IntelligentHQ
IntelligentHQ
Skip to content
Platform Microsoft Foundry and AI services

Microsoft Foundry

Microsoft Foundry is Azure's platform for building generative AI apps and agents: a model catalog, an agent runtime, prebuilt AI tools, knowledge retrieval and governance under one Azure resource with shared access control, networking and policy. It was formerly Azure AI Studio, then Azure AI Foundry.1

Also known as Azure AI Foundry, Azure AI Studio, Foundry

Intended for
AI and application developers, Cloud and platform architects, Data scientists, IT and AI governance teams
Runs on
Azure public cloud regions (feature and model availability varies by region), Foundry Local for on-device and edge inference2
Official Microsoft product page (opens in a new tab) Official documentation (opens in a new tab)

At a glance

What is it?
Microsoft Foundry is a platform service on Azure that groups models, agents and tools under one management resource, the Foundry resource, which holds projects. Microsoft's Learn documentation lists the brand changes: Azure AI Studio and Azure AI Foundry became Microsoft Foundry, and Azure AI Services became Foundry Tools. Azure OpenAI is now offered as Azure OpenAI in Microsoft Foundry Models, and an Azure OpenAI resource can be upgraded to a Foundry resource while keeping its endpoint and keys. The product page lists the services built in: Foundry Models, Foundry Agent Service, Foundry Tools, Foundry IQ, Azure Machine Learning, Foundry Control Plane and Foundry Local. Older hub-based projects stay in a separate Foundry (classic) portal.123
What does it do?
Foundry lets a team browse and deploy models, build prompt agents in a portal or SDK, deploy hosted agents that run their own code in containers, attach tools and knowledge bases, and then trace, evaluate and monitor the result. Access, networking and policy are set once on the Foundry resource with Microsoft Entra ID, Azure RBAC, content filters, network isolation and Azure Policy. Developers work through the Foundry portal, SDKs for Python, C#, JavaScript and Java, the Azure Developer CLI, Visual Studio Code, or coding agents that use the Foundry skill and MCP server. Agents built in Foundry can be published to Microsoft Teams and Microsoft Copilot.14
Who needs it?
Development teams building generative AI applications or custom agents on Azure who want model choice, evaluation, safety tooling and enterprise controls in one place. Microsoft's architecture guide points to Foundry for generative AI and agent work, for comparing and evaluating models from several providers, and for fine-tuning and deploying foundation models with observability and content safety. Teams already on Azure OpenAI are a direct fit, because their resources can be upgraded to Foundry.5
What does it need?24
  • An Azure subscription (browsing Foundry needs no account; building agents does)
  • Role assignments on the Foundry resource and project; Microsoft recommends Microsoft Entra ID with RBAC over API keys for governed workloads
  • A region that offers the models and features you need
  • Python, C#, JavaScript or Java for SDK work, or the Azure Developer CLI for hosted agent projects
What it is not
Foundry is not a low-code business tool: Microsoft's adoption guidance positions Copilot Studio as the low-code SaaS route and Foundry as the PaaS route for pro-code and declarative agents. It is not a general training platform either; the architecture guide says Foundry does not offer broad provisioned training compute or AutoML for tabular data, which is the job of Azure Machine Learning. Foundry has no single price: each service inside it is billed on its own. And not everything is generally available: several Build and Operate features are still marked preview.256

Availability and licensing. Generally available: the new Foundry portal reached GA for Foundry projects (Learn GA overview, checked 26 Oct 2026). Several features remain preview, including monitoring, memory, agent guardrails, Voice Live agents, AI Gateway and the Operate pages; Foundry IQ is GA at API level and preview in the portal. Workflows retire on 1 December 2026. Pricing is consumption based, per service used (product page, checked 26 Oct 2026); Azure offers a free trial of up to 30 days.24

The problem it solves

Teams building with generative AI on Azure used to juggle several resource types: an Azure OpenAI resource, an Azure AI Services resource and a hub, each with its own endpoint, SDK package and access settings. Microsoft's own comparison table describes the older setup as several SDK packages against five or more endpoints.

Foundry consolidates this into one Foundry resource with projects, one project endpoint and one project client. Security, networking and policy are applied once for agents, models and tools together, and tracing, evaluation and monitoring are part of every project.1

How it works

Foundry is organized around a Foundry resource in an Azure subscription that contains one or more projects. Each project exposes one endpoint for models, agents and tools, used by the unified project client (azure-ai-projects 2.x) or an OpenAI client.

  • Models: deploy models from the Foundry Models catalog, or call supported models by name without a deployment (instant access, preview).
  • Agents: Foundry Agent Service runs prompt agents, voice-based prompt agents and hosted agents through the Responses API (Agents v2), which replaced the Assistants API.
  • Tools and knowledge: toolboxes expose built-in and custom tools through one MCP endpoint; Foundry IQ knowledge bases, built on Azure AI Search, ground agents in company data.
  • Operate: tracing, evaluations, red teaming and guardrails, with the Foundry control plane applying Entra identity, RBAC, content filters, network isolation and Azure Policy.

The new Foundry portal is generally available for Foundry projects. Standalone Azure OpenAI resources and hub-based projects are not supported there and continue in the Foundry (classic) portal.45

Microsoft Foundry architecture: components by layer and how they connectCopilot & businessappsAgents & automationAI models & platformData & analyticsDeveloper toolsSecurity, identity &governanceMicrosoft Teams and Microsoft Copilot: Where published agents reach usersMicrosoft Teams andMicrosoft CopilotFoundry Agent Service: Runs prompt, voice-based and hosted agentsFoundry Agent ServiceFoundry resource and project endpoint: Single endpoint and management boundary for models, agents and toolsFoundry resource and projectendpointFoundry Models: Catalog and deployments of Microsoft, OpenAI and partner modelsFoundry ModelsToolboxes and Foundry Tools: Built-in, MCP and prebuilt AI tools that agents callToolboxes and Foundry ToolsFoundry IQ (built on Azure AI Search): Permission-aware knowledge bases for groundingFoundry IQ (built on AzureAI Search)Foundry portal, SDKs and CLI: Where teams explore models, build agents and automate deploymentFoundry portal, SDKs and CLIFoundry control plane: Entra identity, RBAC, content filters, network isolation, policy and tracingFoundry control plane
Diagram as a list

  1. Copilot & business apps

    • Microsoft Teams and Microsoft CopilotWhere published agents reach users

  2. Agents & automation

    • Foundry Agent ServiceRuns prompt, voice-based and hosted agents

  3. AI models & platform

    • Foundry resource and project endpointSingle endpoint and management boundary for models, agents and tools
    • Foundry ModelsCatalog and deployments of Microsoft, OpenAI and partner models
    • Toolboxes and Foundry ToolsBuilt-in, MCP and prebuilt AI tools that agents call

  4. Data & analytics

    • Foundry IQ (built on Azure AI Search)Permission-aware knowledge bases for grounding

  5. Developer tools

    • Foundry portal, SDKs and CLIWhere teams explore models, build agents and automate deploymentConnects to Foundry resource and project endpoint

  6. Security, identity & governance

    • Foundry control planeEntra identity, RBAC, content filters, network isolation, policy and tracing
Components and connections as documented by Microsoft.

Capabilities

  • One resource and one project endpoint1

    A single Foundry resource with projects replaces the earlier hub plus Azure OpenAI plus Azure AI Services setup, with one project endpoint and one unified client.

    Why it matters: Less wiring and one place to set access, networking and policy for all AI assets.

    Limits: Hub-based projects and standalone Azure OpenAI resources are not supported in the new portal; they need migration or the classic portal.

  • Model catalog7

    Projects deploy and call models from the Foundry Models catalog, which Learn puts at more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others.

    Why it matters: Teams can compare providers without separate contracts and integrations.

    Limits: Partner and community models are supported by their providers and billed through Azure Marketplace.

  • Agents from declarative to code14

    Prompt agents are configured in the portal or SDK; voice-based prompt agents run spoken sessions through Voice Live; hosted agents run your own framework code in containers.

    Why it matters: Start without code and move to code when custom logic is needed.

    Limits: Voice Live agents are preview, and Foundry workflows retire on 1 December 2026.

  • Tools and knowledge4

    Toolboxes curate tools once and expose them through a managed MCP endpoint; Foundry IQ provides knowledge bases for grounding.

    Why it matters: Agents reuse governed tools and shared knowledge instead of one-off integrations.

    Limits: Foundry IQ is GA at API level only (portal access is preview), and memory is preview.

  • Observability and evaluation4

    Tracing, evaluations, red teaming and fine-tuning are built into projects.

    Why it matters: Quality and safety can be measured before and after release.

    Limits: Tracing is GA for prompt and hosted agents but preview for workflow and external agents; monitoring is preview.

  • Governance and security14

    Microsoft Entra identity, RBAC, content filters, network isolation and Azure Policy apply across agents, models and tools.

    Why it matters: One security model for the whole AI estate.

    Limits: API key access lacks RBAC granularity, and some features, such as traces and workflow agents, do not yet fully support network isolation.

  • Developer surfaces1

    Portal, SDKs in Python, C#, JavaScript and Java, the Azure Developer CLI, a Visual Studio Code extension, and a Foundry skill and MCP server for coding agents.

    Why it matters: Teams can prototype in the portal and move to code and CI/CD.

    Limits: Hosted agent tooling in the Azure Developer CLI targets hosted agent projects specifically.

Practical use cases

Staff need an assistant that answers from internal documents and can call internal systems.
Approach
Build a prompt agent with a Foundry IQ knowledge base and toolbox tools, then publish it to Microsoft Teams.
Role of Microsoft Foundry
Foundry hosts the agent, the model, retrieval, tools and tracing.
Data, infrastructure and skills
Azure subscription, Foundry project, an Azure AI Search service and a model deployment.
Type of benefit
Faster access to information
Caveats
Foundry IQ portal access is preview; permission enforcement depends on what each source supports.
First step
Create a prompt agent in the Foundry portal and connect one knowledge source.
An existing Azure OpenAI application now needs agents, evaluations and central governance.
Approach
Upgrade the Azure OpenAI resource to a Foundry resource and add projects.
Role of Microsoft Foundry
Foundry becomes the management plane for the existing deployments.
Data, infrastructure and skills
Owner rights on the Azure OpenAI resource and a migration plan.
Type of benefit
Consolidated governance
Caveats
Assistants and v1 agents are not supported in the new portal; keep them in Foundry (classic) until upgraded.
First step
Read Microsoft's upgrade guide and test it on a non-production resource.

Sources 4

A team already has agent code in LangGraph or Microsoft Agent Framework and needs managed hosting with identity.
Approach
Package the code as a hosted agent in Foundry Agent Service.
Role of Microsoft Foundry
Foundry provides the endpoint, scaling, Entra identity and observability.
Data, infrastructure and skills
Container image or source archive, Python or C#, and a region that offers hosted agents.
Type of benefit
Lower operations effort
Caveats
Hosted agents bill CPU and memory per active session in addition to model and tool use.
First step
Deploy the hosted agent quickstart with the Azure Developer CLI.

Sources 8

Who uses it

Works with

Optional integration

  • Azure Machine LearningThe product page lists Azure Machine Learning as an integrated service for model training.
  • Azure AI SearchFoundry IQ knowledge bases are built on Azure AI Search.
  • Microsoft EntraFoundry uses Microsoft Entra identity and RBAC for access control.
  • Microsoft PurviewThe product page names Purview among the governance integrations.

Complementary tools

  • Azure QuantumMicrosoft's quantum solution page lists Microsoft Foundry among related solutions.
  • Microsoft 365 Agents SDKDevelopers can use models and orchestrators from Foundry behind an SDK agent.
  • Microsoft Agent FrameworkHosted agents can run Agent Framework code; Microsoft points workflow users to Agent Framework.

Same family

Alternative approaches

  • Azure Container AppsMicrosoft's agent guidance presents GPU-backed containers as a third build path next to Foundry when you want to run your own models.
  • Microsoft Copilot StudioMicrosoft's adoption guidance presents Copilot Studio (SaaS, low-code) and Foundry (PaaS) as the two main build paths.

Optional integration for

Relationship labels follow Microsoft's documentation. "Alternative approaches" does not mean one is better: each profile says when it fits.

Getting started

  1. Make a first model call

    Follow Microsoft's build with models and agents quickstart to set up a project and send a prompt.

    Check: A prompt sent from code returns a model response from your project endpoint.

  2. Set up the developer environment

    Install the Azure CLI and the Foundry SDK for your language.

    Check: The project client authenticates with Microsoft Entra ID and lists your deployments.

  3. Choose a model

    Compare candidates in the Foundry Models catalog and check region availability.

    Check: The model you picked can be deployed in your project region.

  4. Build a prompt agent

    Create a prompt agent in the portal or with the SDK, then test it in the agents playground.

    Check: The agent answers test questions in the playground.

  5. Add tools and knowledge

    Attach a toolbox and, if needed, a Foundry IQ knowledge base.

    Check: Traces show the agent calling the tools and citing sources.

  6. Keep production on GA features

    Check the GA overview table and restrict preview features with tags or custom RBAC roles, as Microsoft's guidance describes.

    Check: Production workloads depend only on capabilities marked GA.

Official resources

Deciding whether you need it?

Three free tools apply Microsoft's own published guidance to your situation, and say when Microsoft Foundry or any new technology is not needed.

Sources

Each statement above links to the source it comes from. Labels say who reported it.

  1. What is Microsoft Foundry? (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
  2. Microsoft Foundry product page (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
  3. Foundry Models sold by Azure - Microsoft Foundry (opens in a new tab) Microsoft · Vendor-reported
  4. Microsoft Foundry portal general availability overview (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
  5. AI and machine learning products (Azure Architecture Center) (opens in a new tab) Microsoft · Vendor-reported
  6. Technology plan for AI agents (Cloud Adoption Framework) (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
  7. Microsoft Foundry Models overview (opens in a new tab) Microsoft · Vendor-reported · link checked 27 Oct 2026
  8. Hosted agents in Foundry Agent Service (opens in a new tab) Microsoft · Vendor-reported

Fill out the form below to request your copy.

Name(Required)