Microsoft Agent Framework (MAF): Build Enterprise AI Agents

Learn how the Microsoft Agent Framework (MAF) enables model-agnostic and platform-agnostic development for building enterprise AI agents.

Learn to set up a Visual Studio Code project for the Microsoft Agent Framework: create a folder, start a Python virtual environment, and install the framework with its subpackages.

Explore using Microsoft Agent Framework with OpenAI by configuring an OpenAI API key, selecting the OpenAI chat client, and setting environment variables and a .env file.

Provision an Azure OpenAI service in portal.azure.com under Microsoft Foundry, selecting subscription, resource group, region, name, pricing, and networks, then review and submit to deploy LLM for Microsoft Agent Framework.

Create a deployment in Azure OpenAI and configure environment variables for Microsoft Agent Framework (MAF), including endpoint, deployment name (gpt-5.1-chat), and API key.

Explore Microsoft Foundry as the umbrella platform for deploying diverse models—from Llama to Cloud—using the Foundry chat client and default Azure credentials within the Microsoft Agent Framework.

Learn to create a Microsoft Foundry resource in Azure, configure subscription, resource group, region, and identity, then deploy models for use with the Microsoft Agent Framework.

Learn to build a Microsoft agent framework agent inside Azure Foundry, using the Foundry chat client and default Azure credential to run async prompts.

Explore MAF concepts with a deep dive, and learn to build enterprise AI agents, using music-inspired examples.

Explore the Microsoft Agent Framework (MAF) software development kit and library, an open source platform for Python and .NET that blends semantic kernel and autogen for secure, multi-agent orchestration.

Create and use the agent class with a foundry chat client, configure the default azure credential, and name a sample math cloud agent while preparing to explore the messages class.

See a practical demo of the message class: import the message class, create a user message with role and contents, and pass it to agent.run to generate bullet points.

Learn how the tool decorator turns a Python function into a callable tool for an AI agent. It registers tools for the LLM and enables API calls during conversations.

Demonstrate using the tool decorator to turn a simple get weather function into a callable agent tool with a json schema, and swap in a real OpenWeatherMap API call.

Short-term memory in the Microsoft Agent Framework stores the current conversation context, including recent user messages, agent replies, and tool outputs, within a session that resets when the session ends.

See a hands-on demo of short-term memory in the Microsoft Agent Framework, building a session-based in-memory conversation history, and enriching responses with emojis through multi-turn agent.run calls.

Learn long-term memory in the Microsoft Agent Framework, persisting user preferences and past decisions across sessions via context providers and memory providers, using vector stores for retrieval.

Discover how model context protocol (MCP) standardizes tool and data integration for LLMs, acting as a universal USB-C for AI to access external tools like databases, files, and REST APIs.

Integrate MCP with the Microsoft Agent Framework to connect to MCP compliant tools on the fly, removing custom connectors and enabling plug‑and‑play tools across environments.

Explore the three MCP connection tool types for Python: local MCP servers, remote MCP servers, and WebSocket, and learn about MCP-STDIO-TOOL, MCP-STREAMABLE-HTTP-TOOL, and MCP-WEBSOCKET-TOOL.

Learn to use the Microsoft Agent Framework with the Azure MCP server to power an Azure Ops assistant that queries resource groups and storage accounts via the MCP tools.

Explore the agent to agent (a2a) protocol, an open, vendor-neutral standard for secure, scalable cross-agent communication using json rpc over http/https, distinct from mcp.

Explore the building blocks of A2A, including the initiating client, responding server, and JSON agent cards, secured by OAuth, API keys, and mutual TLS for enterprise security.

Learn how the Microsoft Agent Framework enables a2a interoperability, discovery, authentication, and invocation across agents and non-math agents, with secure Azure authentication, structured JSON requests, and streaming results.

Learn to run the code in the Microsoft Agent Framework by starting the Founder Raj server, invoking Shark Tank with Priya, and evaluating a cybersecurity startup pitch.

Build basic agents with real tools using the Microsoft Agent Framework to empower enterprise AI.

Demonstrates building streaming responses with an agent by enabling stream in agent.run, iterating over chunks, and printing chunk.text with end and flush to show tokens as they arrive.

Learn how the Microsoft Agent Framework analyzes images by feeding image data to an LLM to describe the scene and reveal interesting facts.

Explore the human in the loop workflow for enterprise ai agents, with approval, tool calls, session memory, and EU AI Act considerations, illustrated by a weather example.

Demonstrates a human-in-the-loop workflow in the Microsoft Agent Framework, using tool decorators with alwaysRequire to pause for human approval before tool calls and produce an emoji-rich one-page weather report.

Build secure, enterprise-ready AI agents with Microsoft Agent Framework by intercepting and filtering user prompts with middleware, flagging banned words before they reach the LLM.

demonstrates implementing a security middleware in the Microsoft agent framework to block or filter requests before the llm, using a pipeline with context, pre-processing checks, and call_next.

Explore how multi-agent workflows coordinate specialized agents using sequential, concurrent, and handoff orchestration patterns to design scalable agentic systems.

Explore why multi-agent systems address single-agent limitations by coordinating specialized agents within workflows for collaborative intelligence, shared knowledge, and robust adaptive problem solving.

Explore the core workflow components in MAF: executors, edges, workflows, and events, and see how concurrent agents and routing enable enterprise task orchestration and monitoring.

Explore the supported orchestration patterns in Microsoft Agent Framework, including sequential, concurrent, hand-off, group chat, and magentic coordination. Understand when to use each for multi-agent collaboration.

Learn sequential orchestration, where agents process data step by step in a production line, with each output becoming the next input, showing order matters in review, summarize, and approve.

Demonstrate sequential orchestration of analyzer, optimizer, and reviewer agents to polish a social media post with tone analysis, hashtags, emojis, and grammar improvements.

Explore concurrent orchestration where multiple agents act in parallel, independently process tasks, and aggregate diverse outputs into a cohesive insight for parallel data analysis and ensemble decisions.

Plan a trip to Paris using concurrent orchestration: a single prompt runs in parallel across five agents (food, accommodation, activities, transport, budget) to produce a final travel report.

Explore hand-off orchestration for a customer support assistant, routing from the coordinator to billing, technical, and supervisor agents, with real-time event streams via an OpenAI chat client and handoff builder.

Demonstrates handoff orchestration across four agents: support coordinator, billing, technical, and supervisor, sharing one OpenAI chat client and routing via a handoff graph to handle customer prompts and scripted responses.

Execute the code and observe orchestration in the Microsoft Agent Framework routing a damaged order issue through agents and handoffs—from the support coordinator to the specialist and billing teams.

Explore observability in the Microsoft Agent Framework (MAF) to build enterprise AI agents, emphasizing applicability to enterprise-scale agent development.

OpenTelemetry is a vendor-neutral open source observability framework backed by cloud native computing foundation (cncf). Instrument once to generate, collect, and export telemetry data to a back-end for AI agents.

Microsoft Agent Framework embeds observability out of the box with native OTEL integration, generating telemetry, traces, logs, and metrics to monitor multi-agent workflows and support enterprise monitoring.

Build an enterprise AI agent for incident triage using the Microsoft agent framework (MAF), grounded in runbooks and LLMs like GPT 4.0 mini, with OpenTelemetry for observability.

Understand how the triage agent reads alerts, looks up runbooks, and produces a concise triage summary, while implementing open telemetry tracing, spans, and observability for end-to-end visibility.

Watch the demo execute the code and reveal the triage agent’s outputs and next steps, while demonstrating observability with trace ID, Azure Monitor, flame graphs, and token-level metrics.

Create your own Application Insights instance in portal.azure.com to enable application observability within Azure Monitor for the stock market analyst agent, and write telemetry from the agent into App Insights.

Log into Azure with AZ login to run the enterprise AI agents demo in VS Code, select your tenancy, then view simulated stock data and trace IDs for Application Insights.

Build real world agents using the Microsoft Agent Framework (MAF) to create enterprise AI agents.

Explore how to build an ai science tutor within the Microsoft Agent Framework (MAF) to support enterprise ai agents.

Build a real-world AI science tutor named Professor Spark that explains concepts, creates quizzes, grades answers, and suggests experiments using OpenAI chat and session memory.

Explore two code paths—non-streaming and streaming—using async io and openai responses client to build an AI science tutor with guardrails, teaching relativity and Pythagoras with examples.

Build a news reader agent with Microsoft Playwright MCP, a Modern Context Protocol server enabling LLMs to browse pages via structured snapshots.

Demonstrates building an AI agent with the MCP server and Playwright MCP for browser automation, using an OpenAI chat client to fetch BBC sport headlines and summarize with emojis.

Explore how the document insights agent fits into the Microsoft Agent Framework to build enterprise AI agents.

Explore retriever augmented generation to answer from enterprise documents using embeddings and vector databases, inject retrieved context into prompts, and ground responses with the document insights agent.