Skip to content

AI Platform

Teams that build AI applications and agents on Azure need the same building blocks every time: a model endpoint, a place where agents keep their conversations and files, a vector store, monitoring, and access that is controlled through groups. The AI Platform is an optional solution of the platform that provides these building blocks on Azure AI Foundry, in a private network, and gives every workload its own isolated Foundry project.

Optional offering, documented from the Terraform modules

The AI Platform is not part of the core MCP hub and spoke. Whether it is available and how it is set up depends on your platform. Ask your platform team. This documentation is based on the Terraform modules of the solution. The modules are the delivered part of the solution. The environment-specific deployment code (the "deploy roots") is described in the source as planned for a later phase, so the pages describe the reference design: your deployment may differ. See Known issues and open points.

What you get

Area What the solution does Page
Scope What is offered and what is not Service description
Design Shared account, projects, data stores, network and naming Architecture
Responsibilities Who runs what: platform team and workload team Responsibilities
Access Owner, contributor and reader groups per project, and the roles they get Access and RBAC
Building How a workload finds its project and gets its own identity Build on the platform
Operating The Terraform modules, inputs and outputs Platform modules
Open points Known issues and what has to be confirmed Known issues
Decisions The design choices behind the solution Decisions

Terminology

Term Meaning
Foundry account The shared Azure AI Services account (kind AIServices) of the platform. It is the parent of all projects.
Foundry project A workload-scoped namespace in the account, with its own identity, connections and data stores.
Capability host The project setting that tells Foundry to run agents and where agents keep their files, threads and vectors.
Workload identity A user-assigned managed identity of your application, with the roles it needs on the platform resources.
Spoke The Azure subscription of your workload, see What you get.

How it relates to the platform

The AI Platform is deployed in a spoke of the Managed Cloud Platform. It uses the spoke network, private endpoints, the Log Analytics workspace and the Key Vault of that spoke. Requests, changes and support follow the generic way of working of the platform: How to request and Support. Who the platform team is depends on the operating model: XMMS under a managed-service contract, or your own team when you run the platform yourself. See Operating models.