Start with the words you need
Foundry uses a handful of terms again and again. Learn them first, because the rest of the course builds on them.
| Term | What it means |
|---|---|
| Model | An AI model that generates output from a prompt, such as gpt-5-mini. Foundry gives you access to many models from its catalog. |
| Model deployment | Making a model available to your code or an agent. Each deployment has a name, and requests use that name. |
| Agent | A model plus instructions, tools, and prompts, set up to handle a task across a conversation. The instructions stay with the agent, so you don't repeat them every time. |
| Tool | Something an agent can call, such as web search, file search, a code interpreter, or a function you write. |
| Foundry resource | The top-level Azure resource. It owns the model deployments, security, networking, connections, and billing. |
| Foundry project | A workspace inside a resource. A project holds the agents, evaluations, and files for one use case or team. |
| Project endpoint | The web address your code calls to reach a project's agents and models. You copy it from the Foundry portal. |
What Foundry is
Microsoft Learn describes Microsoft Foundry as one management plane for agents, models, and tools, with observability, governance, and security built in. In practice, the resource sets the rules, and projects are where people work.
Storage, Key Vault, and Azure AI Search are separate Azure resources that Foundry connects to. They keep their own access rules, so you manage them alongside Foundry rather than inside it.
The problem it solves
Take an order-support agent as an example. It needs a model, instructions, a tool that looks up orders, a way to test answers before customers see them, logs for when something goes wrong, and access control so only the right people can change it. Without a platform, a team connects five or more separate services. Foundry puts those pieces in one place.
Microsoft's Ignite 2025 announcements make the same point. They say developers need to "design, test, monitor, and improve intelligent systems with the same confidence they have in traditional software." That is Microsoft's view, so treat it as a claim. The rest of the course shows the features behind it.
What it costs to start
The Foundry platform is free to use and explore. Each service is billed at its own rate, so costs depend on the models and services you use. You need an Azure subscription before you build agents, though not to explore.
- To sign up for an Azure account, you need a phone number, a credit or debit card that is not prepaid, and a Microsoft or GitHub account.
- Instant access lets you try a supported model without creating a deployment. It is in preview and limited to some regions, so check the portal.
- Use the pricing links below and the Azure pricing calculator to estimate costs before you build.
- You need an Azure subscription before you build agents, though not to explore. Create a free or pay-as-you-go account: Create an Azure account
- Create a Foundry project and deploy a model by following the quickstart: Create Foundry resources
- If you write code, install the SDK and authenticate by following: Get started with the SDK
- Platform and pricing overview: Microsoft Foundry pricing
- Models: Foundry Models pricing
- Agents: Foundry Agent Service pricing
- Estimate your costs: Azure pricing calculator
How many models? Read the number carefully
Microsoft's pricing page and the Ignite post say more than 11,000 models. The Learn overview says more than 10,000. The catalog keeps growing, so the exact figure matters less than the practical point: you choose a model for each task, and you can compare models before you commit.