Building an AI application used to mean stitching together a model API, a vector database, an evaluation script, and a security layer from five different vendors. Microsoft AI Foundry — Microsoft’s unified platform for building, deploying, and governing AI apps and agents — was built to remove that friction. In early 2026, Microsoft rebranded the product from Azure AI Foundry to Microsoft Foundry, though the platform itself, its portal, and its capabilities carried over without a break.
What is Microsoft AI Foundry?
Microsoft AI Foundry is an enterprise-grade, unified platform-as-a-service on Azure for the full AI application lifecycle — from picking a model, to building an agent, to evaluating it for safety and accuracy, to deploying and monitoring it in production. It brings together a model catalog, agent-building tools, fine-tuning, evaluation, and enterprise security under a single portal and a single set of SDKs, rather than requiring teams to assemble these pieces themselves.
Why Microsoft AI Foundry?
Enterprises adopt Foundry because it collapses several separate concerns into one governed platform:
- One control plane — for models, agents, data connections, and deployments instead of five disconnected tools.
- Enterprise trust built in — identity, network isolation, and content safety are part of the platform, not an afterthought.
- Model choice without lock-in — Foundry hosts models from OpenAI, Microsoft’s own MAI family, Meta, Mistral, DeepSeek, Cohere, and others side by side, so teams can route tasks to the right model on cost or capability grounds.
- A shorter path from prototype to production — thanks to managed agent hosting, built-in tracing, and evaluation loops that feed back into development.
Key Features of Microsoft AI Foundry
1. Model Catalog
Access a centralized catalog of foundation, open-source, and task-specific AI models. Teams can discover, compare, and deploy text, vision, and speech models from a single workspace without managing multiple platforms.
2. AI Agents
Build intelligent AI agents with memory, tools, instructions, and multi-step reasoning. The managed agent runtime supports conversational workflows, automation scenarios, and collaboration between multiple agents in enterprise applications.
3. Prompt Flow
Design, test, and optimize LLM workflows through a visual and code-first experience. Prompt Flow helps teams connect prompts, retrieval steps, APIs, and business logic into reliable end-to-end AI pipelines.
4. Fine-Tuning
Customize models using your organization’s data and business terminology. Fine-tuning improves response quality, domain accuracy, and consistency while providing better control over model behavior.
5. Enterprise Security
Secure AI workloads with Microsoft Entra ID, role-based access control, private networking, and content-safety filters. Built-in security features help organizations meet compliance requirements in regulated industries.
Microsoft AI Foundry Architecture
Foundry organizes work into a two-level structure: a Hub at the top for shared, governed infrastructure, and one or more Projects underneath for individual teams or applications. Each project gets its own endpoint, its own managed identity, and its own connections to external data sources — so teams can share governance while keeping their data and permissions isolated.
Core Components
- AI Hub – the top-level container for shared resources, networking, and governance policies across teams.
- AI Project – a team- or app-specific workspace with its own endpoint, connections, and deployments.
- Model Catalog – the searchable directory of deployable models described above.
- Prompt Flow – the pipeline-design tool for chaining prompts and logic.
- AI Agents – the Agent Service runtime for building and hosting agents and tools.
- Azure AI Search – the retrieval engine most commonly used to ground agents in your own documents (RAG).
- Monitoring – tracing, cost, and performance dashboards for deployed models and agents.
Supported AI Models
Foundry’s catalog includes flagship models from OpenAI, Microsoft’s own MAI family, Meta’s Llama models, Mistral, DeepSeek, Cohere, and many smaller open-source and task-specific models — covering chat, reasoning, vision, and speech use cases, all deployable through the same interface and billed under the same subscription.
AI Development Workflow
A typical project moves through five stages: explore the model catalog, build a prototype (prompts, RAG, or an agent), customize it with fine-tuning or grounding data, evaluate it against quality and safety metrics, and finally deploy and monitor it — looping back to earlier stages as real-world feedback comes in.
How to Build an AI Application Using Microsoft AI Foundry
- Create a Hub, then a Project inside it.
- Browse the Model Catalog and deploy a model as an endpoint.
- Use Prompt Flow or the Agent Service to design your application logic and tools.
- Connect Azure AI Search (or another data source) if your app needs grounding.
- Run evaluations against a test dataset to check quality and safety.
- Deploy the project endpoint and monitor it with built-in tracing and dashboards.
Frequently Asked Questions (FAQs)
Yes. Microsoft renamed Azure AI Foundry to Microsoft AI Foundry in 2026. The features and portal stayed the same; only the name changed.
No. You can use only the features you need, such as the Model Catalog, Prompt Flow, or Agents.
A Hub is shared infrastructure for an organization or team. A Project is a workspace for a specific application or team under that Hub.
You can use models from OpenAI, Microsoft, Meta, Mistral, DeepSeek, Cohere, and other open-source providers through the same platform.
- Customer support
- Healthcare
- Finance
- E-commerce
- Media and entertainment
- Education
No. Azure AI Search is only needed when your agent must search or use your own documents (RAG).
Yes. It includes security features such as Microsoft Entra ID, role-based access control, private networking, and content safety filters, which help support compliance needs.
Conclusion
That covers the fundamentals: what Microsoft AI Foundry is, why enterprises are adopting it, its core features, and how its Hub/Project architecture fits together. With that foundation in place, the platform’s more advanced capabilities — and the details that matter most when deciding whether to build on it — are still ahead. We’ll cover those next in Part 2, including retrieval-augmented generation (RAG), building and hosting agents in depth, security and compliance, real-world use cases, pricing, and what’s new in Foundry for 2026.