In this video, April introduces a different approach in the “Build an Agent” series by demonstrating how to use Azure AI Foundry models with GitHub Copilot instead of building an agent from scratch. She begins by guiding viewers through the initial setup in the Azure AI Foundry portal, where she creates a new project. This involves selecting the Azure AI Foundry resource type, naming the project, choosing the subscription, resource group, and region, and then creating the project. Once the project is created, the portal displays all relevant project information.
Next, April moves to Visual Studio Code, where she installs the AI Toolkit extension, which includes the Azure AI Foundry extension. She explains how to access the AI Toolkit view and its various sections, such as resources, model tools, and agents. Additionally, she installs the Azure Resources extension to sign into Azure, enabling access to the Azure AI Foundry project. After signing in, she sets the default project for the AI Toolkit by selecting the Azure AI Foundry project created earlier, allowing the extension to load the relevant models.
April then demonstrates how to deploy models for use with GitHub Copilot. She filters the model catalog to show Azure AI Foundry models and selects GPT-40 for deployment. She walks through the deployment settings, including deployment name, type, version upgrade policy, model version, tokens per minute, and content filter. After confirming the deployment, the model becomes available in the AI Toolkit. She notes that models can also be deployed directly from the Azure AI Foundry portal but recommends using the AI Toolkit for ease.
With the model deployed, April shows how to integrate it with GitHub Copilot Chat in Visual Studio Code. She explains how to manage models by selecting Azure AI Foundry via AI Toolkit as the provider and adding the deployed GPT-40 model to the list of available models. She then switches to ask mode and submits a prompt about creating a virtual environment in Python, receiving a helpful response from GitHub Copilot powered by the Azure AI Foundry model.
Finally, April covers the option to use local models with GitHub Copilot. She filters the model catalog to find Foundry local models and explains the process of downloading and adding a local model, such as the 5.4 generic CPU model, to the AI Toolkit. She demonstrates selecting the local model in GitHub Copilot and submitting a prompt, noting that response times depend heavily on the hardware’s performance. She concludes by encouraging viewers to explore other videos in the series for agent creation and hints at upcoming AI Toolkit features.
