Microsoft Foundry, set up to scale past the first project
Formerly Azure AI Foundry. We design the projects, access, network boundaries, evaluation, and cost controls before anything ships, so your second use case builds on the first instead of starting over.
Talk to an Expert →Foundry is where AI prototypes become systems
Microsoft Foundry, the platform Microsoft called Azure AI Foundry until its latest rename, brings Microsoft's AI capabilities under one resource and one governance model. It gives you a managed runtime that connects models, tools, and enterprise data, a model catalog to test and fine-tune against, and built-in role-based access, telemetry, and logging.
The platform does a lot for you. It won't decide your project structure, your data boundaries, or which model fits the job. Those choices get made in week one, and they're the ones you live with in year two.
What we set up
Structure and models
- Project and resource structure: how projects map to teams, environments, and subscriptions, designed before anything is provisioned.
- Model selection: matched to the job, with region availability and preview status confirmed instead of assumed.
Access and boundaries
- Identity: role-based access over who can create, run, and change agents, tied to the Entra ID groups you already maintain.
- Network and data: virtual network isolation, data residency, and your own storage, Azure AI Search, and Cosmos DB where compliance requires it.
Safety, quality, and cost
- Content safety: filters set to your risk profile, including prompt injection and cross-prompt injection protection.
- Evaluation: intent resolution, tool call accuracy, and task adherence measured in your pipeline.
- Cost controls: budgets, alerts, and per-project attribution before the first production workload.
Our platform breakdown covers the capabilities in more depth: Azure AI Foundry, the next frontier for enterprise AI agents.
Agent Framework and MCP change what's practical to build
Microsoft Agent Framework brings the AutoGen and Semantic Kernel lines together into one SDK and runtime for multi-agent systems. Agents get built and tested locally, then deployed into Foundry with observability and orchestration intact.
Model Context Protocol (MCP) is an open standard that lets agents call external APIs, functions, and data sources without a custom integration for each one. In practice, that means an agent can reach real business systems.
Together they turn Foundry into a multi-agent platform. They also widen what an agent can touch, so access scoping and logging come first. The workflow design side lives on our agentic AI on Azure page.
Our take: get the data right, then build in Foundry
Most teams ask for Foundry when what they want is an outcome: a service desk that answers from real data, or a process that runs without someone copying between systems. The platform is the easy decision. Whether the data behind it is clean, reachable, and permissioned correctly decides whether the project works.
So we work in a sequence: get the data into shape, build the agent in Foundry, then secure and govern how people use it. When the data isn't ready, we say so up front, because an agent built on bad data is a faster way to be wrong. Our guide to assessing AI data readiness before you build covers the checks.
What we tell clients before they commit
Some Foundry components are still in preview, including parts of multi-agent orchestration and several third-party integrations. Running production work on a preview feature is a decision, and it should be a deliberate one with a fallback.
Team readiness is the other honest constraint. Running agents at scale takes skills in orchestration, observability, and model lifecycle that most infrastructure teams haven't built yet. We'll show you where that gap sits and what it takes to close it, whether that means training your people or embedding one of our Resident Senior Architects.
Why CloudServus
- Azure Expert MSP. Microsoft independently audits how we run environments, and fewer than 1% of partners hold it.
- Solutions Partner designations in Data & AI, Infrastructure, and Digital & App Innovation on Azure.
- 1,600+ engagements delivered, with Azure infrastructure and data and AI work making up the largest share of our live pipeline.
- We run it after we build it. The same team that provisions Foundry can manage the environment underneath it.
Frequently asked questions about Microsoft Foundry
What is Microsoft Foundry?
Microsoft Foundry is Microsoft's platform for building and running AI applications and agents. It was previously called Azure AI Foundry, and before that Azure AI Studio. It combines a model catalog, a managed agent runtime, tool and data connections, and governance features such as role-based access control, telemetry, and logging under one resource model.
Is Azure AI Foundry the same as Microsoft Foundry?
Yes. Microsoft renamed Azure AI Foundry to Microsoft Foundry. The documentation, portal, and role names have moved to the new name, and the Foundry roles that used to be called Azure AI User and Azure AI Owner are now Foundry User and Foundry Owner. Existing projects keep working. The rename doesn't change what you've already built.
How is Foundry different from Azure OpenAI Service?
Azure OpenAI Service gives you model endpoints. Foundry gives you the platform around them: agent orchestration, tool connections, evaluation, content safety, network isolation, and governance in one place. If you only need to call a model from an application you already control, the service may be enough. If you're building agents that act across systems and need an audit trail, Foundry is the right home.
Do we need Foundry to deploy Microsoft 365 Copilot?
No. Microsoft 365 Copilot works across your Microsoft 365 data with no development and doesn't require Foundry. Foundry is where you build the AI capabilities that don't exist in a Microsoft product, against your own systems and logic. Many organizations run both. See our Microsoft Copilot consulting page for the Copilot side.
What does Microsoft Foundry cost?
Foundry is billed through the Azure resources it consumes, including model inference, storage, search, and agent operations. Cost follows usage rather than a fixed platform fee, so it scales with how much your agents do. We model expected consumption during design and set budget alerts and per-project attribution during setup.
Can Foundry run in a regulated environment?
Yes. Virtual network isolation, data residency controls, and the option to bring your own storage and databases keep traffic and data inside your boundaries. The right setup depends on your specific obligations, which we work through during design rather than assume.
Ready to set up Foundry the right way?
No slide decks. Senior Microsoft engineers, real numbers, and a platform your second project can build on.
Talk to an Expert →