Foundry Services: Your AI Toolkit | Guided by A Guide to Cloud

What is Microsoft Foundry?

Simple explanation

Microsoft Foundry is like a workshop with every AI tool you could need — all in one building.

Need to deploy a language model? There’s a workbench for that. Need to search through company documents? There’s a tool for that. Need to build an AI agent that books meetings? There’s a framework for that. Need to extract data from invoices? There’s a machine for that.

Instead of buying tools from 10 different shops, Foundry puts everything under one roof — one portal, one set of SDKs, one billing system.

The Foundry service map

The exam tests whether you know which Foundry service handles which job. Here’s the complete map:

Task Category	Foundry Service	What It Does
Generative tasks	Model deployments (OpenAI, Llama, Phi, etc.)	Text generation, chat, code, summarisation
Grounding	Azure AI Search + Foundry IQ	Ground model responses in your data
Vector search	Azure AI Search (vector index)	Semantic similarity search using embeddings
Agent workflows	Foundry Agent Service	Build agents with tools, memory, and orchestration
Multimodal processing	Multimodal models + Content Understanding	Process text, images, audio, video together
Document extraction	Content Understanding	OCR, layout analysis, structured field extraction
Speech	Azure AI Speech	Speech-to-text, text-to-speech, translation
Translation	Azure Translator	Text and document translation
Evaluation	Foundry Evaluation	Measure quality, safety, groundedness

Foundry architecture: Projects and resources

Every AI solution in Foundry starts with a Foundry Project. Think of it as your workspace:

Component	What It Contains
Foundry Project	Models, agents, data connections, evaluations, deployments
Model deployments	Hosted models you can call via API (pay-per-token or provisioned)
Connected resources	Azure AI Search indexes, storage accounts, Key Vault
Agent definitions	Agent configurations with tools, instructions, and knowledge

Classic vs New Foundry architecture

If you’ve used Azure AI Studio before, you’ll remember the hub-and-project model. The “classic” architecture had hubs as parent containers for projects.

The new Foundry architecture (2026) simplifies this — projects are standalone resources. No more hubs. This makes access control, lifecycle management, and governance simpler.

The exam focuses on the new architecture. If a question mentions “hubs,” it’s likely testing whether you know they’re deprecated.

Grounding and Foundry IQ

Grounding means anchoring AI responses in your actual data instead of letting the model hallucinate. Two key services:

Azure AI Search vs Foundry IQ for grounding
Feature	Azure AI Search	Foundry IQ
What it is	Full search platform with indexing, ranking, and retrieval	Foundry-native knowledge layer for agents
Search types	Keyword, semantic, vector, hybrid	Integrated with agent memory and tools
Best for	RAG applications, custom search pipelines	Quick agent knowledge setup without custom indexing
Setup effort	Medium — configure index, chunking, embeddings	Low — upload docs, Foundry handles the rest
Flexibility	High — full control over indexing and retrieval	Lower — abstracted for simplicity

Exam tip: When to pick Search vs Foundry IQ

The exam will present scenarios where you need to ground an AI response. Key decision:

Need full control over indexing, chunking, hybrid search, enrichment? → Azure AI Search
Quick setup for an agent that needs to search company docs? → Foundry IQ
Both are correct for RAG — the question is about complexity and control requirements

Agent Service

The Foundry Agent Service is where you build AI agents — systems that don’t just chat, but take actions:

Agent Feature	What It Provides
Model selection	Choose which LLM powers the agent’s reasoning
Instructions	System prompt defining the agent’s role, rules, and personality
Tools	Functions the agent can call (APIs, search, code execution)
Knowledge	Data sources the agent can query (Foundry IQ, Search indexes)
Memory	Conversation history and per-thread state
Orchestration	Multi-agent coordination with approval controls

Real-world example: Atlas Financial's compliance agent

Atlas Financial builds a compliance review agent using three Foundry services:

Agent Service — defines the agent’s role (“Review loan applications against regulatory guidelines”)
Azure AI Search — indexes 50,000 pages of regulations for grounding
GPT-4o deployment — provides the reasoning capability

The agent receives a loan application, searches relevant regulations, and produces a compliance assessment with citations. Three services working together.

Content Understanding

Content Understanding is Foundry’s newer multimodal extraction service for processing documents, images, and video. It coexists with Document Intelligence, which remains available for document-specific tasks:

Capability	What It Does
OCR	Reads printed and handwritten text from images and PDFs
Layout analysis	Understands document structure (tables, headings, sections)
Field extraction	Pulls specific fields (invoice number, date, total)
Single-task mode	Optimised pipeline for one extraction type
Pro mode	Multi-capability pipeline combining OCR, layout, and extraction

Exam tip: Content Understanding vs Document Intelligence

These are separate, coexisting services — Content Understanding did NOT replace Document Intelligence. The exam may test when to use which:

Content Understanding — multimodal extraction (documents + images + audio + video), custom schemas, Markdown output for LLM consumption
Document Intelligence — document-specific extraction with pre-built models (invoices, receipts, tax forms), high-accuracy structured output

If the scenario involves multimodal content or feeding output to an LLM/agent → Content Understanding. If it’s purely document extraction with pre-built analysers → either service works, but Document Intelligence has more mature pre-built models.