views on the benefits of using these development assistants to enhance productivity.

Full transparency: I was an average developer at best. I was a late entrant into computers. The first couple of years of my career were spent working for the government in far-flung posts, locations that lacked regular electricity or running water. A return to ‘civilian life’ turned me towards computers, which have been a part of my life for the last three decades. Now, to the main point.

A couple of weeks back, I created three game apps using AI tools and posted them on my website. To explain further, I have not written a line of code in these apps. That might explain their success, but that’s another story.
Let me clarify that I can understand code and, to some extent, what is written in error logs and can also test apps.

Let me now explain how ‘I’ built those apps. I started by defining what I wanted in a few sentences. Next, I copied and pasted this draft into Grok and asked it to write a professional requirements document. I handed this document to Figma Make.

Within seconds, Figma made a prototype with CSS, TSX, and TS files. The end result was a working sample app I could test immediately. I gave a few more instructions and minor tweaks, which it followed like an intelligent developer, and my basic app was ready.

By the way, Figma does all this for free. Once I was satisfied with the end result, I downloaded the files and put them in a folder on my laptop. I opened this folder in Cursor AI and asked it to transform it into a WordPress plugin zip. The plugins are now installed and running on my blog.

My point is this: if a man in his late fifties with a basic understanding of code can do this, imagine what smart, energetic, and intelligent YOU can do.
If you have ideas or see an opportunity to turn your imagination into workable software, now is the time. No need to spend years learning and memorizing all the nuances of a programming language. With basics, you can achieve a lot. Think of yourself as a business person who hires a smart developer to write some code.

The catch is, These apps are not exactly ‘no code’ or even ‘low code’. It is a bit more than low code and involves a lot of manual testing on your part. Also, do not trust the AI tool blindly, as these systems are still in their infancy and learning to walk. You have to hold its hand and gently guide it. Remember the movie Superman when the toddler crash-lands at Kent’s farm. It could lift a car with its hands, but it was still a baby. That is AI for you as of today.

A day may come when it won’t fully obey your commands and tells you to do your coding yourself, but till that day comes, enjoy the show. Have some fun building the app of your dreams and get it done fast.