AI Platform: Service Description

This page describes what the AI Platform offers and where it stops. It is based on the Terraform modules of the solution. Commercial terms (price, hours, response times) are in your contract or onboarding document, not here.

What is offered

Component	What it is
Shared Foundry account	One Azure AI Services account (kind AIServices) per platform, with project management enabled. All projects live in it.
Foundry project per workload	An isolated project with its own managed identity, storage account, Application Insights, connections and Agents capability host.
Agent data stores	Azure Cosmos DB (agent threads and state) and Azure AI Search (vector store). Shared at platform level, connected to each project.
Compute and registry for workloads	The reference design contains an Azure Container Apps environment and an Azure Container Registry for the applications that call the agents.
Access through groups	Per project an owner, contributor and reader group, with roles assigned on the project and its data stores.
Workload identity	A user-assigned managed identity per application, with a list of role assignments.
Lookups	Data-only modules so that workload code finds the platform resources and the project without hard-coded IDs.

The component list follows the design of the solution. The modules of this repository create the Foundry account, the project and its resources, and the workload identity. The Container Apps environment, registry, Cosmos DB account, AI Search service, Key Vault and Log Analytics workspace are looked up by the modules, so they exist before a project is created. Which of them your platform runs, and how they are deployed, is platform specific.

Not included

• Model deployments. Models are deployed at platform level and are not part of the project module. This keeps quota management central and consistent across projects. A project module call does not create a model deployment.
• Application code and agents. The agents, prompts, tools and applications are yours.
• Content and data governance. What data you send to a model, retention of conversations and the rules for using AI are decisions of your organisation.
• A deployment pipeline. The source repository delivers modules. Environment-specific deployment code is planned for a later phase (see Known issues).

Service model

	Platform team	Workload team
Foundry account, shared data stores, network integration	Yes
Create, change and remove a Foundry project	Yes, after a request	Requests
Access groups and role assignments of a project	Yes	Owners of the groups decide the members
Models available in the account	Yes	Asks for a model or quota
Agents, applications, prompts, data in the stores		Yes

Details are on Responsibilities. The platform team is XMMS in the managed-service model and your own team in the self-managed model: Operating models.

Constraints from the source

Constraint	Value
Regions in the reference design	westeurope, swedencentral
Environments	development, test, acceptance, production, nonprod
Workload (project) name	Maximum 8 characters
Tags on all resources	Environment and Workload

These are conventions of the source repository for the planned deployment code. The modules themselves do not enforce them with validation rules. Ask your platform team which values apply on your platform. Generic naming and tags are on Naming and tags.

How to request

Send a request through the channel of your platform (<support channel>), see How to request. For a new project, name the workload, the environment, the region and the members of the owner, contributor and reader groups. The exact intake form for the AI Platform is not defined in the sources.