lagairogo-gai / openai-zendesk

A express app that Zendesk App Framework that uses GPT-3 to triage and summarize tickets.

OpenAI Zendesk Integration

An Node Express app for Zendesk that uses GPT-3 to triage and summarize tickets.

Features

🚦 Ticket Triage: Use GPT-3 to triage new incoming tickets to include a priority, summary note, and group assignee.

📝 Summarize Tickets: reduce long ticket discussions for use-cases like agent handoff.

Demo Video

Pre-Requisites

• Zendesk Support: You can create a free trial by following these instructions.
• Zendesk CLI: Follow these instructions.
• OpenAI account: You can create one here.
• Ngrok: If you're running the app locally, you'll need a tunnel service like Ngrok so that Zendesk webhooks can reach your localhost.

Running Locally

App Setup

Clone the sample app, install dependencies, and create a .env file to store API keys:

# Clone the repository:
$ git clone [repo name]

# Switch to the app's directory:
$ cd openai-zendesk

# Install dependencies:
$ npm install

# Copy the example .env file:
$ cp .env.example .env

Adding Secrets to your .env

Configure four values in your .env file:

• Zendesk Email: This can be your email or someone else's. You might want to create a separate user account named "Agent Assist Bot" so it's clear who is posting on the tickets.
• Zendek API Key: You can create one in the Zendesk Admin portal under Apps & Integrations > Zendesk API.
• Zendesk Remote URI: The API URL for your Zendesk instance. It will look like: https://{{your_custom_subdomain}}.zendesk.com/api/v2.
• OpenAI API Key: you can create one here.

Running Your App

When running locally, the app uses the Zendesk CLI to serve the front-end assets folder to an iframe in Zendesk. The express app runs as a back-end and connects Zendesk to the GPT-3 API.

Use NPM to start up your app backend like this:

# Restart automatically if you make changes:
$ npm run dev

# Just run the app:
$ npm run start

You can serve the app's front-end using the Zendesk CLI:

$ zcli apps:server

Now open a Zendesk Support ticket, adding ?zcli_apps=true to the end of the URL, for example: https://{{your_custom_subdomain}}.zendesk.com/agent/tickets/1?zcli_apps=true

If all goes well, you'll see the Ticket Summary interface loaded in the sidebar. If you click on the "Summarize Ticket" button a note should appear on the ticket summarizing the conversation.

Triaging New Tickets

The openai-zendesk integration can auto-triage tickets so that each ticket receives a priority, group assignee, and summary comment. This process has three steps:

1. Run Ngrok as a tunnel to your local environment
2. Create a Zendesk Webhook
3. Create a Zendesk Trigger that calls your webhook

To receive webhooks locally, you'll need to spin up Ngrok pointed at your express app:

$ ngrok http 3000

Once Ngrok is running, create a Zendesk Webhook:

• In Zendesk Admin go to Apps & Integrations > Webhooks.
• Click on Create Webhook and then click on Trigger or Automation.

---

---

• Give your webhook a name, and set {{your-ngrok-url}}/triage as the Endpoint URL.
• Then, click create webhook.

---

To create a Zendesk Trigger, follow these steps:

• Go to Zendesk Admin > Objects and Rules > Triggers > Click on Add Trigger
• Under conditions, determine when you want the trigger to fire. In our case, we fire the trigger when the ticket is new, has no priority, and doesn't have an owner.

---

• Under the Actions section, select Notify active webhook and then select the name of the webhook you created.
• Set the JSON body of the request to include the ticket id:

{
   "ticket_id": "{{ticket.id}}" 
}

---

To test our your webhook integration, create a new ticket. Your app should receive a webhook within 3 - 5 seconds triggering the ticket to be processed and triaged.