Copilot Studio: Use Azure Foundry Models

Key insights

• Bring your own model from Azure AI Foundry now runs inside Copilot Studio.
  Microsoft demonstrated how makers can plug their deployed chat‑completion models into Copilot Studio prompts to power agents and apps.
• Azure AI Foundry model catalog gives access to thousands of models, including popular and task‑specific options.
  Developers can pick public models or use custom fine‑tuned models already deployed to Foundry.
• Model Choice and Data Grounding boost relevance: choose the model that fits your domain and combine it with enterprise data sources for context‑aware answers.
  This improves accuracy for specific workflows and compliance needs.
• Create an agent, select a model, build prompts, and deploy to Microsoft 365 apps.
  The process stays inside Copilot Studio so low‑code teams can iterate faster and ship AI assistants to Teams, SharePoint, and more.
• BYOM and Multi‑model access unlock flexibility, and metered billing lets organizations pay based on usage.
  Enterprises gain control over model behavior, costs, and governance while scaling securely on Azure.
• Makers and enterprises benefit from faster prototyping, better domain fit, and unified management.
  The integration shortens development time for specialized assistants while keeping security and monitoring under Azure policies.

Overview of the demo and origin

The original post, authored by Microsoft, summarizes a recorded demo that shows how to bring your own model from Azure AI Foundry into prompt workflows inside Copilot Studio. The video comes from a Microsoft 365 & Power Platform community call held on July 22, 2025, and it demonstrates a step-by-step scenario for makers and developers. As a result, the demo highlights practical steps and decisions required to connect models to agents, flows, and apps, while also showcasing structured outputs and support for image prompts.

Importantly, the demo aims to bridge two parts of Microsoft’s AI tooling: the model catalog and the no-code/low-code experience. Consequently, viewers can see how teams might move from model selection to live deployment without leaving the Copilot Studio environment. This makes the material relevant both to technical teams and to business users who want to adopt AI into everyday workflows.

What the demo showed and setup steps

First, the presenter creates an agent in Copilot Studio and connects that agent to the target scenario, such as Microsoft Teams or SharePoint. Next, within the prompt configuration, the demo selects models located in Azure AI Foundry, including options for custom fine-tuned models that organizations have uploaded. Then the workflow demonstrates building prompts that call the chosen model and configuring the agent to return structured outputs or images as needed.

After configuring the prompt and model, the demo deploys the agent across Microsoft 365 endpoints to show how a user would interact with it in live contexts. Along the way, the speaker explains authentication, billing considerations, and how model selection affects runtime behavior. Thus, the video serves as a practical guide for teams planning to integrate purpose-built models into business applications.

How the integration works in practice

The integration relies on a catalog model lookup and a prompt binding in Copilot Studio that routes requests to the chosen model instance in Azure AI Foundry. In practice, this means prompts in Copilot Studio can call thousands of catalog models or a custom model your team owns, while the studio handles orchestration and response formatting. Consequently, the approach simplifies experimentation because makers can swap models without rebuilding the agent from scratch.

Moreover, the system supports structured outputs and image prompts, enabling more complex interactions than simple text responses. However, teams must design prompts carefully to ensure reliable, predictable outputs, especially when combining multiple model calls or agents. Therefore, practical testing and input validation remain essential before broad deployment.

Benefits and common uses

One major benefit is model choice: teams can select from a broad set of models to match domain needs, accuracy requirements, or cost targets. Additionally, combining model selection with Copilot Studio’s low-code interface reduces the time from idea to deployment, so business users can iterate faster while IT retains governance. Consequently, organizations can build tailored assistants for customer support, document summarization, and workflow automation more easily than before.

The integration also improves data grounding because Copilot Studio agents can leverage enterprise indexes and contextual sources to generate more accurate responses. At the same time, Azure’s infrastructure offers enterprise security, monitoring, and flexible billing, which makes scaling to many users more feasible. Thus, the combination appeals to teams that need both agility and governance.

Tradeoffs, challenges, and considerations

Despite clear benefits, the approach requires tradeoffs around cost, complexity, and testing. For example, choosing a powerful model may improve accuracy but will increase runtime costs and latency, so teams must balance performance against budget. Furthermore, customized models promise better domain fit but add overhead for fine-tuning, validation, and ongoing maintenance.

Another challenge lies in ensuring consistent, safe outputs across different models and update cycles, which demands robust prompt design, monitoring, and fallback logic. Likewise, integrating models from a large catalog increases the need for governance policies to control data usage, access, and compliance. Therefore, teams should plan for ongoing operations, including metering, auditing, and iterative quality checks before scaling broadly.

Implications for organizations and next steps

Overall, this demo demonstrates a practical path to bring custom AI models directly into business-facing agents, which could accelerate adoption of AI across Microsoft 365 environments. Moving forward, organizations will need to align technical owners and business sponsors, select appropriate models, and define governance to manage costs and risks. As a result, teams that invest in careful testing and monitoring can reap faster innovation while maintaining enterprise controls.

In short, the video from the community call presents a clear workflow and shows how Copilot Studio and Azure AI Foundry can work together to deliver tailored AI experiences. Consequently, this integration is worth watching for teams planning to deploy conversational agents and automation at scale within Microsoft 365.

Keywords

bring your own model azure foundry, azure foundry byom, copilot studio custom model, prompt engineering copilot studio, deploy custom model azure, integrate azure foundry copilot, BYOM copilot studio, use custom LLM in Copilot