Auto Categorise Outlook Emails with AI Using n8n

Introduction

In this guide, we're going to streamline and optimise how you manage your inbox. We'll be using n8n, a brilliant workflow automation tool, combined with AI to automatically categorise your Outlook emails, keeping you on top of your inbox.

You can follow this guide as a stand alone article, or use is as an accompanying guide to the below YouTube video.

What You'll Learn

Setting up a free static domain with ngrok
Configuring Microsoft Azure for Outlook authentication
Setting up n8n locally using Docker on Windows Subsystem for Linux (WSL)
Integrating Ollama's AI models for email categorisation
Creating and optimising the n8n workflow for email processing
Handling errors and exceptions in the workflow
Testing and refining the automation

Why This Setup?

Now, you might be wondering why we're going through all this trouble. Well, this automation setup is absolutely brilliant, and here's why:

✨ It saves you loads of time by automatically sorting and tagging emails

📈 Improves email management for those of us drowning in correspondence

🔧 It's customisable to fit your specific categorisation needs

🤖 Uses AI for more accurate and context-aware categorisation

🔒 It's a self-hosted solution, giving you better privacy and control

Prerequisites

Before we dive in, ensure you have the following components installed and set up:

Component	Requirement	Purpose
Operating System	Windows 10/11	Base system
Subsystem	WSL2	Linux environment
Container Platform	Docker Desktop	Service containerisation
IDE	Visual Studio Code	Code editing
Version Control	Git	Code management
Cloud Account	Microsoft Azure	Authentication
Email Service	Microsoft Outlook	Email processing

Note: This guide assumes you're somewhat familiar with basic programming concepts. Even if you're new, don't worry! We'll walk you through each step. Make sure you have the following installed on your Windows machine:

Windows Subsystem for Linux (WSL2): Allows you to run Linux distributions on Windows. Follow Microsoft's WSL installation guide.
Docker Desktop: Needed to run Docker containers. Download it from Docker's website.
Visual Studio Code (VS Code): A powerful code editor. Download it from here.
Git: For cloning repositories. Download it from Git's website.

If you don't have WSL, Docker Desktop and VS Code set up, follow this guide first

Step-by-Step Guide

1. Setting Up the Environment

🎥 Watch this section 00:33 - 04:22

1.1 Set Up ngrok for a Static Domain

We'll use ngrok to expose our local environment securely to the internet. This is necessary for webhooks and authentication callbacks.

Sign Up for ngrok
- Go to ngrok's website and sign up for a free account.
Create a Static Domain
- After logging in, navigate to the Dashboard.
- Click on Reserved in the left-hand menu and select Reserved Domains.
- Click Create Domain.
Note Down the Static Domain URL
- This will be something like https://yourname.ngrok.io.
- Keep this handy; we'll need it later.

1.2 Prepare the n8n Environment

Now, let's set up n8n in your development environment.

Open Visual Studio Code
- If you haven't installed it yet, download and install it from here.
Connect to WSL
- Install the Remote - WSL extension in VS Code.
- Click on the green >< icon in the bottom-left corner of VS Code.
- Select New WSL Window.
- This will open a new VS Code window connected to your WSL instance.
Clone the n8n Setup Repository
- Open the terminal in VS Code by pressing Ctrl + Shift + \ (backtick) or navigating to Terminal > New Terminal.
- In the terminal, run the following command:
```
git clone https://github.com/WayneSimpson/n8n-ngrok.git
```
- This will clone the repository into your WSL filesystem.

💡

⚠️ Important: Make sure you're using the correct repository URL as provided above. This is crucial, folks!

1.3 Configure the Docker Compose File

Navigate to the Cloned Repository Folder
- In the VS Code Explorer pane, open the n8n-ngrok folder.
Open docker-compose.yml
- This file defines the Docker services we'll be running.
Review the Defined Services
- PostgreSQL Database: For n8n's database.
- n8n: The workflow automation tool.
- Vector Database: Optional, for advanced AI features.
- ngrok: To expose your local services to the internet.

1.4 Set Up the Environment Variables

We’ll configure the .env file, which will store necessary environment variables securely. Follow these steps:

Create Your .env File
- Open the terminal in Visual Studio Code.
- Run the following command to copy the example file:
```
cp .env.example .env
```
- This creates a new .env file from the example template.
Update the .env File
- Open the .env file in VS Code for editing.
- We’ll need to set several keys, as described below:
Generate a Random Encryption Key
- Visit RandomKeygen and generate a 256-bit key.
- Copy this key and set it as n8n_ENCRYPTION_KEY in your .env file.
Generate a Random JWT Secret
- At RandomKeygen, generate another key to use as your JWT secret.
- Set this value as n8n_JWT_SECRET in your .env file.
Add Your OpenAI API Key (Optional)
- If you plan to use OpenAI’s API, set OPENAI_API_KEY to your OpenAI key in the .env file.
- If not, you can skip this step.
Set Your ngrok Authentication Token
- In the ngrok dashboard, under Setup & Installation, find your authentication token.
- Copy and set this as NGROK_AUTHTOKEN in the .env file.
Set Your ngrok Static Domain
- Both WEBHOOK_URL and VUE_APP_URL_BASE_API should be set to your ngrok static domain, e.g., https://yourname.ngrok.io.

Example .env File:

POSTGRES_USER=root
POSTGRES_PASSWORD=your-super-secret-password
POSTGRES_DB=n8n

N8N_ENCRYPTION_KEY=your-n8n-encryption-key
N8N_USER_MANAGEMENT_JWT_SECRET=your-super-secret-jwt-token

N8N_AI_OPENAI_API_KEY=your-open-ai-key #if you want to use n8n ai for code nodes etc

NGROK_AUTHTOKEN=your-ngrok-auth-token
WEBHOOK_URL=https://your-nrok-url
N8N_HOST=your-nrok-url

1.5 Start the n8n Environment

Navigate to the Directory Containing docker-compose.yml
- Ensure you're in the n8n-ngrok directory.
Start All Services
- In the terminal, run:
```
docker-compose up -d
```
- This will start all the services defined in docker-compose.yml.
Check the Logs
- To ensure everything is running smoothly, check the logs:
```
docker-compose logs -f
```
- Press Ctrl + C to exit the logs view.
Access the n8n Editor
- Open your web browser and navigate to your ngrok static domain URL, e.g., https://yourname.ngrok.io.
- You should see the n8n editor interface.

💡

📝 Note: If you encounter issues during setup, verify your environment variables and service status. You can also check the Docker Desktop application to see if the containers are running.

2. Importing the Workflow Template

Now that n8n is up and running, we'll import a pre-built workflow template that automates email categorisation using AI. This template will be the foundation for the next steps.

2.1 Download the Template

Download the template directly from this link.

2.2 Import the Template into n8n

In the n8n editor, click on the menu icon (three horizontal lines) in the top corner.
Select Import.
Choose From File and select the downloaded template file.
The workflow will be imported into your n8n instance.

2.3 Review the Workflow

Take a moment to explore the imported workflow.
You'll see various nodes configured to fetch emails, process them with AI, and update their categories.
Don't worry if it looks complex; we'll guide you through configuring it to suit your setup.

3. Setting Up Ollama for AI Integration

🎥 Watch this section 04:22 - 06:50

To enable AI-powered email categorisation, we'll set up Ollama, which allows us to run local AI models.

3.1 Install Ollama for Windows

Ollama can be installed in multiple ways, but we'll focus on installing it directly on Windows.

Download and Install Ollama

Visit the Ollama website and download the Windows installer.
Run the installer and follow the on-screen instructions.

3.2 Download an AI Model

Depending on your system's resources, you can choose different AI models.

Select an AI Model

For powerful systems:
- Gwen 2.5 (14B model) is recommended for better results.
For less powerful systems:
- Llama 3.2 7B (7B model) or Llama 3.2 13B (13B model) can be used.

Pull the Model

Open Command Prompt or PowerShell and run:

ollama pull gwen2.5-14b

Replace gwen2.5-14b with the model you choose.

3.3 Verify Ollama is Running

Ollama runs as a background service after installation.

Ensure it's running by checking your system tray or services.
To list available models, run:

ollama list

You should see the model you pulled listed.

💡

Note: Downloading larger models may take time and require more disk space.

4. Configuring the Workflow for Ollama

Now we'll connect the imported workflow to Ollama.

4.1 Update the AI Agent Node

In the n8n editor, open the imported workflow.
Locate the AI Agent node.
Double-click the node to open its settings.

4.2 Connect to Ollama

Under Model Provider, select Ollama.
Host: Enter localhost.
- If you're running n8n in a virtual machine or WSL, you might need to use your machine's IP address.
  - To find your IP address, open Command Prompt and run ipconfig, then note your IPv4 Address.
Port: Use the default port 11434.
Click Refresh Model List.
Select the model you pulled earlier from the Model dropdown.

4.3 Save and Test the Node

Click Save to apply the changes.
Optionally, you can execute the node to test the connection.
- The first response may take longer as the model loads into memory.

5. Setting Up Microsoft Azure for Outlook Integration

🎥 Watch this section 09:18 - 12:37

To interact with Outlook and automate email categorisation, we'll set up an application in Microsoft Azure.

5.1 Register a New Application in Azure

Access Microsoft Azure Portal

Go to the Azure Portal and log in with your Microsoft account.

Navigate to App Registrations

In the search bar at the top, type App registrations and select it.

Register a New Application

Click + New registration.
Fill in the details:
- Name: n8n Outlook.
- Supported account types: Choose Accounts in any organizational directory and personal Microsoft accounts.
- Redirect URI: Set it to your n8n OAuth callback URL, e.g., https://yourname.ngrok.io/rest/oauth2-credential/callback.

5.2 Configure API Permissions

In your newly registered app, set the necessary permissions.

Add Permissions

Click on API permissions in the left-hand menu.
Click + Add a permission.
Choose Microsoft Graph > Delegated permissions.
Select the following permissions:

Permission	Type	Description
`Mail.Read`	Delegated	Read user mail
`Mail.ReadWrite`	Delegated	Read and write mail
`Mail.Send`	Delegated	Send mail
`User.Read`	Delegated	Read user profile

Click Add permissions.

Grant Admin Consent

Click Grant admin consent for your tenant.
Confirm to grant permissions.

5.3 Create a Client Secret

In the left-hand menu, click on Certificates & secrets.
Click + New client secret.
Add a description (e.g., n8n Secret) and set an expiration.
Click Add.
Copy the Value of the client secret. This is only shown once, so save it securely.

6. Connecting n8n to Microsoft Outlook

Now we'll set up the credentials in n8n to connect to Outlook.

6.1 Add Microsoft Outlook Credentials

In the n8n editor, select an Outlok node and click on Credentials.
Click + New and select Microsoft Outlook.
Fill in the configuration details:
- Credential Name: Outlook OAuth.
- Client ID: Paste your Azure app's Application (client) ID.
- Client Secret: Paste the client secret you generated.
- OAuth Callback URL: It should match the Redirect URI you set in Azure.
  - Typically, this is https://yourname.ngrok.io/rest/oauth2-credential/callback.

6.2 Connect and Authorize

Click Connect.
You'll be redirected to log in to your Microsoft account and grant permissions.
After successful authentication, your credentials should be saved.

7. Configuring the Workflow for Outlook

7.1 Update Outlook Nodes

In your workflow, locate all the Microsoft Outlook nodes.
For each node:
- Double-click to open the node settings.
- Under Credentials, select the Outlook OAuth credentials you just created.
- Verify other settings like Operation, Folder, and Fields as per your requirements.

7.2 Adjust Node Parameters

Get Emails Node

Operation: Get Many
Limit: Set to a number you're comfortable with (e.g., 10).
Fields: Select fields like flag, from, importance, replyTo, sender, subject, toRecipients, body, categories, isRead.

Apply Filters

To fetch only relevant emails, use the following custom filter in the Filters section:
```
flag/flagStatus eq 'notFlagged' and not categories/any()
```
Folder: Set to your Inbox or another folder as needed.

8. Customising Categories

🎥 Watch this section 15:51 - 16:26

The workflow includes predefined categories for email classification. You can customize these to suit your needs.

8.1 Update Categories in the Workflow

Locate the Category Node

Find the Set node named varID & Category or similar.
This node sets the categories used in the AI Agent and subsequent nodes.

Edit the Categories

Double-click the node to open its settings.
Find the field where categories are defined, typically a string or an array.

Update the categories to reflect those you want to use.

"action", "junk", "receipt", "SaaS", "community", "business", "other"

You can add or remove categories as needed.

8.2 Reflect Changes in the AI Agent Node

Open the AI Agent node settings.
Ensure the System Message and Prompt reflect the categories you've defined.
Update any references to categories to match your customized list.

8.3 Update the Switch Node

Locate the Switch node that routes emails based on the category.
Add or modify rules to handle your new categories.
- For each category, set a condition:
```
{{$json["output"]["category"]}} equals "YourCategory"
```
Ensure that all categories are accounted for, and add a fallback route if necessary.

8.4 Adjust Outlook Update Nodes

For each category, there may be corresponding Microsoft Outlook nodes to update or move emails.
Ensure these nodes are correctly configured to handle your customized categories.
Update folder selections or actions as needed.

9. Finalising and Testing the Workflow

Now that all configurations are in place, it's time to test the workflow.

9.1 Run the Workflow Manually

Set the Trigger node to Manual mode for testing.
Click Execute Workflow at the top-right corner.

9.2 Monitor the Workflow Execution

As the workflow runs, click on each node to view its input and output data.
Pay special attention to the AI Agent node to see how emails are categorised.
Check for any errors indicated by red nodes.

9.3 Verify in Outlook

Log in to your Outlook account.
Check if emails have been categorised and moved according to the workflow logic.

9.4 Troubleshooting

If you encounter issues:
- Ensure all nodes have the correct credentials selected.
- Double-check the API permissions in Azure.
- Verify that Ollama is running and accessible.
- Review error messages for guidance on what needs adjustment.

10. Automating the Workflow

Once you're satisfied with the workflow, you can set it to run automatically.

10.1 Change the Trigger Node

🎥 Watch this section 25:32 - 26:06

Replace the Manual Trigger node with an Outlook Trigger node.
Set it to On Message Received.
Configure any filters or options as needed to suit your requirements.

10.2 Save and Activate the Workflow

Save the workflow.
Toggle the Active switch at the top-right corner to activate the workflow.

11. Tips and Best Practices

11.1 Adjusting AI Prompts (Optional)

If you wish to customize how the AI categorises emails, you can adjust the System Message and Prompt in the AI Agent node.
- Keep prompts clear and concise for better performance, especially with smaller models.
- Ensure that the output format is maintained to prevent errors.

11.2 Error Handling

🎥 Watch this section 20:04 - 21:08

Implement error handling to ensure the workflow runs smoothly even if issues occur.
Use a Set node or Function node to sanitize and validate the AI's output.
- For example, use regular expressions to clean the JSON output if the AI includes unwanted characters.
Enable Continue On Fail in critical nodes to prevent the entire workflow from stopping due to a single error.

11.3 Managing System Resources

Model Selection: Choose an AI model appropriate for your system's capabilities.
- Larger models offer better accuracy but require more resources.
Batch Processing: Adjust batch sizes in the Get Emails node to prevent overloading your system.
- Start with smaller batches and increase gradually as you monitor performance.

11.4 Regularly Update and Monitor

Keep your n8n instance and Ollama models up to date for the latest features and security patches.
Monitor the workflow's performance and accuracy over time.
- Adjust AI prompts and categorization logic as needed based on email content changes.

Conclusion

Congratulations! You've successfully set up an advanced email categorisation system using n8n and AI. Here's what you've accomplished:

Set up a local n8n instance with ngrok for secure, public accessibility.
Imported and configured a workflow template for AI-powered email categorisation.
Installed Ollama and integrated an AI model for processing emails.
Configured Microsoft Azure and connected n8n to Outlook.
Customized the workflow to suit your categorization needs.
Tested and automated the workflow for continuous operation.