Azure AI Foundry Agent Service

Overview

Azure AI Foundry Agent Service helps teams move from AI prototypes to production-ready automation. Initially introduced as Azure AI Agent Service, it now anchors the broader Azure AI Foundry platform.

What it does:

• Goes beyond chat: process documents, manage workflows, automate multi-step business tasks
• Orchestrates models and tools with governance and observability included by default
• Handles state, retries, content safety, and enterprise integration primitives

Scale:

• Supports Azure OpenAI models and selected open-source models (e.g., Llama, Mistral), depending on region
• 1,400+ Logic Apps connectors for enterprise systems
• Multi-agent orchestration available (some capabilities remain in preview)

Availability depends on region. See Azure Products by Region.

Architecture

The service follows a conceptual “assembly line” approach.

1. Model Selection: Azure OpenAI + selected OSS models
2. Customization: Prompts, fine-tuning, RAG
3. Tool Integration: Bing Search, SharePoint, Azure AI Search, Logic Apps, Functions, custom APIs
4. Orchestration: Tool calls, state, sequencing, retries
5. Trust & Security: Entra ID, RBAC, content safety, VNet isolation
6. Observability: Azure Monitor + Application Insights tracing

Multi-Agent Orchestration

Two supported patterns:

• Connected Agents: A primary agent delegates tasks to specialized sub-agents (preview)
• Multi-Agent Workflows: Stateful workflows spanning multiple agents (preview/GA mix depending on feature)

Example: A support agent routes billing issues to a finance agent, product questions to a knowledge agent, and compliance checks to a policy agent.

The runtime brings together concepts from Semantic Kernel and AutoGen and supports external frameworks such as CrewAI, LangGraph, and LlamaIndex.

---

Enterprise Security

• Identity: Microsoft Entra + RBAC + on-behalf-of authentication
• Compliance: Integrates with Microsoft Purview for AI governance workflows
• Network: Private endpoints, VNet integration, isolation options
• Threat Protection: Defender for Cloud + integrated monitoring

Development Experience

Multiple development paths:

• Agent Playground: Visual/no-code design and testing
• Azure AI Foundry SDKs: Python, C#, TypeScript/JavaScript, Java
• Evaluation Tools: Metrics for tool robustness and task adherence
• CI/CD Friendly: Works with Azure DevOps, GitHub Actions, and standard pipelines

Use Cases

• Customer Service: CRM-linked intelligent routing and response
• Document Processing: Invoices, contracts, knowledge extraction
• Business Automation: Approvals, supply chain, reporting flows
• Research & Analysis: Trend analysis, summarization, monitoring
• DevOps: Code review helpers, environment checks, operations bots

Summary

Azure AI Foundry Agent Service bridges the gap between AI experiments and enterprise-ready automation.
Its opinionated runtime — with built-in security, observability, and governance — helps organizations move beyond demos and into scalable, trustworthy agent workflows.

Resources

• Service Overview
• General Availability Announcement