Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Real-time agents enable natural, context-aware conversations across voice and digital messaging channels. Customers can interact directly with an AI-powered agent and receive immediate, relevant responses without relying on traditional interactive voice response (IVR) menus or scripted experiences. For voice channels, real-time agents support fully voice-driven interactions over PSTN and SIP, including dual tone multifrequency (DTMF) input, barge-in, voice activity detection (VAD), and other telephony capabilities.
You can use the same real-time agent configuration across both voice and digital messaging channels, providing a consistent experience regardless of how customers choose to engage.
Note
- Real-time agents is in preview for digital messaging channels. Preview features aren't meant for production use and might have restricted functionality. These features are subject to supplemental terms of use, and are available before an official release so that customers can get early access and provide feedback.
- Learn about basic agents in Basic agents overview.
- Learn more about developing agents in Overview of real-time agent design optimization.
The following table describes the capabilities provided by real-time agents:
| Capability | Description |
|---|---|
| Natural language understanding | Understands customer intents without requiring specific phrases or menu options. |
| Multi-channel deployment | Deploy the same agent to voice, digital messaging (preview), or both. |
| Voice-native design | Purpose-built for spoken interaction with a natural conversation flow. Applicable to voice channel only. |
| Real-time responsiveness | Minimal latency with instant turn-taking and natural pauses. |
| Context awareness | Maintains conversation state and references previous interactions. |
| Multilingual support | Understands and responds in multiple languages. |
| Flexible integration | Connects with CRMs, knowledge bases, APIs, and Power Automate flows. |
| Deterministic control | Combines AI conversation with structured topic flows for compliance and consistency. |
| Text LLM voice model | Uses a text-based voice architecture that converts speech to text, uses GPT-5-Chat (Preview) model to understand intent and generate responses, and then synthesizes the response back to speech. This model supports advanced voice customization through Microsoft Neural Text-to-Speech (TTS), including custom voice models and broader deployment flexibility. |
Prerequisites
Before you begin, ensure you have the following license, role, and permission prerequisites in place:
- Dynamics 365 Contact Center with the following channels configured per your requirement. To learn more, see Provision channels in Dynamics 365 Contact Center.
- Voice channel and call routing
- Digital messaging channels
- Omnichannel administrator role with permissions to configure voice or messaging channels and the required workstream.
- Copilot Studio for agent creation and configuration.
- Copilot Studio maker role with permissions to create and configure agents.
Note
- For the most complete and up-to-date Copilot Studio licensing and billing information, refer to the Microsoft Copilot Studio Licensing Guide.
- For the most complete and up-to-date Dynamics 365 Contact Center licensing and billing information, refer to the Microsoft Dynamics 365 Licensing Guide.
Regional prerequisites
This feature uses GPT-Realtime, GPT-Realtime-Mini (Preview), or GPT-5-Chat (Preview).
- GPT-Realtime is hosted in North America and might process data outside the AI resource's Azure geography when you use a Global Standard deployment.
- GPT-Realtime-Mini (Preview) is hosted in Australia and might process data outside the AI resource's Azure geography when you use a Global Standard deployment.
- GPT-5-Chat (Preview) supports in-region data processing in the United States, European Union, United Kingdom, and Australia when deployed in a supported region.
Data storage stays in the AI resource's Azure geography.
Learn more in Understanding deployment types in Microsoft Foundry Models and Data, privacy, and security for Azure Direct Models in Microsoft Foundry.
Note
As of July 2026, the following regional limitations apply:
- North America customers can use GPT-Realtime and GPT-5-Chat (Preview) without extra regional configuration.
- Australia customers can use GPT-Realtime-Mini (Preview) and GPT-5-Chat (Preview) without extra regional configuration.
- EU Data Boundary customers can use GPT-5-Chat (Preview) with in-region data processing but can't use GPT-Realtime or GPT-Realtime-Mini (Preview) because those models require cross-geo processing.
- UK customers can use GPT-5-Chat (Preview) with in-region data processing, but GPT-Realtime and GPT-Realtime-Mini (Preview) require cross-geo processing.
- Customers in other regions must allow cross-geo processing for GPT-Realtime and GPT-Realtime-Mini (Preview). GPT-5-Chat (Preview) regional availability depends on deployment region support.
Responsible AI and compliance notice
Microsoft's safety and responsible AI evaluations found safety-relevant behavioral limitations that might result in a higher risk that the agent produces potentially harmful content. Customers should take steps to mitigate these risks as further detailed in the transparency note for real-time agents.
You're solely responsible for using Dynamics 365, this feature, and any related feature or service in compliance with all applicable laws. Customers are also responsible for complying with Microsoft policies. These policies include the Microsoft Enterprise AI Services Code of Conduct.
Known limitations
The following limitations apply to real-time agents for digital messaging channels while the feature is in preview:
- Authentication: OAuth and Microsoft authentication aren't supported. Configure the agent with No authentication. Learn more in Configure user authentication.
- Conversation length: Conversations that exceed 60 minutes automatically escalate to a human agent.
- Analytics: Copilot Studio analytics reports might be incomplete for real-time agents. Learn more in Analytics overview.
Conversation transcript logging for Text LLM voice agents is limited. Some conversation turns and agent responses might not be captured in call transcripts, which can affect analytics and monitoring.