Microsoft Foundry
Microsoft Foundry: AI apps in your Azure environment.
Microsoft Foundry provides models, agents and tools in your Azure subscription. We use it to build applications with their own model, workflow or interface.
For development teams on Azure whose application needs a specific model, a workflow of its own in code or its own interface.
- Foundry Agent Service
- Azure OpenAI
- Claude
- Azure AI Search
- MCP
- Entra ID

Models, agents, tools
In Microsoft Foundry you choose the model from a catalogue and run it in your region with your identities and networks. The Agent Service runs agents: as a prompt agent without code, as a voice agent on the phone or as a hosted agent with your own code in a container. Tools such as Azure AI Search, SharePoint and MCP servers attach through a toolbox, and every decision is traceable in Application Insights.
What we build on Foundry
Foundry pays off where an application needs more than Copilot Studio offers: a specific model, a workflow of your own in code, processing without a chat or an interface outside Microsoft 365. These four applications are the ones we build on it most often; what companies use them for is described under Enterprise AI and Agentic AI.
- Document pipelineReads incoming contracts, quotes or reports, extracts the agreed fields, checks them against your rules and stores the result in Dataverse or your line-of-business system, without a chat and in a batch overnight.Result: in the morning the data sits checked in the business system, and your team works through the flagged exceptions.
- Specialist agent with the right modelAnswers questions from your documents through Azure AI Search, with the model that performs best for your language and field, for example Claude for long contracts.Result: answers with a source in your application, in Teams or on the website.
- Phone agentTakes calls, understands the request, asks for missing details and creates the case or transfers the call.Result: every enquiry outside office hours is recorded, and the team calls back with complete details.
- Hosted agent with your own codeRuns a multi-step workflow, for example check the order, ask the supplier, obtain approval, and waits at the point where a person decides, even for days.Result: a workflow your team understands, traces and adapts.
Region, sources, consumption
A Foundry application needs three things. An Azure subscription with a region where the chosen models and services are available; deployment type and data zone determine where processing happens. Sources that are right, and one role that owns what the application does. And one person who reads the consumption: models, tools and hosting are billed by use. For personal data, we record region and logging before the start.
What your team takes over
An application that runs in your Azure subscription and is tested on your examples: test set, evaluation and traces in Application Insights. A handover with source code, deployment as infrastructure code and responsibilities. One person on your team who adjusts model, sources and instructions and can stop the application. And a cost frame from the pilot before it is opened to everyone.
Sound familiar? Tell us where you stand today.
ContactWhat's new.
As of: 30 September 2026
- As of 11 September 2026
Hosted agents run in Switzerland North and Switzerland West.
Your own agent code, built with Agent Framework, LangGraph or no framework at all, runs as a container in Foundry Agent Service: with its own Entra ID, its own endpoint, an isolated sandbox per session and traces in Application Insights. Both Swiss regions are on the list of supported regions.
Source: Microsoft Learn: Hosted agents in Foundry Agent Service - As of 25 September 2026
Three agent types, publishing to Teams and Copilot.
Foundry Agent Service offers prompt agents without code, voice-based prompt agents for phone and browser (preview) and hosted agents with your own code. Published agents appear in Microsoft Teams, Microsoft Copilot and the Entra Agent Registry. The A2A protocol 1.0 for agent-to-agent communication is generally available.
Source: Microsoft Learn: What is Microsoft Foundry Agent Service? - As of 21 September 2026
Claude runs in Foundry, globally or in the US data zone.
Claude models are partner models through Azure Marketplace, hosted on Azure or on Anthropic's infrastructure. All versions run as a Global Standard deployment; a data zone exists for some models in the US only. Authentication uses Entra ID or an API key; the Mythos models accept Entra ID only.
Source: Microsoft Learn: Deploy and use Claude models in Microsoft Foundry
Models, boundaries and cost.
For choosing the model, drawing the line with Copilot Studio and the cost in operation.
GPT models in your Azure subscription, with quotas and deployment types.
Claude in FoundryWhen Claude is the better choice and how it runs in Azure.
GrokModels from SpaceXAI for research and market monitoring.
Local models with OllamaWhen models should run in your own data centre.
Open models from Hugging FaceSelect models with licence and origin and mirror them into your network.
Copilot Studio or a custom applicationThe dividing line, with examples from projects.
Knowledge with sourcesAnswers from your documents, with a reference to the passage.
What a Foundry agent costsWhich items arise for model, tools and hosting.
Hidden costs of RAGWhat your own search index in Azure means in operation.
Model diversity and approvalsWhy a second model checks the work and MCP tools ask before consequential actions.
Azure AI in our projects.
An agent in operation at BMC, two concepts with Azure OpenAI and Azure Speech, and an interface that makes line-of-business systems reachable for agents.

BMC Switzerland AG
BMC develops its Copilot agent with support from CNEXT
View project
CNEXT SI Labs
Peter: A concept for AI support in initial assessment
View project
CNEXT SI Labs
Misty: Exploring speech interaction and business data through robotics
View project
Berner Bildungszentrum Pflege AG
In progressModernising the central BZ Pflege API
View projectAgents, ready to use.
Two ready-made agents from our catalogue, for veterinary practices and for HR data in Abacus.
Related to your task
When the agent lives in Microsoft 365 and gets by with connectors.
Power PlatformDataverse and flows as storage and trigger for Foundry applications.
Entra ID and accessIdentities for people and agents, roles and conditions.
Enterprise AIOne AI platform for every department, with data boundaries, control and cost.
Agentic AIAgents that get tasks done in your systems, on Foundry and other platforms.
Support & operationsHave Foundry applications maintained and developed further after handover.
What development teams ask about Foundry.
Foundry or Copilot Studio?
Copilot Studio when the agent lives in Microsoft 365, answers from SharePoint and Dataverse and gets by with connectors. Foundry when you need a specific model, a workflow of your own in code, processing without a chat or an interface outside Microsoft 365. The two can be combined when an agent in Microsoft 365 should use a Foundry application.
Does our data stay in Switzerland?
That depends on the model and the deployment type. Hosted agents run in Switzerland North and Switzerland West. Of the OpenAI models, as of September 2026 GPT-4.1, GPT-4o and the embedding models run as a regional deployment in Switzerland North; newer GPT models process in the EU data zone or globally. Claude models run globally, and a data zone exists for some of them in the US only. We check where processing happens before the model choice and record it in the proposal.
Which model do we use?
The one that performs best on your examples. Foundry provides models from OpenAI, Anthropic, SpaceXAI, Mistral, DeepSeek and Meta, and switching the model does not change the agent code. In the pilot we compare two or three models on the same test set for quality, response time and consumption.
Can the application act on its own?
Yes, through tools in a toolbox: your own interfaces via OpenAPI, MCP servers, Azure AI Search, SharePoint and code execution. Each tool gets its own permissions through the identity of the agent in Entra ID. Where an action has consequences, a hosted agent stops and waits for the decision of a person, even for days. As of August 2026 Microsoft runs this approval step as a preview.
How do we keep costs under control?
Models are billed per token, tools per call, hosted agents by processor and memory of active sessions. In the pilot we read the consumption per application in Azure, set budgets and alerts and choose the deployment type by load. So you know the frame before the application is opened to everyone.
Do we need developers for this?
Not for prompt agents: they are configured in the Foundry portal. Yes for hosted agents and pipelines, in Python or C#. We build the first application and hand it over so that your developers or we carry it on.
Who runs the application after handover?
Your team, with our support if you want it. We hand over source code, deployment as infrastructure code, the test set and one person who adjusts model, sources and instructions. Microsoft retires models with announced dates; with the test set we show how the switch works.
Agree on a useful next step
Tell us about your current workflow, the systems involved and the result you need. In our first conversation, we identify what else we need to know to plan a useful next step.
