Microsoft Foundry Goes GA: Enterprise-Grade Agent Platform Now Available

Microsoft has announced the general availability of its Foundry platform, marking a significant milestone in enterprise AI agent development. The platform now offers production-ready tools for building intelligent agents that can plan, reason, and collaborate across data, tools, and people while maintaining enterprise-grade security and observability.

Next-Generation Foundry Agent Service

The cornerstone of this release is the Foundry Agent Service, now generally available with production-ready SDKs across Python, JavaScript, Java, and .NET. Built on OpenAI's Responses API, the service offers wire-compatible agents that allow teams already using the Responses API to migrate to Foundry with minimal code changes while immediately gaining enterprise security, private networking, and observability features.

Developers can publish agents through REST endpoints or directly to Microsoft 365 and Microsoft Teams. The platform supports both prompt (declarative) agents in GA and hosted agents in public preview, with regional expansion covering East US, North Central US, Sweden Central, Southeast Asia, Japan East, and more. This global footprint enables organizations to meet data residency requirements and reduce latency for international deployments.

For teams using frameworks like Microsoft Agent Framework or LangGraph, hosted agents provide a path to operationalize agent logic without managing infrastructure. Multi-agent workflows in LangGraph can now connect directly to Foundry Agent Service, allowing developers to compose complex graphs while keeping individual agents running in Foundry's secure environment.

Voice Live Integration

Voice Live API integration with Foundry Agent Service is now in public preview, making voice a native, production-ready channel for agents. This unified, real-time API provides premium speech capabilities with foundation models for natural speech-to-speech interactions. The integration enables seamless, secure voice interactions for scenarios like customer service and field support, with Voice Live handling all audio input and output while inheriting Foundry's security and monitoring features.

Gulf Air's Falcon Eye platform demonstrates this capability in action, allowing executives to speak naturally and receive immediate, accurate spoken answers grounded in live operational data. Voice-driven navigation can also open dashboards and analyses, creating a conversational mission-control experience for real-time decision-making.

Observability and Quality Management

As organizations move agentic AI systems into production, quality and safety require continuous measurement rather than one-time testing. Foundry Control Plane's observability features are now generally available, enabling systematic assessment of performance across the AI lifecycle.

Teams can leverage out-of-the-box evaluators for coherence, relevance, groundedness, retrieval quality, and safety across generation and retrieval augmented generation (RAG) scenarios. Custom evaluators in preview allow organizations to define LLM-as-a-judge and code-based custom evaluators aligned to business logic, internal standards, and regulatory requirements.

Continuous evaluation with monitoring and alerting automatically assesses sampled production traffic using built-in and custom evaluators, surfacing quality, safety, availability, and performance signals through Azure Monitor. Tracing capabilities capture requests, tool calls, and responses to understand system behavior, diagnose failures, and debug issues across agents, models, and tools in production.

Prompt optimizer is also in preview, applying prompt-engineering best practices to restructure, clarify, and enhance instructions with transparent, paragraph-level reasoning for every change.

Enterprise Security and Private Networking

Foundry Agent Service supports standard setup with private networking, allowing organizations to bring their own private virtual network (BYO VNet) for isolated environments. This means no public egress, container/subnet injection into networks, and access to private resources via the platform network.

Private networking extends to tool connectivity, including VNet support for Model Context Protocol (MCP), Foundry IQ by Azure AI Search, and the Fabric data agent. The platform supports a full spectrum of authentication methods for MCP servers, including key-based authentication, Microsoft Entra Agent ID, Foundry project managed identity, and OAuth identity passthrough.

Organizations can disable preview features entirely and standardize on GA-only capabilities for production development, ensuring consistent governance across their AI estate.

Partner Security Integrations

Microsoft Foundry is expanding customer choice through partner-native security integrations directly accessible in the platform. Palo Alto Networks' Prisma AIRS now integrates with Foundry in preview, enabling policy-driven runtime inspection of prompts and model responses to help detect risks such as prompt injection and toxic content.

Zenity integration is also in preview, extending built-in guardrails and controls in Foundry Control Plane with runtime security capabilities for AI agents. By analyzing agent intent and execution paths as agents interact with tools and systems, Zenity helps identify and stop unexpected actions, unsafe tool usage, or unintended data access before they impact production systems.

Unified Microsoft Foundry Portal

The new Microsoft Foundry portal at ai.azure.com is now generally available, providing a unified UI experience for teams to build, optimize, and govern AI apps and agents in one place. The portal brings together Foundry Agent Service, Foundry Models, Foundry Tools, Foundry IQ, and Foundry Control Plane with integrated governance.

This unified interface helps teams move from experimentation to production with the right controls in view, streamlining the development lifecycle and reducing the complexity of managing multiple AI services.

Real-World Impact

Corvus Energy, which powers battery systems across 1,500+ maritime vessels, demonstrates Foundry's practical value. The company uses Foundry and Microsoft Fabric to move from manual inspection workflows to agent-driven operational intelligence across their global fleet. Andreas Hunderi, VP of Global IT at Corvus Energy, notes that "Microsoft Foundry has delivered real value for our team. The updated interface makes our services and workflows faster, more capable, and easier to use — and we're now able to connect things that simply weren't possible before."

Looking Ahead

With Foundry Agent Service, Observability in Control Plane, and the Foundry portal now generally available on a production-ready backend, Microsoft is focusing on making the platform even easier to use. Future developments include deeper workflow authoring, richer guidance, extended governance, and flexible integration into existing enterprise ecosystems.

The goal remains simple: any model, any tool, any framework — governed, observable, and ready for production. As AI agents become increasingly central to enterprise operations, platforms like Foundry provide the foundation for building intelligent systems that can operate securely and reliably at scale.

For teams ready to get started, Microsoft offers comprehensive resources including the Foundry SDK, developer blogs, Microsoft Learn courses, and the AI Builder's Guide to Agent Development and Management whitepaper. The platform's general availability represents a significant step forward in making enterprise-grade AI agent development accessible to organizations of all sizes.