Quality of response analyzes a sample set of successfully answered generative AI questions to provide insights about the quality of generative AI responses.

The feature uses AI to check the relevance, completeness, and grounding of each response in the sample set. It assigns a quality score for each response. The feature uses the quality score to classify each response as having good or poor quality.

Makers see a visual indicator that shows the breakdown of answered generative AI questions between good and poor quality. Makers can view the reasons for poor quality responses. They can see if the response isn't fully relevant, if the response is incomplete, or if the response isn't fully grounded.

Copilot admins and makers can get actionable insights directly on the agent's Analytics page:

• See insights for unanswered Generative AI questions.
• Get insights that focus on specific themes of questions.
• Receive guidance on how to fix the issue and boost performance.
• View the list of related conversations with unanswered questions so you can easily investigate the issue.

Insights are generated daily by checking all Gen AI unanswered questions from the past seven days.
This feature doesn't support SharePoint and Graph Connector knowledge sources.

Makers can get user feedback on agent responses and use that feedback to improve an agent's performance. By default, the user feedback feature is on for all new and existing Copilot Studio custom agents. Makers can turn off this feature in the agent settings.

Users can give feedback in these channels:

• Test chat
• Demo website
• Custom website
• Webchat SDK
• Teams
• Microsoft 365 Copilot

Agent users can add comments when they send their feedback. You can see the comments in the conversation transcript table in your Dataverse store.

Analytics collects user feedback and shows the total number of reactions. It also shows how many reactions are thumbs-up and how many are thumbs-down.

You can now analyze the business impact of your Microsoft Copilot Studio agents directly in Viva Insights. This feature gives you a consolidated report that shows usage and business impact trends across all your published Copilot Studio agents. You see key stats like hours saved, usage analysis, and resolution rate at the tenant level. You can drill down to view stats for each agent. The report uses regular usage data that Copilot Studio sends to Microsoft 365.

By default, this feature is enabled if you meet the following criteria:

• You have at least 50 Microsoft 365 Copilot licenses (standalone or bundled with Viva Insights).
• You publish at least one Copilot Studio agent in a production environment.

You control the feature in the Power Platform admin center. The feature is automatically enabled for all public customers in a commercial cloud when the criteria are met.

Users with the Insights Analyst role in Microsoft 365 can view the Copilot Studio agent report in Viva Insights.

Environment admins can turn off the report from the Power Platform admin center.

If your Copilot Studio data is in a different geographic location or region from your Microsoft 365 tenant, an environment admin must turn on data sharing across geographic boundaries.

Computer use is a feature in Copilot Studio that enables your agent to interact with any system with a graphical user interface (GUI), such as websites or desktop applications. It performs actions like pressing buttons, selecting menu options, and typing into fields—just like a human user. You simply describe the task in natural language, and the agent executes it on a configured computer with a virtual mouse and keyboard.

With this feature, agents complete tasks even when no API is available. It's useful for automated data entry, invoice processing, and data extraction.

Computer use runs on Computer-Using Agents (CUA), an AI model that combines computer vision and reasoning to navigate and interact with GUIs. It adapts to interface changes, so it continues working even if buttons or layouts shift. Best of all, it requires no coding—just plain language instructions.

With this feature, makers connect to SAP through a low-code or no-code interface. They don't need to copy or ingest tabular data manually.

Data stays external in SAP, but Copilot uses real-time reasoning to answer natural language queries.

Makers pick specific Databricks tables to use as knowledge sources. Agents answer grounded questions about the data with accuracy and context. You don't need manual integration or scripting to bring the data.

Makers add the knowledge source in Copilot Studio when they create a new agent or configure an existing agent.

Makers don't need to create SQL queries. Copilot Studio uses natural language instructions to query, join, and filter the right tables.

The agent returns data and reasons over it. Then, the agent parses the data conversationally within the chat for the agent user.

The portal automatically enables this feature for you. Find it in the Agent Add Knowledge section. Use it by selecting SAP as a knowledge source.

With this capability, makers can use Confluence Spaces and Pages content in unstructured storage as a knowledge source for their agents in Copilot Studio.

This capability enables agents to provide high-quality answers based on operational data within your Confluence configuration.

Administrators can manage access to this data by configuring the Confluence connector in the Administration Portal.

Makers can connect to Confluence as a knowledge source and select unstructured Spaces and Pages content that the agent uses when generating results or answering queries. You configure connections to knowledge sources within Copilot Studio when you create a new agent or configure or update an existing agent.

Dataverse processes and indexes content so it can give accurate, detailed results to natural language queries from agent users. Users get answers only from data they can access. The system checks user access tokens against Confluence content before it returns a response. Users can access the original articles through citation references.

Scheduled sync keeps the knowledge base data up to date, within a 4-6 hour window.

With this capability, makers can use OneDrive file and folder content as a knowledge source for their agents in Copilot Studio.

This capability enables agents to provide high-quality answers based on operational data stored in files on OneDrive.

Administrators can manage access to this data by configuring the OneDrive connector in the Administration Portal.

Makers can connect to OneDrive as a knowledge source and select specific files or folders of files that the agent uses when generating results or answering queries. You configure connections to knowledge sources in Copilot Studio when you create a new agent or when you configure or update an existing agent.

Dataverse processes and indexes content so it can provide accurate, detailed results to natural language queries from agent users. Users see only answers based on data they can access. Dataverse validates user access tokens against OneDrive content before it returns a response. Users can access the original articles through citation references.

Scheduled sync keeps the knowledge base data up to date within a four- to six-hour window.

Administrators manage availability of this data by managing the OneDrive connector in the Administration Portal.

With this feature, you can connect to Oracle through a low-code or no-code interface. You don't need to manually copy or ingest tabular data.

Key Capabilities

• Real-time reasoning over external data: Copilot performs real-time reasoning and responds to natural language queries while data stays external in Oracle.
• Natural language querying: Use natural language instructions—no SQL needed—to query, join, and filter Oracle tables.
• Conversational responses: The agent reasons over data and presents it conversationally in the Copilot chat interface.
• Grounded answers with context: Select specific Oracle tables as knowledge sources. Agents answer grounded questions with accuracy and context without manual integration or scripting.
• Seamless integration in Copilot Studio: Add Oracle as a knowledge source when you create a new agent or update an existing one.
• Automatic availability: The feature automatically enables and appears in the Agent > Add knowledge section.

With this capability, makers can use Salesforce Knowledge Base content in unstructured storage as a knowledge source for their agents in Copilot Studio.

This capability enables agents to provide high-quality answers based on operational data within your Salesforce configuration.

Administrators manage access to this data by configuring the Salesforce connector in the Administration Portal.

Makers connect to Salesforce as a knowledge source and select Knowledge Base unstructured sources for the agent to use when generating results or answering queries. You configure connections to knowledge sources in Copilot Studio when you create a new agent or configure or update an existing agent.

Dataverse processes and indexes Knowledge Base content. This process gives agent users accurate, detailed results for natural language queries. Users see only answers based on data they can access. The system checks user access tokens against Salesforce content before it returns a response. Users can access the original articles through citation references.

Scheduled sync keeps the Knowledge Base data up to date. Sync happens within a 4 to 6-hour window.

With this capability, makers can use ServiceNow knowledge base content in unstructured storage as a knowledge source for their agents in Copilot Studio.

This capability enables agents to provide high-quality answers based on operational data within your ServiceNow configuration.

Administrators can manage access to this data by configuring the ServiceNow connector in the Administration Portal.

Makers can connect to ServiceNow as a knowledge source and select knowledge base unstructured sources for the agent to use when generating results or answering queries. You configure connections to knowledge sources in Copilot Studio when you create a new agent or when you configure or update an existing agent.

Dataverse processes and indexes knowledge base content. This process gives agent users accurate, detailed results for natural language queries. Users get answers only from data they can access. The system checks user access tokens against Service Now content before it sends the response. Users can access the original articles through citation references.

Scheduled sync keeps the knowledge base data up-to-date. Sync happens within a 4 to 6-hour window.

Connect Copilot Studio directly to your Snowflake data with a no-code interface. Your data stays secure in Snowflake while Copilot provides real-time responses to natural language queries.

Makers select specific Snowflake tables to use as knowledge sources. Agents answer grounded questions over the data with accuracy and context, without manual integration or scripting to surface the data. Makers add the knowledge source within Copilot Studio when creating a new agent or when configuring or updating an existing agent. Makers don't need to create SQL queries, as Copilot Studio parses natural language instructions to query, join, and filter the appropriate tables. The agent returns data and reasons over it before parsing it conversationally within the chat to the agent user.

With this capability, makers can use ZenDesk knowledge base content in unstructured storage as a knowledge source for their agents in Copilot Studio.

This capability enables agents to provide high-quality answers based on operational data within your ZenDesk configuration.

Administrators can manage access to this data by configuring the ZenDesk connector in the Power Platform admin center.

Makers can connect to ZenDesk as a knowledge source and select knowledge base unstructured sources for the agent to use when it generates results or answers queries. You configure connections to knowledge sources in Copilot Studio when you create a new agent or when you configure or update an existing agent.

Dataverse processes and indexes knowledge base content. This process gives agent users accurate, detailed results for natural language queries. Users see only answers based on data they can access. The system checks user access tokens against ZenDesk content before it sends a response. Users can access the original articles through citation references.

Scheduled sync keeps the knowledge base data up to date. Sync happens within a 4- to 6-hour window.

Understanding the efficiency of your agents and their associated costs is crucial for creating cost-effective agents. With this feature, agent creators can see how efficient their agent is and track its cost over time.

Admins can:

• Get detailed billing consumption analysis that fits their needs and business—by billed sessions or by message consumption.
• See consumption analysis over time to spot changes in consumption patterns.
• Understand the specific events that contribute to message consumption, so they can see how efficient their business's agents are and find ways to optimize them.

Analytics for autonomous agents gives you a comprehensive view of how autonomous runs perform. It includes:

• Total number of agent runs.
• Breakdown of runs by triggers.
• Overall success rate for agent runs.
• Success rate for autonomous agent runs over time.
• Tools analysis so you can see tools usage and success rate.
• Details on how long autonomous agent runs take, with a breakdown of successful and failed runs.
• Billing consumption analysis for autonomous agents.

In the Activity tab on your agent, you can review your agent's sessions. You can see transcripts of your user's interactions with your conversational agents when users choose to share them. This information helps you understand how the agent behaves in real-world scenarios so you can effectively debug and improve the agent. Agents that use generative orchestration have this functionality by default.

In the Activity list, you can see more information, including step count and last action, to help you identify the session you want.

When you view a specific agent run, you see what triggered the run, which actions the agent took, and the inputs and outputs for each action. You also see details about which knowledge sources the agent used and a detailed thought process to help with troubleshooting and understanding agent behavior.

In Copilot Studio, knowledge sources work with generative answers to enhance your agents' capabilities. When you integrate knowledge sources, your agents can use enterprise data from Power Platform, Dynamics 365, websites, and external systems to give your customers relevant information and insights.

This release introduces support for adding SharePoint as a knowledge source for makers building autonomous agents. Only authenticated users can access SharePoint, so information sharing is secure and reliable. With this release, makers can access SharePoint knowledge using their credentials. Soon, you'll be able to access SharePoint knowledge with end-user credentials.

To add SharePoint as a knowledge source, use the following steps:

1. Open the agent.
2. Go to the Overview or Knowledge pages, and select Add Knowledge.
3. Select SharePoint from the list of options.
4. Enter the SharePoint URL.
5. Confirm your selection.
6. Select Add to finish adding SharePoint as a knowledge source.

With generative orchestration, your agent chooses the best actions, knowledge, and topics to answer user queries or respond to event triggers. In contrast, classic orchestration uses only one topic to respond to a user, based on manually authored trigger phrases.

When you activate generative orchestration, it mainly relies on the descriptions of actions and topics to determine which elements to use in response to the last user input or event. When enabled, generative orchestration orchestrates the conversations users have with your agent. This orchestration helps provide users with a more natural and fluid conversation.

The following table compares agent behavior between classic orchestration and generative orchestration.

To turn on generative orchestration for your agent, follow these steps:

1. Go to the Overview page for your agent.
2. In the Details section, under Orchestration, turn on Use generative AI to determine how best to respond to users and events.

After an agent is created in Copilot Studio, makers will see a new section Generative response model with a list of available models.

The Default model is automatically selected if the user doesn't make any change.

With this feature, the maker can now chose to switch to the Preview GPT 4.5 model. After they switch, the agent's answer generation workflow will use GPT 4.5 model instead of the existing default (GPT 4o). The maker can experiment and explore the quality and character of the responses in the test chat or with agent users.

They can switch back and forth between the Default and Preview models at any time.

Microsoft provides a catalog of managed agents that you can browse and install from within Copilot Studio. These agents are complete solutions. They're prebuilt with everything you need to immediately connect to related services and provide autonomous capabilities to your users.

Customize these agents with add-ons to help them specialize in a specific domain. Add-ons are packages of connections and capabilities that each managed agent solution uniquely provides.

On the Create page, under Managed agents, you can see which managed agents are available for your current environment. Select an agent to learn more about its capabilities and requirements.

You can extend Microsoft 365 Copilot with a custom agent by turning on the Microsoft 365 Copilot channel. Submit your custom agent to Teams or Microsoft 365. You can submit it to an admin for publishing in the line of business catalog, or you can deploy it just for yourself or to a small team by using a security group.

Microsoft 365 Copilot supports most common custom agent chat features. These features include messages, quick replies, adaptive cards with Action.Submit, multiturm interactions, and multimessage responses to a user interaction. Because it's a custom agent, it also benefits from Copilot Studio's rich usage analytics and governance provided by Power Platform Admin Center. Use your existing custom agent from Copilot Studio without any changes. Just add it to the Microsoft 365 channel.

Graph connectors help organizations connect their enterprise data. They expand the range of content sources available in Microsoft Copilot Studio, Microsoft 365 Copilot, and the broader Microsoft ecosystem.

With this feature, makers can access more data sources that fit their organization and the agents they want to create. They don't need to build custom or complex solutions. The feature also boosts the quality of agents with graph connectors in Copilot Studio by supporting semantic search for the best results.

These new preconfigured graph connectors include Stack Overflow, Salesforce Knowledge, and GitHub.

Makers can use these connectors as knowledge sources when they create new agents. Administrators can set up new connections in the Admin Center, manage data crawling, and set permissions.

Currently, user input triggers agents in Microsoft Copilot Studio. When an end user asks a question or sends a message, the copilot reviews a series of maker-defined topics and actions. It uses this information to respond appropriately to the end user.

When you add automated triggers to Copilot Studio, you enable an agent to run without any end-user input. The agent responds to events like an email arriving, a document being created, or a table row being updated. It follows the instructions you define for it.

We add an activity tab to the agents in Microsoft Copilot Studio. Use the activity tab to track each historical interaction with your agent. The activity tab records each time the agent triggers a topic or action as a conversation. You can drill into the interaction and review each action that the agent calls and why it calls the action. For this version of activity, the agent records only interactions with you.

Use analytics in Copilot Studio to:

• Monitor the performance of your copilots.
• Understand the top themes in questions that users ask.
• Get detailed insights on user satisfaction, including a breakdown of user behaviors that drive satisfaction status.
• Monitor result statuses, along with the top themes of queries without search results.
• Get visibility into how Microsoft moderates and blocks toxic queries to ensure the compliance of your copilots.

They say a picture is worth 1,000 words. With image analysis, Copilot users can convey complex scenarios, objects, or ideas without typing lengthy descriptions. Copilot gets context from visual cues, which broadens its conversational range and yields more informed responses.

For example, when a user uploads an image of a car and asks, "What model is this car?", Copilot analyzes the image and responds with the car model. Later in the conversation, if a user asks, "What is the range of this car?", Copilot uses the previous image as context to answer the question.

Whether it's explaining a recipe from a food photo, getting advice on troubleshooting a broken printer, or identifying a rare bird species, Copilot provides detailed insights through image analysis.

Makers can use prebuilt indexes (indexes that pro developers built) with minimal friction and low cost. Use these indexes as grounding data for your copilots that you create with Microsoft Copilot Studio. With these copilots, users can ask questions in natural language about their enterprise systems that they use within their Power Platform tenants. Microsoft builds on the natural language-to-query approach that it released with Microsoft Dataverse knowledge and extends the same approach to copilots that use Azure AI Search index as knowledge.

Solution management in Copilot Studio brings Power Platform application lifecycle management (ALM) directly into the Copilot Studio app.

When you create an agent in Copilot Studio, you add it to a Power Platform solution. With Power Platform ALM, you can create custom solutions to manage your agents across multiple environments. You can also use pipeline deployments and other ALM scenarios in the Power Platform admin center.

By adding solution management directly into the Copilot Studio app, you can view, export, and import solutions. You can set your preferred or default environment, configure pipelines, and customize and manage your solutions without leaving your agent-building environment.

The solution manager in Copilot Studio enables you to:

• Create agents with solutions.
• Set the preferred solution in Copilot Studio.
• Create agents and components with a preferred solution.
• Edit and extend agents and components with a preferred solution.
• Deploy agents and components by using pipelines in Power Platform with a single click.
• Share agents and components with a user or team.

For more information about the solution manager, including how to use it and how it connects to your broader organizational ALM scenarios, see Create and manage solutions in Copilot Studio.

Copilot extensions in Microsoft Copilot for Microsoft 365 are created by employees in your organization to retrieve information or perform actions through Copilot.

With this feature, you can create and share these extensions in a simple canvas that all Microsoft Copilot users can access. Employees in your organization can quickly and easily create work-specific extensions in a familiar experience. They don't need to learn a new app or service. The extensions are instantly available in Microsoft Teams or at https://microsoft365.com/chat.

Microsoft 365 Copilot is a seat-licensed product that includes the Microsoft Copilot Studio license, so anyone can be a maker. These new makers can use Copilot Extensions in Microsoft 365 Copilot. They can also browse the Microsoft 365 Copilot Extension store to install an extension. Makers see the create a copilot extension call to action and go to the studio.

For example, an employee might want to create an extension for Copilot to handle internal queries and requests for creating, managing, and approving tasks within an expense reporting system.

When the maker creates their copilot extension, they can expand its knowledge with data sources from SharePoint and Microsoft Graph. This expansion enables richer, more informed answers that are contextually scoped. Makers can also create actions. These actions connect the extension to enterprise and internet systems to perform tasks, like getting the weather or resolving a ticket in a customer management portal.

When you define the knowledge and actions for a copilot extension, publish the extension to Microsoft Copilot. End users can then get information from the knowledge sources and interact with the capabilities the actions offer, all within Microsoft Copilot.

Admins can trust these copilot extensions. All extensions use the same Microsoft large language model. Power Platform admins govern which Power Platform tools are available in a copilot action. Microsoft 365 admins control who can use the copilot extension.

With Copilot Studio plugin extensibility for Microsoft 365, you use the user's authentication token to invoke the plugin. This approach makes capabilities such as generative answers more useful because the plugin uses corporate resources on the user's behalf, including Microsoft SharePoint.

By using parameters, plugin creators define the parameters Microsoft Copilot collects before it calls the plugin. These parameters are available within the plugin to call external functions, apply business logic, or use in any way the creator chooses.

Copilot Studio simplifies the process of defining parameters and adding descriptions for Microsoft Copilot. With this information, Copilot knows how to prompt the user to provide the necessary details.

This new user experience organizes the features and capabilities available to makers in Microsoft Copilot Studio.

A refreshed navigation menu and tab-based structure makes it easier to understand your place in the navigational hierarchy and what you can access:

• The Home screen has a Describe with Copilot section that simplifies creating a custom copilot. It also offers custom copilot templates to help you get started.
• The Library page shows your most recent copilots and extensions, so you can pick up where you left off. It also displays all your environment-level components.
• The Configure page gives you an at-a-glance view of what a selected copilot can do.
• The top menu bar provides access to knowledge, topics, actions, analytics, and channels, so you can see detailed views for each component.
• Test your copilot anytime with the refreshed test pane.

The solution applies safer defaults and gives opinionated in-product communication when makers change those defaults. For example, the solution changes two significant security defaults. All bots require authenticated access, and makers don't have access to transcripts by default.

The solution also introduces these improvements:

• Granular governance levers that give admins more control over the copilots in their tenants without affecting maker adoption.

  • Customers commonly request this feature. The solution invests in granular controls for many Copilot Studio features, including generative AI, knowledge sources, app insights, and authentication.

• Audit log and inventory capabilities for admins.

  • Admins can see comprehensive audit logs, including viewing tenant-wide usage, inventory (including API support), and tenant hygiene (DLP violations and inactive copilots) reports to monitor business impact.

• Show sensitive labels to the copilot creators and users when the response includes generative answers from SharePoint files.

This feature enables makers to enrich their custom copilots with knowledge sources.

To get enterprise value from AI, customers need to integrate enterprise knowledge with foundational models. This integration includes enterprise data, data in application silos, and the Microsoft 365 Graph. Custom copilots that have knowledge make it easy to ground LLM responses in enterprise data. They give users relevant business insights in a secure and compliant way.

Supported types are public websites, SharePoint and OneDrive, file uploads, Dataverse (including Dynamics 365 and Power Apps data), and other external data sources indexed in Microsoft Graph.

Create a custom copilot with an initial description.

The conversational creation experience asks for additional information to refine the scope and behavior of the copilot.

Add knowledge sources, such as website URLs, to the copilot so it can respond to information on those sites.

Switch to the Configure tab to refine the copilot, see the conversation starter prompts, or add extra details.

In the right pane, view the summary of instructions that the AI model uses to create the copilot.

When the conversation completes, the creation process autogenerates a name that matches the copilot's function. You can preview or test the copilot and go to Copilot Studio to add more features to the copilot.

You can add these knowledge sources to your agents without any coding. Within the Microsoft 365 environment, the supported knowledge sources are SharePoint and Microsoft Graph connectors.

Start from the configure page for your agents in Microsoft Copilot Studio. When you add knowledge sources, the next screen shows all the supported knowledge sources. Choose the knowledge source you want to add. For example, you can choose a SharePoint document library or an existing Microsoft Graph connection, such as the ServiceNow knowledge base or Confluence.

In conjunction with multilingual bot support, topics and bots that use generative answers adapt to the current conversation's language to answer questions.

Customers can create their own logic with Power Fx to choose which sources to use for a request. The logic can use global variables. When generative answers connects to a data source in the user's language, the copilot takes the request from the copilot user in their language and returns the response in the same language.

Generative answers supports all languages that Copilot Studio supports, except Chinese (Traditional). For a list of supported languages and their levels of support, see Supported languages.