User Profile
AmirJafari
Microsoft Employee
Joined 3 months ago
User Widgets
Contributions
Retirement of Fabric data agent integration in Copilot in Power BI
Nearly a year ago, we announced the integration between Fabric data agent and Copilot in Power BI, allowing users to access a Fabric data agent directly from Copilot in Power BI. This integration enabled users to explore report data and interact with data agents within the same workflow. When you asked a question about your data, Copilot worked with you to understand your topic of interest, then used that context to identify the right data agent, along with other data sources, to answer from. We are retiring this integration Beginning August 26, 2026, the integration between Copilot in Power BI and Fabric data agents will be retired. Copilot in Power BI will no longer connect directly to Fabric data agents. After that date, users will no longer be able to ask Copilot in Power BI questions that are answered through a Fabric data agent. This change does not affect the Fabric data agent experience itself. Fabric data agents remain available through other supported experiences and continue to provide access to the data and capabilities through those experiences. Why we are making this change The Copilot in Power BI integration was built on the OpenAI Assistant API. As that API is being retired, we are retiring this integration as well. Rather than investing in a replacement for this specific integration, we are focusing on supported experiences that continue to provide access to Fabric data agents today. focusing on supported experiences that continue to provide access to Fabric data agents today. Where you can continue to use Fabric data agents Fabric data agents remain available through different supported experiences across the Microsoft ecosystem and beyond, including: Directly in the Fabric data agent experience Microsoft Foundry Microsoft Copilot Studio Microsoft 365 Copilot A Model Context Protocol (MCP) server endpoint, which enables integration with your own supported apps and tools. The underlying Fabric data agent remains unchanged by this retirement. Security controls and permissions continue to be enforced in accordance with the configuration and capabilities of the experience being used, including your data sources, security rules such as row level security and column level security, and permissions. Planning your transition If you currently access Fabric data agents through Copilot in Power BI, consider transitioning to one of the supported experiences previously listed. The most appropriate option will depend on how your organization uses Fabric data agents today and the workflows you want to support. For example, organizations already using Microsoft 365 Copilot may choose to access data agents there, while organizations building custom experiences may find Microsoft Copilot Studio, Microsoft Foundry, or MCP-based integrations better aligned to their needs. Questions? Thank you for building with us and for your continued feedback. We remain committed to supporting Fabric data agents through supported experiences across the Microsoft ecosystem and beyond. If you have questions, please reach out and we will help you find the right fit.Fabric data agents in Microsoft Foundry: Easier to connect, easier to trust
Last year, we announced the integration between Fabric data agent and Microsoft Foundry. The idea was simple: build a data agent in Fabric on top of your data in Fabric OneLake, then bring that agent into Foundry so your Foundry agents can answer questions grounded in your enterprise data. That foundation helped teams connect their data to AI agents, and your feedback highlighted two areas where we could make the experience better: simplifying how to connect a Foundry agent to Fabric data agents and improving visibility into how they operate once deployed. Today, we’re introducing a set of enhancements that make Fabric data agents easier to discover, integrate, and monitor in production. This update helps organizations build AI agents that can access trusted business data with greater confidence and transparency. Bringing Fabric data agents into Foundry through Model Context Protocol Fabric data agents now integrate with Foundry through Model Concept Protocol (MCP), an emerging industry standard for agent-to-tool communication. This means your Fabric data agents appear as tools that Foundry agents can invoke when they need access to enterpise data that lives in Fabric OneLake. When your Foundry agent decides it needs data, it calls the Fabric data agent the same way it would call any other tool. MCP provides a standard way to connect Fabric data agents and Foundry agents, making the integration easier to extend over time without requiring changes to how you configure it. Importantly, this means organizations can combine the strengths of both platforms: Fabric provides trusted, governed business context, while Foundry provides the platform for building and orchestrating agents. Together, they enable developers to create AI experiences grounded in enterprise data. Discover and add agents from the OneLake Catalog One of the most common pieces of feedback we received was that connecting a Fabric data agent required too much manual work. Previously, builders needed to locate workspace IDs and artifact IDs before they could connect a data agent to their Foundry agent. If you had more than one agent, or you were not the person who built it, tracking down those IDs was cumbersome. The following figures display the previous experience of selecting the Fabric Data Agent tool and filling in the IDs manually. Figure: The old experience: selecting the Fabric Data Agent tool. Figure: The old experience: entering the workspace ID and artifact ID manually. Now, data agents can be discovered directly through the OneLake Catalog experience. You simply add the Fabric IQ (OneLake Catalog) tool, filter for data agents, and pick the ones you want. The following figures display the new experience. Figure: The new experience: adding the Fabric data agent via the Fabric IQ (OneLake Catalog) tool. Figure: The new experience: browsing data agents in the OneLake Catalog and adding them. No more searching for IDs. You see the agents by name, along with their location and sensitivity, and you add them with a click. If you have permission to use an agent, it’s right there in the list. Connect multiple data agents to a single Foundry agent Business questions rarely live within a single domain. The previous integration allowed only one Fabric data agent per Foundry agent. The limitation often forced builders to choose which data source to prioritize. Now, you can connect multiple Fabric data agents to a single Foundry agent, allowing the agent to draw from specialized sources of business knowledge. For example, a Foundry agent can simultaneously leverage a sales analytics agent, a supply chain agent, and a customer support agent, selecting the most relevant source based on the user’s question. Tracing and logs through Foundry Observability Adding a data agent is one thing. Understanding what it does once it is running is another, and that is where the second part of this update comes in. When you add a Fabric data agent to your Foundry agent, you can now view logs and traces through Foundry Observability. This provides insight into how requests move through the agent workflow, including which tools were invoked, how long individual steps took, and what results were returned. Figure: Foundry agent trace with the Fabric data agent's tool calls, input, and output. Observability is important because agents operate as chains of decisions rather than single transactions. When an answer appears incorrect or a workflow becomes slow, builders need visibility into the entire execution path, not just the final output. Traces show you the path a request took and where time was spent, while logs provide detailed records of what occurred during execution. As organizations move agents from experimentation to production, this level of visibility becomes critical. Teams need reliable ways to troubleshoot issues, understand agent behavior, and maintain confidence in the systems they deploy. Foundry Observability helps makes that possible. Figure: Span metadata for a single tool call: timing, status, and data source. Where we’re headed Our goal is simple: help organizations build AI agents they can confidently run in production. That means making enterprise knowledge easier to discover, simplifying how agents connect to trusted business data, and providing the observability needed to understand and troubleshoot agent behavior at scale. With MCP-based integration, OneLake Catalog discovery, support for multiple data agents, and deeper observability, we’re continuing to enhance Fabric and Foundry as a unified foundation for enterprise AI. This update is rolling out to all regions over the coming days. To learn more about connecting Fabric data agents with Microsoft Foundry, explore the Observability for Fabric data agents in Microsoft Foundry documentation.Fabric data agent API is now public: Build Fabric data agents into your tools and pipelines
The Fabric data agent API is now public. Developers can now automate how Fabric data agents are created, configured, updated, and published from their own tools, pipelines, and backend services. The primary purpose of the management-plane public API is lifecycle management: managing and updating data agent configuration in a supported, repeatable way. With new public API support, the Fabric data agent SDK can run outside Fabric, giving teams a consistent way to manage data agents programmatically while keeping the in-product experience unchanged. What changed? Until now, most of what you did with a Fabric data agent happened inside Microsoft Fabric. You created a data agent in the Fabric portal, wired up your data sources, configured the agent, and tested it. You could also use the data agent SDK, but only from within a Fabric notebook. Either way, the work stayed in one place. With this update the Fabric data agent SDK now runs on the Fabric public API. After authenticating to Fabric, you can now programmatically: Create and manage data agent artifacts. Add, remove, and configure data sources. Configure agent instructions, data source instructions, and example queries. Update and publish data agents from your own code. Integrate data agent management into local development, CI/CD pipelines, internal portals, containers, Azure Functions, and backend services. Manage and update data agent configurations for lifecycle management. Why this matters If you’re only working in the Fabric portal, this update won’t change your day-to-day. The impact mostly shows up once you need to automate or scale — when managing data agents across environments, tools, or pipelines becomes part of your workflow. A few examples follow: You manage many workspaces and want to provision the same data agent shape across all of them without clicking through each one. You are building an internal tool or portal where your team manages data agents without ever opening Fabric. You want to plug data agent provisioning into your existing CI/CD pipeline. You are an ISV building on Fabric and you want data agents to be part of your product, not a side trip for your customers. That is what a public API is for. It turns a clickable feature into a building block. How this fits with the rest of Fabric The SDK is built on the Fabric public API, the same REST surface you already use for workspaces and items. It uses the same authentication and the same patterns, so it fits naturally next to the rest of your Fabric automation. It is useful to separate management from runtime consumption. The public API and SDK are the management-plane surface for creating, configuring, updating, and publishing data agents. After a data agent is published, the MCP endpoint is the main runtime and consumption endpoint for querying it from tools, applications, and agent experiences. This complements Fabric Git integration and deployment pipelines. Git and deployment pipelines help teams track data agent configuration as files and promote changes across development, test, and production environments. That is set up at the workspace level. The public API adds a programmatic management layer so teams can create, update, configure, and publish data agents directly from their own code. 1. Authenticate to Fabric (from your local machine, outside Fabric) from azure.identity import AzureCliCredential from fabric.analytics.environment.credentials import ( SetFabricAnalyticsDefaultTokenCredentialsGlobally, ) SetFabricAnalyticsDefaultTokenCredentialsGlobally(AzureCliCredential()) 2. Create a data agent in your workspace from fabric.dataagent.client import create_data_agent # WORKSPACE_ID: the Fabric workspace that will own the new data agent WORKSPACE_ID = "<your-workspace-id>" agent = create_data_agent( data_agent_name="Quickstart Data Agent", workspace_id=WORKSPACE_ID, ) 3. Configure the data agent and add a data source # AGENT_INSTRUCTIONS: the agent's system prompt (what it is for, tone, formatting rules) AGENT_INSTRUCTIONS = "<your agent instructions>" # DATASOURCE_ID: the artifact ID of an existing lakehouse, warehouse, semantic model, or KQL database DATASOURCE_ID = "<your-datasource-id>" agent.update_settings(ai_instructions=AGENT_INSTRUCTIONS) agent.add_staging_datasource( artifact_name_or_id=DATASOURCE_ID, workspace_id_or_name=WORKSPACE_ID, ) 4. Publish the data agent agent.publish_staging(description="Initial publish") Figure: The data agent SDK running in VS Code on a local machine, authenticated to Fabric. Getting started To get started, authenticate to Fabric using user or service principal token, select a workspace, and use the data agent SDK to create your first data agent from a script. You can find the Fabric data agent SDK and the Fabric REST API reference in the documentation, including the Data Agent Items APIs for managing the agent lifecycle and the Staging APIs for editing data sources and example queries. Start small: add one data source, configure instructions and example queries, publish the agent, and then use the MCP endpoint as the main consumption endpoint for querying the published data agent from your tools or applications. Enabling Fabric data agents to be programmable outside Fabric is a foundational step. It gives developers and teams a supported and consistent way to build data agents into their own tools, pipelines, and products. We look forward to seeing what you build with it.Update to required permissions for Semantic Models in Fabric Data Agents
We’re announcing an update to the permissions required to interact with semantic models in Fabric data agents. Today: As a creator, you must have access to the workspace where the semantic model lives and Build permission on the semantic model to add semantic model to a data agent. As a consumer, you need Read access to the semantic model to use a data agent to ask questions over the semantic model. We’re simplifying this experience. With this update, creators and consumers only need Read access on the semantic model to interact with it through a data agent. You no longer need workspace access to add a semantic model to a data agent or to ask questions from it. If you have Read access, you can add and use the semantic model within the data agent experience. This change applies only to interactions through the data agent. If you need to modify the semantic model or use features such as Prep for AI, you will still need Write permission on the semantic model. This update will take effect on April 6th. We’re sharing this in advance so you can review and adjust permissions if needed before the change goes live.Fabric Data Agent now supports CI/CD, ALM Flow, and Git Integration
Fabric data agents now support CI/CD, ALM flow, and Git integration. These enhancements help you implement best practices for managing your data agent artifacts and tracking changes, leading to more reliable and collaborative development. Benefits of CI/CD and ALM Flow Fabric data agents support a variety of data sources, including Lakehouse, Warehouse, Power BI Semantic Models, and KQL databases. Managing these configurations effectively is critical as they evolve. CI/CD and ALM provide a structured way to handle updates such as schema selections, data source configurations, example queries, and custom instructions for each data source. All changes are stored in a Git repository, which ensures that every update is tracked, reviewable, and reversible. Meaning you can apply standard source control practices—committing changes, opening pull requests for review, and rolling back if necessary. By integrating with Git, CI/CD and ALM turn the management of data agents into a repeatable, auditable process that aligns with modern software engineering practices. Git Integration for Version Control and Collaboration Git integration brings versioning and collaboration directly into the lifecycle of a data agent. Every modification—whether you add or update example queries for a data source, provide instructions for the agent, or define instructions specific to each data source—they are all recorded with full history of what changed, when, and by whom. If an issue arises, you can quickly revert to a previous version. Collaboration also becomes streamlined. Team members can experiment independently and safely on separate branches, validate changes, and merge them into the main branch once approved. This workflow reduces risk, accelerates development, and ensures that updates are introduced in a controlled, transparent manner. Team members can version and track different updates through Git, giving them a reliable history of changes and making it easier to collaborate across the organization. Managing Data Agent Changes Across Environments Git version control not only tracks your changes but also aligns naturally with application lifecycle management (ALM) practices, where changes move through dedicated environments before reaching production. A recommended approach is to separate your work into three dedicated workspaces: Development workspace—where you design, build, and test updates to enhance the data agent’s performance. This includes modifications to agent instructions, data source instructions, and sample queries etc. Test workspace—used to validate updates in an environment that closely resembles production and consumption channels. Production workspace—dedicated to running the data agents relied on by your end users and stakeholders across consumption channels. For example, if your team updates the data agent to support a new data source, you can test this change in the development workspace, validate it in test workspace, and then deploy it to production workspace without disrupting existing users. This separation ensures all changes are thoroughly validated before reaching the production environment. With Fabric’s deployment pipelines, you can transfer changes between workspaces efficiently and safely, reducing risk and supporting consistent, reliable operations. By combining CI/CD, ALM flow, and Git integration, you gain a predictable, auditable, and scalable way to manage changes to the Fabric data agent. Getting Started To get started with CI/CD, ALM flow, and Git integration for your Fabric data agent, visit our documentation at Fabric data agent | Microsoft Learn for step-by-step instructions.Fabric Data Agents + Microsoft Copilot Studio: A New Era of Multi-Agent Orchestration (Preview)
Additional authors: Joanne Wong The preview of the integration between Fabric data agents and Microsoft Copilot Studio is now available, introducing a robust capability that enables agents to interoperate seamlessly across tools. This enhancement is designed to streamline the development, deployment, and scaling of intelligent agents across enterprise data environments, reinforcing automation and extensibility within the Fabric ecosystem. What are Fabric Data Agents? Data agents in Microsoft Fabric are AI-powered assistants that go beyond simple data retrieval from OneLake—they engage in natural language conversations about it. These agents can understand your enterprise data schema, enforce your governance policies, and interpret your business context to surface insights that are timely, relevant, and actionable. By embedding Fabric data agents into Copilot Studio, organizations can now build agents that are not only intelligent but also deeply informed by their most trusted data sources. How the Fabric Data Agent Integration with Copilot Studio Works Connecting your custom Copilot agent to a Fabric data agent enables agent-to-agent collaboration using Model Context Protocols (MCP), a framework that lets multiple agents share goals, memory, and reasoning context. This allows the agents to work together more effectively and deliver richer, more complete responses, reflecting a growing trend in AI systems. In connected-agent scenarios where collaboration is key, your Copilot agent will tap into the Fabric data agent’s deep, data-grounded expertise to deliver richer, more accurate, and fully contextualized responses across your organization’s queries. You can further enrich your Copilot agent by: Adding knowledge sources (like SharePoint, public websites, or internal documents). Integrating tools. Extending its capabilities to handle domain-specific queries or additional task flows. Once configured, you can publish your Copilot agent to multiple channels, including Microsoft Teams. What Is Multi-Agent Orchestration? Previously, AI agents operated in silos—each handling a specific task with limited collaboration. With multi-agent orchestration, Microsoft Copilot Studio now enables agents, to delegate tasks, collaborate across systems, and work in concert to achieve complex, business-critical goals. Imagine a scenario where: A Fabric data agent pulls real-time sales data from your enterprise data lake. A Microsoft 365 agent uses that data to draft a proposal in Word. An Azure AI agent schedules follow-up meetings in Outlook and updates your CRM. This seamless handoff between agents can completely transform workflows for the enterprise. Why this matters This integration is a strategic shift in how enterprises employ agentic AI; it empowers business users—many of whom work primarily in Teams—to ask questions and get data-driven answers without ever leaving their chat window or needing direct access to Fabric or Copilot Studio. Now your entire organization can benefit from these AI experts right where your users work every day. Scale AI across departments without duplicating effort. Ensure data consistency and compliance by centralizing access through Fabric. Accelerate innovation by enabling agents to collaborate like teams would. Getting started Now in preview, the Fabric data agent integration with Copilot Studio will roll out globally in the coming days. We encourage you to start exploring how your existing agents can be connected and enhanced through this new capability. Dive into the documentation and start building! Resources Check out the demo to learn how to build multi-agent solutions with Fabric data agents. Join the community discussion, report issues and suggestions, and share your feedback.Extracting deeper insights with Fabric Data Agents in Copilot in Power BI
Co-author: Joanne Wong We’re excited to announce the upcoming integration of Fabric data agent with Copilot in Power BI, enhancing your ability to extract insights seamlessly. What's new? A new chat with your data experience is launching soon in Power BI-- a full-screen Copilot for users to ask natural language questions and receive accurate, relevant answers from their available Fabric resources. As part of this launch, users can also discover and leverage Fabric data agents directly in the Copilot in Power BI experience. With this integration, you no longer need to jump between different items to find what you're looking for. It simplifies your workflow by helping you focus on extracting insights instead of navigating through various sources. Previously, you may have access to multiple resources, but finding the right data to answer specific questions can be challenging. Copilot was limited to the right pane of a single report, allowing questions only about that open report. This new standalone Copilot in Power BI addresses this by streamlining the process. Now, when you ask a question, Copilot scans the Power BI semantic models, reports, and Fabric data agents you have access to. It ranks and suggests the most relevant items based on your query, giving you options to choose from. How does it work? Fabric data agents are key when your question involves accessing data from other sources in Fabric, such as lakehouses, warehouses, or KQL databases in addition to Power BI semantic models. Fabric data agent serves as the bridge to these diverse data sources, enabling you to retrieve relevant information from OneLake for your queries. Moreover, if you already know which data agent to use, you can manually add it to the Copilot session and interact with it directly for more relevant results. Once Copilot determines that the Fabric data agent is necessary, it rephrases the query for clarity, sends it to the data agent, and retrieves the answer from the most relevant data source—whether it’s a lakehouse, warehouse, semantic model, or KQL database. Security protocols, such as Row-Level Security (RLS) and Column-Level Security (CLS), are enforced based on your permissions, and the answer is then delivered directly in the Copilot conversation interface. The integration of Fabric data agent with Copilot in Power BI takes natural language querying to the next level, making it easier to find insights without switching between resources. Whether you know exactly what data you need or if you want to rely on Copilot’s suggestions, this integration helps you focus on gaining insights from your data. Getting Started Stay tuned for the launch of the standalone Copilot in Power BI experience in the coming weeks. To create a Fabric data agent, explore the documentation. When you have access to the standalone Copilot experience, turn on the 'Users can access a standalone, cross-item Power BI Copilot experience' tenant setting. Once the tenant setting is enabled, you can manually add Fabric data agent(s) to your Copilot session. What’s Next? Explore the documentation to learn more about Fabric data agents. Read more about the latest updates in Copilot in Power BI. Learn more about how Copilot in Fabric works. Watch our Copilot and AI in Fabric session at Build.Empowering agentic AI by integrating Fabric with Azure AI Foundry
Additional author: Joanne Wong Data plays a critical role in agentic AI, enabling AI agents to operate independently, make informed decisions, and take meaningful actions. And that's why we are expanding capabilities and deepening integrations between our data and AI platforms. Starting today, organizations can use Azure AI Foundry to connect customized, conversational agents, created in Fabric. A critical component of Azure AI Agent Service is the ability to securely ground AI agent outputs in enterprise knowledge, so that responses are accurate, relevant, and contextually aware. Data agents in Fabric, formerly known as AI skills, can retrieve knowledge across different data sources - from lakehouse and warehouse data to Power BI semantic models and KQL databases - using a number of specialized query language tools that help AI to generate SQL, KQL and DAX to extract, process, and present data effectively and delivers precise, actionable insights. Fabric data agents can determine when to use specific data, how to combine it, and what insights matter most. By combining Fabric’s sophisticated data analysis over enterprise data with Azure AI Foundry’s cutting-edge GenAI technology, businesses can create custom conversational AI agents leveraging domain expertise. This seamless integration enables organizations to develop agents that are not only based on unstructured data in Azure AI Search or SharePoint but also integrate with structured and semantic data in Microsoft OneLake, thereby enhancing data-driven decision-making. Customers like NTT DATA are leveraging data agents in Microsoft Fabric to extract real time insights. NTT DATA built a suite of HR-focused data agents, helping users interact directly with real-time data to uncover patterns in staffing, chargeability, and productivity. “We see data agents as a conversational capability layer we can use to ‘talk’ to our data, understand it, and derive different insights in support of our daily decision making. By significantly improving real-time actionable insights, Azure AI and Fabric help elevate business outcomes as well as human potential.” Genis Campa, Head of Data Products Strategy, NTT DATA We invite you to explore the preview and experience firsthand how this unified approach can transform your data into a powerful asset for more insightful decision-making. Learn more about this integration in the Azure AI blog. Next steps Join the conversation on the Fabric Community Learn more in the Fabric documentation Give us your feedback on Fabric IdeasAnnouncing Public Preview of the "Share" Feature for Fabric AI Skill
The Share capability for Fabric AI Skill is now in public preview. This feature enables AI Skill creators to share their AI Skill with colleagues, allowing them to consume the AI Skill and gain valuable insights from the data. How Sharing Works Once you have created and validated the effectiveness of your AI Skill, you can first publish it and then share it with your team or colleagues who want to gain insights from the data. The default 'Share' capability allows others to simply query the published version of the AI Skill, enabling them to ask questions and receive responses. This simplifies how team members can consume the AI Skill for Q&A, helping them leverage data insights for informed decision-making. Beyond querying, you can also share your AI Skill with additional permissions: Co-create (Edit and view details): Collaborate with colleagues by inviting them to co-develop the AI Skill. View Configuration (View details): You can grant viewing access to the AI Skill’s configuration settings, allowing others to understand its setup without making any changes. These flexible sharing options empower you to control how your AI Skill is used and developed, making it a valuable tool for collaboration and data-driven insights. Additionally, please note that when sharing the AI Skill, you also need to share access to the underlying data it uses. The AI Skill honors all user permissions to the data, including Row-Level Security (RLS) and Column-Level Security (CLS). These flexible sharing options empower you to control how your AI Skill is used and developed, making it a valuable tool for collaboration and data-driven insights. Enhanced Publishing Experience To further support your work, when you publish your AI Skill, you can now include a description that outlines its purpose and functionality. This description will help users understand what your AI Skill does, making it easier for colleagues to leverage it effectively for their data queries. Streamlined Version Management Additionally, we have introduced flexible version management. After publishing the AI Skill, you can switch between the published version and the current version you are developing. This allows you to run the same set of questions on both versions, facilitating performance comparisons and providing insights into how your updates impact the AI Skill’s performance. More Ways to Use AI Skill We are actively working on additional ways for you to leverage AI Skill across different platforms. Soon, you’ll be able to integrate AI Skill with Microsoft Copilot Studio, Microsoft Azure AI Studio, and more, providing even greater flexibility and accessibility to consume AI Skill within the Microsoft ecosystem. We’re also developing ways for you to use AI Skill within your custom applications, even those outside of Microsoft ecosystem. These future integrations will allow you to seamlessly incorporate AI Skill into your existing workflows, unlocking new possibilities for customized and context-specific AI experiences. We encourage you to explore AI Skill and let us know what you think. You can learn more about AI Skill here and you can fill up this form to let us know how else you envision to use AI Skill.
