This guide explains how to deploy a chat application directly from the Azure AI Foundry playground to Azure App Service.
Azure AI Foundry provides a built-in capability to publish playground experiences as web applications. This accelerator deploys the required infrastructure (App Service, managed identity, networking) so you can publish directly from the Foundry playground.
- Completed deployment of this accelerator (
azd up) - Access to the AI Foundry project via the Jump VM
- An AI Search index with your data (created via OneLake indexer or manually)
Since all resources are deployed with private endpoints, you must access AI Foundry through the Jump VM:
- Go to the Azure Portal
- Navigate to your resource group
- Select the Jump VM (Windows Virtual Machine)
- Click Connect → Bastion
- Enter the VM credentials (set during deployment)
- Once connected, open a browser and navigate to AI Foundry
- In AI Foundry, select your Project
- Navigate to Playgrounds → Chat playground
- Configure your deployment:
- Select your GPT model deployment (e.g., gpt-4o)
- Add your AI Search index as a data source
- Configure the system prompt for your use case
- Test the chat experience in the playground
- Verify responses are grounded in your indexed data
- Adjust system prompts and parameters as needed
- Click Deploy → Deploy to a web app
- Configure deployment options:
- Create new or Update existing web app
- Select your Subscription and Resource group
- Choose the App Service deployed by this accelerator (if updating)
- Review authentication settings (Entra ID is recommended)
- Click Deploy
After deployment completes:
- Navigate to the App Service in Azure Portal
- Copy the Default domain URL
- Access your application (authentication may be required)
If your App Service is deployed with private endpoints, you'll need to access it through the Jump VM or configure Azure Front Door for public access.
Ensure the App Service has the correct managed identity permissions to access:
- Azure OpenAI / AI Services
- Azure AI Search
- Azure Storage (if applicable)
Verify that:
- Your AI Search index contains data
- The playground is configured to use the correct index
- The deployed app has the same index configuration