Hi Rob; Let’s see what we can do for you. There is also this CloudBank Community entry from a couple years back in 2023 that points to a video resource.
Here first is a list from Microsoft Learn docs; and I follow this up with some terminology and a simple version of “how to get there”.
- Create and deploy an Azure OpenAI resource in AI Foundry — step-by-step for standing up the resource and deploying your first model, via portal, CLI, or PowerShell. This is the “how do I actually get access” starting point. How-to: Create and deploy an Azure OpenAI in Microsoft Foundry Models resource (classic) - Microsoft Foundry (classic) portal | Microsoft Learn
- Microsoft Foundry architecture overview — explains the layered model (Foundry resource for governance → projects for isolation → connected Azure services), which is the mental model for organizing work and access. Microsoft Foundry architecture - Microsoft Foundry | Microsoft Learn
- Create a Foundry resource (multi-service) — describes how a single resource brings agents, model deployments, and Foundry Tools together behind one endpoint and key, with projects for separating use cases. Create a Foundry resource - Foundry Tools | Microsoft Learn
- Upgrade Azure OpenAI to Microsoft Foundry — useful for understanding the current rebrand and how an existing Azure OpenAI endpoint maps into Foundry without rebuilding. Upgrade Azure OpenAI to Microsoft Foundry - Microsoft Foundry | Microsoft Learn
- AI and Machine Learning products (Azure Architecture Center) — compares Azure’s AI/ML platforms by audience and workload, which helps distinguish the agent-assistant path (AI Foundry) from building ML/data pipelines (Azure Machine Learning). AI and Machine Learning Products - Azure Architecture Center | Microsoft Learn
Terms:
- My terminology: Azure Object – an abstract container used to tie related services together as in a bundle. Main example: Resource Group. OpenAI example: Foundry Resource.
- Azure OpenAI — Microsoft’s hosted version of OpenAI’s models (GPT-4o, o-series, etc.), running inside Azure’s security, compliance, and billing perimeter rather than on OpenAI’s own servers.
- AI Foundry — the umbrella workbench platform (portal + APIs + SDKs): Get models, build agents, bring in data.
- Foundry resource — the top-level Azure Object: Governance and billing. Holds model deployments, agents, connections. One endpoint, one key.
- Project — a sub-container / workspace inside a Foundry resource that isolates use cases
- Deployment — We don’t call models in Azure directly; we create named “deployments” (like instances) of some model and our code calls that name.
- SDK — a Software Development Kit: the code library you install to call the service from Python or whatever
Access to Azure OpenAI:
-
Have an Azure subscription: check!
-
Create a Foundry resource in a region that hosts the models you want (model availability varies by region). 3 modes of access:
- The Azure portal (point-and-click) (CloudBank suggests use this 1 or 0 times; then look into the following options…)
- Azure CLI / PowerShell / ARM templates (scripted, repeatable, reversible)
- Dive into the AI Foundry portal
-
Deploy a model into that resource — pick a model, name it, choose deployment type (Standard, Global Standard, or Provisioned Throughput).
-
Get your endpoint URL + key (see also the Azure service for this: Microsoft Entra ID: keyless auth via managed identities/roles).
-
Call the model from code.
Microsoft is standardizing on the OpenAI-compatible /v1 API and the stable OpenAI SDK — you point the standard OpenAI client at your Azure endpoint.
(Do not use the deprecated “Azure AI Inference beta SDK”)