Microsoft Foundry: AI Agents, Foundry IQ, MCP, A2A & RAG

Discover how generative AI and AI agents use foundation models, large language models, and tools to dynamically plan, orchestrate APIs and plugins, and enable human-in-the-loop workflows.

Discover how Microsoft Foundry unifies AI agents, model catalog, and enterprise knowledge through the agent service, Foundry IQ, and control plane to productionize scalable, governed AI.

Explore the Microsoft Foundry SDK, compare chat completions and assistance APIs, and learn to build agents on models from OpenAI, llama, Mistral, and Cohere with plugins and data in Python.

discover the differences between hub-based and standalone foundry projects, including organization-wide collaboration vs lean team use, sdk versions, shared connections, and infrastructure management.

Deploy a standalone Microsoft Foundry project in the Azure portal, explore the foundry portal, and compare models like GPT 5.1 and Cloud Opus 4.5 to guide selection.

Access the GitHub repo for course demos, clone or fork the Microsoft Foundry code base written in Python, and start demos in VS Code with Microsoft Foundry extensions.

Deploy a GPT-4 chat completions model in Microsoft Foundry using the Foundry SDK 2 and the chat playground, then call it from VS Code via the OpenAI client.

Deploy a cloud model from the foundry catalog, configure environment variables, and call it via the anthropic sdk with the messages api, demonstrating a translation task.

Install the Azure resources extension and the Microsoft Foundry extension in VSCode, sign in with your Azure account, and use the model playground and remote agent playground.

Create a demo agent in the Microsoft Foundry portal, choose a GPT four model, enable versioning, and explore tracing, conversation IDs, memory store, tools, YAML, and Foundry IQ integration.

Create and test a basic Batman agent in a code-first microsoft foundry lab by wiring the foundry sdk, openai client, and conversation IDs to manage chat history.

Create a web searcher agent in Microsoft Foundry with the foundry sdk, configure environment variables, attach the web search tool, and run GPT-4 powered real-time queries with sourced results.

Create a code interpreter agent in the Microsoft Foundry ecosystem, using a CSV data asset and code interpreter tool to generate a matplotlib column chart in a sandbox.

Create an agent with the Foundry SDK and an OpenAPI tool using a weather OpenAPI definition to fetch current weather for a location, demonstrated with New York City.

Integrate a public MCP server with a foundry agent via registered tools, env vars, and connection IDs, using a GPT-4 powered agent to retrieve learning paths for Azure AI.

Explore multi-tool capabilities in a foundry agent by orchestrating a weather OpenAPI tool and an MCP server to answer weather queries and locate Microsoft Learn modules.

Learn to use the browser automation tool with a Microsoft Foundry agent, setting up a Playwright workspace and a browser automation connection for a hands-on lab.

Discover how Microsoft Foundry memory store provides a built-in, token-efficient vector database with embedding and retrieval to persist memories across user sessions.

Grant the Foundry project's managed identity the Azure AI user role at the resource group level in portal.azure.com to access the OpenAI LLM and embedding models for memory store lab.

Learn to work with the foundry memory store and its components. Define memory, update memories, and retrieve grounding knowledge for prompts.

Explain retrieval augmented generation and how retrieval, augmentation, and generation deliver grounded answers from enterprise documents via similarity search on vector databases.

Explore vector embeddings, using Ada 002's 1536 dimensions, to enable semantic retrieval with cosine similarity and retrieval augmented generation using Azure AI search, Cosmos DB, and Redis cache.

Learn how Azure AI search ingests unstructured and multimodal data, indexes it, and supports vector, full text, and hybrid search for retrieval augmented generation with AI enrichment.

Explore multi-modal rag in Azure AI search, deploying image verbalization and layout-aware document intelligence to preserve structure, generate citations, and enable retrieval augmented generation with vector embeddings.

Deploy a RAG infrastructure on Azure by provisioning a storage account, enabling anonymous access, uploading PDFs, and building an AI search index with Ada 002 embeddings for Foundry agent chat.

Create a multimodal rag index in Azure AI search by importing blob data, verbalizing images with GPT-4, and generating text embeddings for semantic ranking.

Develop a hands-on lab to build a multi-modal rag agent using Azure AI search, Foundry, and vector semantic hybrid queries.

Azure API management serves as the gateway for Microsoft Foundry, mediating API access, policies, monitoring, and monetization across developers, publishers, MCP servers, and generative AI tools.

Explore the model context protocol as a universal, server-based interface that enables AI agents to access APIs via remote or local MCP servers, streamlining maintenance and integration.

Explore MCP server architecture, where a host app runs a client connecting to local and remote servers via a data and transport layer, using JSON-RPC and streamable HTTP.

From a solution architect perspective, this lecture shows how to expose remote MCP servers behind an Azure API management gateway, apply policies, and enforce identity-based access control.

Set up AI gateway in a Foundry project, explore API management, MCP servers, and monetization through products and subscriptions, with region options and pricing tiers.

Configure token-per-minute rate limits for a GPT-4 deployment via the AI gateway. Govern MCP servers through the API gateway and apply inbound rate-limiting policies to demonstrate quota control.

Deploy an MCP server in Azure Container Apps, build and push its fastapi-based image to Azure Container Registry, then expose it behind an api gateway for secure governance.

Attach the Azure Container Apps hosted MCP server to the API Management service that acts as the API gateway for the Foundry portal, then configure the base URL and policies.

Connect your MCP server to the Microsoft Foundry portal, attach it to an agent, and apply a customizable rate-limiting policy under API management to control inbound calls.

Explore how Foundry IQ unifies enterprise knowledge from SharePoint, one lake, and the 365 ecosystem into a single source of truth for AI agents.

Explore how agentic retrieval in Foundry IQ orchestrates source selection, query planning, and parallel knowledge queries, then uses L2/L3 ranking and a fine tuned language model to ground answers.

Understand how BM25 ranking and semantic reranking in Azure AI search boost relevance for retrieval-augmented generation. Learn how inverted indexes, lexical analysis, TF-IDF, and semantic configurations shape scoring and results.

Set up Foundry IQ lab by uploading documents to Azure Blob storage and SharePoint, index them in Azure AI Search, and create a knowledge base with a grounded Foundry agent.

Create a Foundry IQ agent in the portal, attach the Carbon Ops knowledge base with Azure Blob and SharePoint sources, configure skills and embeddings, then run multi-source queries.

Create a grounded foundry iq agent in a code-first Python notebook by configuring environment variables, MCP endpoints, and knowledge base connections, then test with an ESG-focused query.

Understand multimodality in AI, where predictive and generative AI partner to process text, images, and charts. Learn how to preserve document layout, extract insights, and generate charts using agents.

Learn how Azure Language Service masks PII to protect personal data, with pre-built features and custom features; see how a Foundry agent routes queries to the service for compliant workflows.

Conduct a hands-on lab for PII text redaction with Azure Language Service and Foundry agents, including environment setup and redacting names and emails in travel planning chats.

Azure Document Intelligence uses OCR and bounding boxes to extract data from invoices, receipts, and IDs with pre-built and custom models, returning JSON with confidence scores for foundry agents.

Analyze invoices with azure document intelligence and pii redaction via azure language service, then use a foundry agent with code interpreter to create a chart of invoice totals.

Discover multi-agent orchestration workflows in Microsoft Foundry, a low-code solution for citizen developers to design scalable, human-in-the-loop processes and hand them to developers.

Build a sequential orchestration workflow that acts as a Ms Learn path builder, coordinating three agents—topic builder, Ms Learn module picker, and study plan generator—for a personalized Microsoft Learn roadmap.

Explore building a human-in-the-loop sequential workflow for AI agents, including user feedback loops, conditional routing, and end-to-end orchestration with study plan generation and topic builder agents.

Design a two-agent chat workflow in the foundry orchestration tool, with a recipe agent and a judge agent that score recipes and loop feedback to iterate.

Explore the Microsoft agent framework as the successor to semantic kernel and Autogen, and learn how interoperable, memory-enabled agents operate across Foundry and multi-agent workflows.

Explore hands-on lab steps to build a basic chat agent with Microsoft Foundry and the Microsoft Agent Framework, including cloud and local agents, environment setup, and a streaming demo.

Build a local chat agent with the Microsoft Agent Framework, creating a runtime-bound identity that does not persist in Microsoft Foundry, using an OpenAI chat model.

Create a foundry-centric agent using the Microsoft Agent Framework with a hosted MCP tool pointing to the Microsoft Learn MCP server to answer Microsoft Learn content.

Learn to build a family-centric multi-tool agent using Microsoft Foundry, attaching a code interpreter and a Microsoft Learn MCP tool, with sandbox Python execution.

Learn to set up dev ui to visualize Microsoft Foundry agent workflows, create a docs agent, run the python script, and inspect agent trace stacks and token usage.

Explore structured output streaming in Microsoft Agent Framework to populate an Excel sheet using a predefined schema and a Pydantic model.

Build a sequential workflow with two agents in Microsoft Foundry—researcher and writer—where grounding knowledge on given topics informs article creation.

Visualize a sequential workflow with the Microsoft agent framework using dev UI, executing researcher and writer agents, with tracing and token usage visible.

Design and run parallel travel workflows with Microsoft Agent Framework and Foundry, wiring location picker, destination recommender, cuisine, weather, and itinerary planner agents to generate a complete itinerary.

Visualize parallel workflows in the Microsoft agent framework with Dev UI, coordinating location picker, destination, weather, cuisine, and itinerary planner agents, with live traces and visualization.