How to build and ship an AI agent in the GitHub Copilot app
GitHub walks through building and deploying a production-ready AI agent directly inside the GitHub Copilot app, using Azure AI Foundry features (Foundry Skill and Foundry Canvas) to provision Azure resources, deploy a model, test grounding to reduce hallucinations, and ship the agent with a secure proxy and website chat widget.
Overview
The video shows an end-to-end, in-app workflow for creating a grounded “backer concierge” agent for the Tailspin Toys scenario, starting from a GitHub issue prompt and finishing with a deployed agent.
What the walkthrough covers
- Preparing the data used to ground the agent
- Creating required Azure resources from within the GitHub Copilot app
- Selecting and deploying a model
- Building a hosted agent
- Deploying the agent to Azure AI Foundry
- Building a secure Azure Functions-based proxy to protect credentials
- Adding a website chat widget
- Testing the agent, including checks aimed at reducing hallucinations
Self-paced workshop
- Workshop link: https://github-samples.github.io/copilot-workshops/real-world-development/app/8-foundry-canvas/
Video chapters
- 0:00 Introduction
- 0:58 Prepare the data
- 1:39 Create Azure resources
- 2:04 Select and deploy a model
- 5:08 Build the hosted agent
- 5:51 Deploy the agent to Foundry
- 7:45 Build a secure functions proxy
- 8:11 Add the website chat widget
- 8:28 Test the backer concierge agent