Microsoft Foundry: Your AI Command Center
Microsoft Foundry is where you deploy AI models, build agents, and manage your AI projects. This module walks you through the portal — from creating a project to deploying your first model.
What is Microsoft Foundry?
Microsoft Foundry is your AI workshop — it’s where you go to build, test, and deploy everything AI in Azure.
Think of it like a fully equipped workshop. The model catalog is your tool shelf (hundreds of AI models to choose from). The playground is your workbench (test models without writing code). The project is your workspace (keeps everything organised).
You don’t need to install anything — it’s a web portal at ai.azure.com. Sign in with your Azure account and start building.
The Foundry portal layout
When you sign in to ai.azure.com, you’ll see:
| Section | What It Contains |
|---|---|
| Home | Recent projects, quick actions, getting started guides |
| Model catalog | Browse and deploy hundreds of models (OpenAI, Microsoft, Meta, Mistral, etc.) |
| Playground | Chat with deployed models, test prompts, configure settings — no code needed |
| Build | Create agents, build apps, write and test prompts |
| Operate | Monitor deployments, view metrics, manage resources |
| Tools | Foundry Tools (AI services) — Speech, Vision, Language, Content Understanding |
Key concepts
Foundry resource
A Foundry resource is the top-level Azure resource that contains everything. Think of it as your AI headquarters.
Projects
A project is a workspace within a Foundry resource. Each project has:
- Its own deployed models
- Its own agents
- Its own API keys and endpoints
- Its own monitoring and logs
Priya scenario: Priya creates a Foundry resource for her university work. Inside it, she creates two projects: one for her coursework chatbot and one for her thesis research assistant.
Model deployments
A deployment makes a model available through an API endpoint. When you deploy GPT-4o, you get a URL that your applications can call to send prompts and receive responses.
Deploying your first model — step by step
Here’s how Priya deploys GPT-4o in the Foundry portal:
- Sign in to
ai.azure.com - Select a project (or create one)
- Go to Model catalog → search for “GPT-4o”
- Click Deploy → choose deployment type (Global Standard is easiest)
- Name your deployment (e.g., “gpt4o-coursework”)
- Set configuration — rate limits, content filters
- Deploy — within seconds, you have an API endpoint
The Playground: testing without code
The Playground is where you test your deployed model interactively:
| Playground Feature | What You Can Do |
|---|---|
| Chat | Have a conversation with the model, just like ChatGPT |
| System prompt | Set the system message (role, rules, tone) |
| Parameters | Adjust temperature, max tokens, top-p in real time |
| View code | See the equivalent Python/C#/JavaScript code for your chat |
| Compare models | Run the same prompt against different deployments side by side |
Key exam concept: The Playground is the fastest way to test prompts and parameters without writing any code. It generates the equivalent SDK code you can copy into your application.
Foundry evolution: what changed and why
Understanding the naming history helps on the exam:
| Previous Name | Current Name |
|---|---|
| Azure AI Studio | Microsoft Foundry (portal) |
| Azure AI Services | Foundry Tools |
| Azure OpenAI Service | Part of Foundry (OpenAI models) |
| Hub + AI resource | Foundry resource + projects |
| Assistants API | Responses API |
The exam may still reference older terminology in some questions. Know that “Foundry” is the current umbrella term.
Foundry Tools (formerly Azure AI Services)
Foundry Tools are the specialized AI services available within Foundry:
| Tool | Capability |
|---|---|
| Azure AI Language | Text analysis — sentiment, entities, keywords, summarisation |
| Azure AI Speech | Speech-to-text, text-to-speech, translation |
| Azure AI Vision | Image analysis, OCR, face detection |
| Azure Content Understanding | Document, image, audio, video extraction |
| Azure OpenAI | GPT models, GPT-image, embeddings (model access, not a standalone tool) |
🎬 Video walkthrough
Flashcards
Knowledge Check
Priya wants to test how GPT-4o responds to different system prompts and temperature settings before writing any code. Which Foundry feature should she use?
DataFlow Corp needs to organise their AI work: one team builds customer support bots, another works on document processing. How should they structure their Foundry setup?
Next up: Building a Chat App with the Foundry SDK — writing your first Python code to interact with AI.