3. Foundry Primer

Generative AI & Agents

Generative AI models can produce text, code, images, and audio. Agents extend this by adding tool use, memory, and autonomous decision-making. Foundry provides a managed service for both.

SDK Overview

SDK	Package	Purpose
Projects	`azure-ai-projects`	Agent lifecycle, tools, memory
Inference	`azure-ai-inference`	LLM chat completions, embeddings
Search	`azure-search-documents`	Vector + keyword search
Identity	`azure-identity`	Authentication

Hub vs Standalone Projects

Hub — Shared AI Search, Storage, Key Vault, and monitoring. Best for teams.
Standalone Project — Independent resources. Best for experiments and individual dev.

Deploying Models in Azure

Open the Model Catalog in Foundry
Select a model (e.g., gpt-4o, Llama-3.1-70b)
Choose deployment type (Global Standard, Provisioned Managed)
Configure capacity and content filters
Deploy and get the endpoint URL + key

Demo Repositories

Microsoft provides starter repos on GitHub:

Azure-Samples/azure-ai-agent-sdk — Agent SDK samples
Azure-Samples/azure-ai-search-rag — RAG pipeline samples
Azure-Samples/azure-ai-gateway — AI Gateway samples

✏️ Exercise: Deploy a gpt-4o model from the Foundry model catalog. Note the endpoint URL and authentication method. Write a Python script that sends a test prompt to the endpoint.