Ship an AI agent without leaving your terminal
Microsoft Developer demonstrates an end-to-end workflow for building and shipping a hosted AI agent from the terminal using GitHub Copilot CLI and Microsoft Foundry, including planning, resource creation, model selection, evaluation-driven improvements, and a secure Azure Functions proxy to connect the agent to a website chat experience.
Overview
The video walks through going from an idea to a live, hosted AI agent without leaving the terminal. The scenario is a “Backer Concierge” for Tailspin Toys that:
- Recommends games
- Remembers the conversation (conversation memory)
- Answers only from the product catalog (catalog-grounded responses / RAG-style constraint)
Key links shared in the description:
Build an AI Game Concierge
The presenter builds an AI concierge agent for a toy/game catalog, focusing on a production-style workflow where the agent is hosted and then connected to a website.
Define the agent’s trust boundary
The workflow explicitly defines what the agent is allowed to answer from and how it should be constrained:
- The agent should answer only from the catalog
- The solution includes a proxy layer intended to keep credentials safe
Plan with GitHub Copilot CLI
GitHub Copilot CLI is used from the terminal to:
- Plan the work
- Scaffold the agent
- Propose a model choice
- Iterate on fixes based on evaluation results
The workflow is approval-driven: the developer approves each step before changes are applied.
Create Microsoft Foundry resources
The video shows creating the required Microsoft Foundry resources from the terminal, using Foundry and Azure skills via GitHub Copilot CLI.
Select and deploy a model
Copilot proposes a model, and the workflow includes selecting and deploying that model as part of the agent build-out.
Build the hosted agent
The agent is built as a hosted agent in Microsoft Foundry, with the concierge behavior and conversation memory.
Evaluate and improve responses
An evaluation set is used to catch issues in the agent’s responses. Copilot then helps fix what the evals detect, and the agent is iterated until it meets expectations.
Deploy the agent to Foundry
After improvements, the agent is deployed as a hosted agent in Foundry.
Build a secure Azure Functions proxy
The solution includes an Azure Functions-based proxy layer designed to:
- Keep credentials out of the client
- Provide a safer boundary between the website and the hosted agent
Add the website chat widget
The hosted agent is wired into a website via a chat widget, using the Functions proxy as the backend connection point.
Review the complete solution
The video ends with a review of the full end-to-end setup:
- Terminal-driven planning and scaffolding with GitHub Copilot CLI
- Foundry resource creation and model deployment
- Hosted agent build and deployment
- Evaluation-driven iteration
- Secure proxy with Azure Functions
- Website chat widget integration