Skip to main content

Azure GPT: Integrate Azure OpenAI with LoyJoy

For your own Azure OpenAI deployment, you need an existing Azure enterprise account that you have access to. If your company already has a normal Azure account, you can simply use it for this guide.

This guide will do following steps:

  • In your Azure account create Azure OpenAI.
  • In Azure OpenAI add models for gpt-4o, gpt-4o-mini and text-embedding-ada-002.

Step 1: Create Azure OpenAI

  • In Azure type OpenAI into Azure search and select Azure OpenAI
  • Click button Azure OpenAI
  • The screen Create Azure OpenAI should show up.

Create Azure OpenAI Basics

  • In Azure OpenAI > Basics
    • Select a subscription and if necessary at Resource group click Create new with arbitrary name such as openai
    • Select a region. Sweden Central is recommended for Europe, as it has the newest models. European regions other than Sweden Central lack gpt-4o-mini, which is required when enabling reranking in GPT Knowledge. You can check the list of available regions for model availability.
    • In name enter e.g. companyname-openai.
    • As pricing select Standard S0.
    • Click Next
  • In Azure OpenAI > Network keep the preselected value All networks, including the internet, can access this resource. and click Next
  • In Azure OpenAI > Tags click Next
  • In Azure OpenAI > Review + submit click Create

Create Azure OpenAI Submit

  • Deployment is in progress will show up. Then Your deployment is complete will show up.
  • Click Go to resource to open the deployment.

Create Azure OpenAI Deployment

  • Under menu item Keys and Endpoint you find access keys and an endpoint URL. Copy this data to send it to LoyJoy support team later.

Step 2: Deploy Models

To use Azure OpenAI, now models have to be deployed.

  • In Azure OpenAI click Azure OpenAI Metrics Dashboard to open Azure AI Foundry | Azure OpenAI Service.
  • Azure OpenAI Service should open, optionally with a sign in screen. You might need to select your Azure subscription to proceed.
  • In Azure OpenAI Service open menu item Deployments.

Deploy Models

Create following models:

  • text-embedding-ada-002:
    • Click Deploy model and Deploy base model. Select text-embedding-ada-002 in version 2 and enter a name such as companyname-text-embedding-ada-002.
  • gpt-4o-mini:
    • Click Deploy model and Deploy base model. Select gpt-4o-mini in version Standard and enter a name such as companyname-gpt-4o-mini.
  • gpt-4o:
    • Click Deploy model and Deploy base model. Select gpt-4o in version Standard and enter a name such as companyname-gpt-4o.

Step 3: Content Filter

Microsoft Azure filters completion requests for inappropriate content by default. This might lead to error messages in your chatbot such as The response was filtered due to the prompt triggering Azure OpenAI's content management policy. Please modify your prompt and retry..

Create Content Filter

This filter can be lessened by configuring the Content Filter in Azure OpenAI Service. To do this, follow these steps:

  • In Azure OpenAI Service click Safety + security in the left menu.
  • Click Create content filter.
  • Enter a name such as CompanyNameContentFilter.
  • As input filter thresholds select High in all categories. A high threshold means that Azure will be less restrictive in filtering out inappropriate questions from customers.

Input Filter

  • As output filter thresholds select High in all categories. A high threshold means that Azure will be less restrictive in filtering out inappropriate answers generated by the AI.

Output Filter

  • Apply to all deployments.
  • Click Create filter.

Step 4: Configuration in LoyJoy

To use the deployment in LoyJoy please select an Azure model and enter the following details:

  • From step 3 one of the keys and the endpoint URL
  • The model deployment names, in this example companyname-text-embedding-ada-002 and companyname-gpt-4o and companyname-gpt-4o-mini.

LoyJoy Azure GPT settings

Step 5: Fill out form "Modified Content Filters and Abuse Monitoring"

Optionally, you can request that Microsoft internally should turn off human abuse monitoring based on Microsoft internal staff.

To request this fill out this form.