Microsoft Foundry Adds Native Voice Agents With GPT Realtime

Microsoft is bringing real-time voice agents to Microsoft Foundry, allowing enterprises to build, deploy, monitor, and evaluate voice-based AI through the same Agent Service platform used for other agents.

The new capability is available in public preview for prompt-based and hosted agents. Microsoft has integrated voice directly into Foundry Agent Service instead of adding a separate speech layer on top of a text-based agent.

Voice agents support GPT Realtime, MAI, and custom models

Developers can build voice agents using GPT Realtime, Azure Realtime, Microsoft’s MAI models, and custom models.

Microsoft says the agents support more than 80 languages and 140 locales. They can also handle natural turn-taking and interruptions during conversations.

Enterprises can customize their agents with domain-specific speech models, custom voices, and photo or video avatars.

Voice agents can use existing enterprise tools and data

Voice agents can access the same enterprise data, tools, and governance policies already configured for other Microsoft Foundry agents.

Companies can deploy them through web apps, Microsoft Teams, Teams Phone, and Twilio-powered inbound and outbound calling systems.

Microsoft Foundry also provides voice-specific tracing, monitoring, transcripts, and evaluation tools, giving organizations more ways to track agent performance and interactions.

The addition expands Foundry beyond traditional text-based AI agents and builds on Microsoft’s earlier work with hosted agents.

In other AI news

Microsoft has reportedly killed the Copilot+ PC branding without providing a clear explanation for the change.

The company is also rolling out Claude Opus 5.5 and GPT-6 Sol to Copilot across Word, Excel, and other Microsoft apps.

Meanwhile, Meta Muse has reached around 700,000 daily users as reports of service issues have started to appear.