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

Administrators can manage access to use 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 to be used by the agent when generating results or answering queries. Connections to knowledge sources are configured within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

Content is processed and indexed in Dataverse to provide accurate, detailed results to natural language queries from agent users. Users will only receive answers based on data they have access to, as user access tokens are validated against Confluence content before the response is returned, and they can access the original articles through citation references.

The knowledge base data is kept up-to-date through scheduled sync, 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 enables agents to provide high-quality answers based on operational data stored in files on OneDrive.

Administrators can manage access to use 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 to be used by the agent when generating results or answering queries. Connections to knowledge sources are configured within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

Content is processed and indexed in Dataverse to provide accurate, detailed results to natural language queries from agent users. Users will only receive answers based on data they have access to, as user access tokens are validated against OneDrive content before the response is returned, and they can access the original articles through citation references.

The knowledge base data is kept up-to-date through scheduled sync, within a 4-6 hour window.

Administrators will be able to manage the availability to use this data through the management of the OneDrive connector within the Administration Portal.

This feature allows makers to connect to Oracle using a low-code or no-code interface, eliminating the need to manually copy or ingest tabular data.

Key Capabilities

• Real-time reasoning over external data: Data remains external within Oracle, while Copilot performs real-time reasoning and responds to natural language queries.
• Natural language querying: Makers can use natural language instructions—no SQL required—to query, join, and filter Oracle tables.
• Conversational responses: Data is reasoned over by the agent and presented conversationally in the Copilot chat interface.
• Grounded answers with context: Makers can select specific Oracle tables as knowledge sources. Agents then answer grounded questions with accuracy and context, without requiring manual integration or scripting.
• Seamless integration in Copilot Studio: Makers can add Oracle as a knowledge source when creating a new agent or updating an existing one.
• Automatic availability: This feature is automatically enabled and available 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 enables agents to provide high-quality answers based on operational data within your Salesforce configuration.

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

Makers can connect to Salesforce as a knowledge source and select Knowledge Base unstructured sources to be used by the agent when generating results or answering queries. Connections to knowledge sources are configured within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

Knowledge Base content is processed and indexed in Dataverse to provide accurate, detailed results to natural language queries from agent users. Users will only receive answers based on data they have access to, as user access tokens are validated against Salesforce content before the response is returned, and they can access the original articles through citation references.

The Knowledge Base data is kept up-to-date through scheduled sync, within a 4-6 hour window.

This feature allows makers to connect to SAP using a "low-" or "no-code" interface, eliminating the need to copy or ingest tabular data manually.

Data remains external within SAP, yet Copilot can perform real-time reasoning and respond to natural language queries.

Makers can select specific Databricks tables to use as knowledge sources, and agents can answer grounded questions over the data with accuracy and context without manual integration or scripting to surface the data.

Makers can add the knowledge source within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

They don't need to create SQL queries, as Copilot Studio will parse natural language instructions to query, join, and filter the appropriate tables.

Data is returned and reasoned over by the agent, before being parsed conversationally within the chat to the agent user.

This feature is enabled automatically for the user and can be found in the Agent Add Knowledge section. It can be used by selecting SAP as a knowledge source.

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

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

Administrators can manage access to use this data by configuring the Service Now connector in the Administration Portal.

Makers can connect to Service Now as a knowledge source and select knowledge base unstructured sources to be used by the agent when generating results or answering queries. Connections to knowledge sources are configured within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

Knowledge base content is processed and indexed in Dataverse to provide accurate, detailed results to natural language queries from agent users. Users will only receive answers based on data they have access to, as user access tokens are validated against Service Now content before the response is returned, and they can access the original articles through citation references.

The knowledge base data is kept up-to-date through scheduled sync, within a 4-6 hour window.

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

Makers can select specific Snowflake tables to use as knowledge sources, and agents can answer grounded questions over the data with accuracy and context without manual integration or scripting to surface the data. Makers can add the knowledge source within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent. They don't need to create SQL queries, as Copilot Studio will parse natural language instructions to query, join, and filter the appropriate tables. Data is returned and reasoned over by the agent, before being parsed conversationally within the chat to the agent user.

When a maker is authoring their copilot extension, they can add suggested prompts that the copilot will support.

In the case of the Microsoft 365 Chat experiences, these will be highlighted in the center of the page, with title and description, rotated and prominently displayed at the start of each session.

For other channels this functionality will be presented with appropriate experiences for that channel.

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

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

Administrators can manage access to use 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 to be used by the agent when generating results or answering queries. Connections to knowledge sources are configured within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

Knowledge base content is processed and indexed in Dataverse to provide accurate, detailed results to natural language queries from agent users. Users will only receive answers based on data they have access to, as user access tokens are validated against ZenDesk content before the response is returned, and they can access the original articles through citation references.

The knowledge base data is kept up-to-date through scheduled sync, within a 4-6 hour window.

Understanding the efficiency of your agents and their associated costs is crucial for creating cost-effective agents. Now, agent makers can understand how efficient their agent is and monitor its cost over time.

With this feature, admins can:

• Receive full billing consumption analysis the way that works for them and your business—by billed sessions or by message consumption.
• View consumption analysis over time to identify changes in consumption behavior.
• Understand the specific events that contribute to message consumption, so they can understand the efficiency of your business's agents and ways to optimize them.

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

The feature uses AI to check relevance, completeness, and grounding of each response that is part of the sample set and assign a quality score for each. Based on the quality score, the response is classified as having a good or poor quality of response.

Makers will see a visual indication for the breakdown of answered generative AI questions between good and poor quality. Makers can view the reasons for poor quality responses and understand if the response is not fully relevant, the response is incomplete or the response is not fully grounded.

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’ll see key stats like hours saved, usage analysis, and resolution rate rolled up at the tenant level and drillable down to each agent. It’s all powered by regular usage data sent from Copilot Studio to Microsoft 365.

This feature is enabled by default if you meet the following criteria:

• You've got at least 50 Microsoft 365 Copilot licenses (standalone or bundled with Viva Insights).
• You’ve published at least one Copilot Studio agent in a production environment.

The feature is controlled in the Power Platform admin center and will be automatically enabled for all public customers in a commercial cloud that meets the criteria.

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

Environment admins can also disable it from the Power Platform admin center.

If your Copilot Studio data resides in a different geographical location or region than your Microsoft 365 tenant, an environment admin needs to manually enable data sharing across geographical boundaries.

Computer use is a feature in Copilot Studio that enables your agent to interact with any system that has a graphical user interface (GUI), such as websites or desktop applications. It can perform 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 using a virtual mouse and keyboard.

This allows agents to complete tasks even when no API is available, making it useful for things like automated data entry, invoice processing, or data extraction.

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

Makers will be able to use pre-built indexes (indexes that were built by pro-developers) with minimal friction and low cost, as grounding data for their copilots built using Microsoft Copilot Studio. Such copilots allow users to ask questions in natural language about their enterprise systems being used within their Power Platform tenants. Microsoft will build on the natural language to query approach released with Microsoft Dataverse knowledge and extend the same to copilots using Azure AI Search index as knowledge.

Makers can receive user feedback to agent responses and use it to optimize an agent's performance. The user feedback feature is enabled by default for all new and existing Copilot Studio custom agents, and makers can choose to disable the feature from the agent's settings.

The ability to provide feedback is available in the following channels:

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

Agent users can also provide comments before sending their feedback. The comment is available in the conversation transcript table in your Dataverse store.

Analytics aggregates user feedback and shows the total number of reactions, with a split between thumbs-up and thumbs-down ratings.

You can configure custom agents to perform actions or call topics in response to events using triggers. Unlike topic triggers, which require user input, event triggers allow your agent to act autonomously when a defined event occurs.

This new feature lets you create triggers for autonomous agents that run using end users' credentials. This means you can create an agent that operates entirely in the background, triggered by external events, and share it with end users so they can configure the agent to run on their behalf.

As a maker, you will:

1. Create an agent and add a trigger.
2. Configure the trigger to run with end-user credentials.
3. Publish and share the agent.
4. Notify the end user that an agent has been shared with them and prompt them to configure and enable the agent.
5. Once configured and enabled, the agent will begin to operate on behalf of the end user.

This feature provides a seamless way to automate tasks and improve efficiency by leveraging event triggers and end-user credentials. It ensures that agents can act independently and be easily shared and configured by multiple users.

You can now add structured and unstructured data from Microsoft and non-Microsoft systems as grounding data for your copilots built using Microsoft Copilot Studio. Such copilots allow users to ask questions in natural language about the enterprise systems being used within their Power Platform tenants. Microsoft will build on the natural language query approach released with Microsoft Dataverse knowledge and extend the same to select non-Microsoft services.

• Microsoft Services:
  • One Drive for Business
  • SharePoint Lists
• Non-Microsoft Services: *Confluence - Structured and Unstructured *Databricks - Structured *Salesforce - Structured and Unstructured *ServiceNow - Structured and Unstructured *Snowflake - Structured *Zendesk - Structured and Unstructured

Extending Microsoft 365 Copilot with a custom agent is as easy as turning on the Microsoft 365 Copilot channel and then submitting it to Teams or Microsoft 365, either by submitting to an admin to publish to the line of business catalog or deploying just for yourself or to a small team using a security group.

Microsoft 365 Copilot supports most common custom agent chat features; messages, quick replies, adaptive cards (with Action.Submit), multi-turn interactions, and multi-message responses to a user interaction. In addition, because it's a custom agent, it 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 change, just add it to the Microsoft 365 channel.

Analytics for autonomous agents provides a comprehensive view on the the performance of autonomous runs, including:

• Overall number of agent runs.
• Breakdown by triggers.
• Overall success rate of agents runs.
• Success rate of autonomous agent runs overtime.
• Tools analysis—understand tools usage and success rate
• Details on duration of autonomous agent runs with a breakdown between successful and failed runs.
• Billing consumption analysis for autonomous agents.

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

• View insights for unanswered Generative AI questions.
• Each insight focuses on specific theme of questions.
• Receive guidance on how to address the problem and increase performance.
• View the list of related conversations with unanswered questions so you can easily drill down to investigate the problem.

Insights are generated daily, by looking at all Gen AI unanswered questions from the last seven days. The feature does not support SharePoint and Graph Connector Knowledge sources.

The custom search node allows makers to run a search action on the selected knowledge source and save the raw data output to a variable. This allows flexibility in how the search results are post-processed.

Here is how the new node should be used:

1. Open Copilot Studio and sign in with your account.

2. Navigate to the topic in which you would like to perform the call to the model.

3. If you wish to include in the model call grounding data, use the new custom search action.

   This new node allows you to use RAG to extract the raw data, as opposed to a Generative Answer node which will also summarize the raw data.

4. To include the user message context, put the system variable Activity.Text as input. The search flow will start with getting all required context and create a search query.

   Select the relevant or all knowledge sources in the Data sources (under action properties, Knowledge sources).

   You can also set moderation level and latency message.

5. Save the search output in a new variable.

This node will be available for all Copilot Studio agent makers.

In the Activity tab on your agent, you will be able to review your agent's sessions. You can see transcripts of your user's interactions with your conversational agents where you have chosen to share. This allows the maker to understand how the agent is behaving in real world scenarios to effectively debug and improve the agent. This functionality is available by default for agents that use generative orchestration.

In the Activity list, you can now see more information, including step count and last action, to identify the session you are looking for.

When you drill into a specific agent's run, you can see what triggered the run, which actions were taken, and the inputs and outputs of each action. Additionally, you can see details on which knowledge sources were used and a more detailed thought process, to help with troubleshooting and understanding agent behavior.

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

Customize these agents using add-ons that allow them to further specialize in a specific domain. Add-ons are packages of connections and capabilities unique to each managed agent solution.

See what managed agents are available for your current environment on the Create page, under Managed agents. Select one to learn more about its capabilities and requirements.

We are introducing two new autonomous agent templates, showing how to build an agent with a trigger that handles tasks autonomously.

The new agent templates, the Lead Manager and Customer Brief, are designed to accelerate customer adoption of autonomous agents.

The Lead Manager template focuses on top-of-funnel sales development by managing incoming leads, offering functionalities like duplicate handling and automated email responses. It integrates with Salesforce and Outlook while allowing customization such as setting employee count thresholds for meeting scheduling.

The Customer Brief template streamlines meeting preparation by generating executive summaries from Salesforce data and automating email workflows.

Both agents are installed via Copilot Studio "Managed Agents". The templates can be edited before publishing and customized to your business processes.

Bring your own model (BYOM) to response generation allows you to plug in your own custom model (deployed in AI Foundry), enabling more tailored, domain-specific answers.

First, navigate to the agent settings and use the response models drop down list to add a new model, and provide the AI Foundry deployment connection details of your model.

Next, you define the model prompt, containing instructions and global variables. With full control over the prompt, you can ensure the model receives rich, contextual input for generating accurate responses.

You can also choose whether you want to include citations.

With the custom model connected and the prompt configured, you can have full control over the agent response generation.

You can now include a SharePoint Lists within a SharePoint site as a knowledge source for your agents.

SharePoint Lists can provide additional grounding for agents for several scenarios, and generative answers can include relevant results from SharePoint Lists.

Makers can use SharePoint Lists by adding SharePoint as a knowledge source and then selecting relevant lists within recent lists or my lists.

Selecting a specific list will ensure all rows and columns are available for the agent based on the ACL restrictions specified for the SharePoint List.

This feature integrates the benefits of the Power Platform catalog to Microsoft Copilot Studio, enabling administrators to discover, install, and manage agents across Dataverse environments.

Before installing an agent, administrators can review and make corrections to the configurations, connections, and environment variables associated with the agent.

Installation status, details, and results can be tracked without deviating from the underlying work.

Admins and Makers can connect to third-party data sources using prebuilt extensions. Extensions combine connectors, topics, flows, and configuration files to quickly allow agents to access third-party data for domain-specific use cases such as prospect to cash, configure price quote, benefits and compensation, and IT helpdesk.

What are file groups?

With file groups, makers can group up to 500 files into a single knowledge source, where agents can narrow their scope of search when queried about relevant and related data.

Allowing to query over a collection of files enables the extension of query scenarios to include proper determination of content.

File group requirements

• Two files minimum.
• Only for local file upload.

File group example scenario: Imagine Contoso international airlines want to build a consumer bot for their customers where they can ask questions about their upcoming travel. One of the most common questions they get is about baggage policies.

Situation: We know baggage policies differ based on the type of travel (domestic vs international), customer tier (economy, business, first class) and perhaps other attributes.

Solution: Contoso has a collection of PDFs where each policy is written in detail, but they have a document for each customer tier. In this case, each criteria such as “type of flight” and “seat type” can be set up as a variable that helps the agent filter through the knowledge to find the most relevant chunk that satisfies the criteria and retrieve the most accurate answer.

What makes file groups?

File groups are made of three components:

• Name and Description: Helps the agent know what the knowledge is about.
• Instructions: Instructs the agent how to find the knowledge for each given query.
• Knowledge: The desired files with the information to ground the agent.

How are file groups different from individual files?

While individual files are each considered their own knowledge sources, a file Group uses multiple files as a single knowledge source, and uses instructions to help the agent understand the right chunk to find and provide the right information back to the user.

When using individual files as a knowledge source, all knowledge sources will be searched on simultaneously, and the answer provided to the users depends on one or more chunks the agent was able to find in the one or more files, regardless of important variables that may help narrow down the scope of the answer.

When using file groups, searching for the right information happens in two stages. First, the agent looks at the agent level instructions to understand and get to the right file. Next the agent looks for the file group instructions to understand how to find the right chunks in the file to answer the user’s query. This way, the agent provides more accurate answers from the right file.

An example of a file group is for a scenario where an organization has information that is nuanced, depending on specific dimensions, such as geographic location, or designation.

How to create a file group

This is a Public Preview enabled feature. Once enabled, the feature can be seen when uploading more than two files at the same time, which is indicated via a toggle that says, "Group these files (preview)".

Once the toggle is turned on, the name and description are changed to indicate a group of files, and a new "Instructions" field is added. Here the user can add specific instructions with variables to help the Agent find the right knowledge for a given situation.

When a file collection is added as a knowledge source, it is represented using a folder icon. This can be seen in the Overview page (Knowledge section) or the Knowledge page itself.

When to create a file group

File groups can be created at the time of adding new knowledge, via the “Add knowledge” button.

Is it possible to create file groups using pre-existing files?

Currently it is not possible to create a group using pre-existing files: the same files can be re-uploaded as a new group.

Is it possible to create file groups and upload individual files at the same time?

While it is possible to have both individual files and file collections as knowledge sources for the same agent, it is currently not possible to upload individual files and file collections at the same time: either a set of files is uploaded as a group or as individual files. Makers can do multiple individual files and file group uploads to an agent as needed.

Is it possible to create multiple file groups?

It is possible to create multiple file groups, one at a time, up to 500 knowledge sources.

Is it possible to select a folder to upload a file group?

While direct folder upload is not supported, it is possible to select all acceptable files inside of a folder to create a file group.

How to remove a file group

When adding new knowledge to an agent, the user has the option to either upload individual files or a group (for more than two files). Before pressing the “Add” button to add the files to the agent, the user can un-toggle the grouping at any time, deciding to upload them as individual knowledge sources.

For existing file groups, by hovering on the knowledge source, a hover menu becomes visible. When selected, there is an option to delete the entire file group.

How to know which files are in each file group

To see the details of a file group such as the individual files that make it, the user can select the name of the file or the hover menu to "Edit" the group. This takes the user to the Knowledge details page.

In the Knowledge Details page, individual files can be deleted within a file group, up to two files that are a requirement to make the group.

Is it possible to update or replace a file within a file group?

To update a file inside a group, the old file must be manually removed (deleted) and the new file manually added to the file group:

1. In the Overview or Knowledge pages, select the desired file group.
2. In the Knowledge details page of the file group, select Edit.
3. Remove the old file by selecting the hover menu, followed by Delete.
4. Drag and drop or select the option to upload the desired updated file.

Can more files be added to a file group?

At any time, new files can be added to a file group:

1. In the Overview or Knowledge pages, select the desired file group. 1 In the Knowledge details page of the file group, select the “Edit” button.
2. Drag and drop or select the option to upload the desired file(s).

Can files be removed from a file group?

At any time, files in an existing file group can be removed until there’s only two files left. Two files is the minimum requirement for a file group to exist.

To remove a file inside a file group:

1. In the Overview or Knowledge pages, select the desired file group.
2. In the Knowledge details page of the file group, select the “Edit” button.
3. Remove the desired file by selecting the hover menu, followed by “Delete”.

Repeat the process for all files to be removed.

What are the instructions for the file group?

Instructions is a rich text field, where users can include variables such as a user’s location, or Date of Birth, to create requirements for the agent to perform a knowledge search.

What happens if instructions are not provided in a file group?

If instructions are not provided in a file group, the agent will search through all the knowledge contained in that file group, instead of taking certain variables into account and going straight to the most relevant information.

By introducing the WhatsApp channel in Copilot Studio, your makers can deploy their agents into WhatsApp in a few clicks.

The WhatsApp channel uses Azure Communication Services (ACS) as the bridge between the agents you create in Copilot Studio and your WhatsApp endpoint, the chat or group that the users of your agent will use to interact with the agent.

When you publish an agent to WhatsApp (after obtaining and connecting a WhatsApp for Business account to ACS), Copilot Studio fetches your WhatsApp configuration from ACS and lets you choose the number or endpoint where your agent should appear.

You can now deploy custom agents created in Copilot Studio directly to SharePoint sites. This lets your team interact with the agent in the context of shared documents, plans, or collaboration pages, without switching interfaces.

Once the SharePoint channel is configured, users will see a Copilot icon on the site. Clicking the icon opens the agent experience, allowing users to engage with the agent directly within the SharePoint environment.

The SharePoint channel respects the Authenticate with Microsoft setting by default, ensuring secure, identity-aware responses. It also follows existing data loss prevention (DLP) policies, and admins can manage its availability through the Power Platform admin center using DLP rules.

Makers must have write access to the SharePoint site in order to configure and deploy agents.

Often you will have a need to use an action in multiple Copilot deployments, however today you need to recreate the same action in each deployment to have the same experience.

This feature allows you to select a previously published connector action and publish it to additional Copilot deployments in their organization. For example, you can use a previously published connector from Copilot for Sales and publish it again for Copilot for Customer Service or other Copilot for Sales distributions.

This means you will be able to reuse important functionality without requiring them to reconfigure the same action multiple times.

The interface provides a Publish button that provides a list of all available endpoints within the environment that you can select from. You can then submit the actions to additional selected endpoints for administrator approval with a few simple clicks.

This feature is turned on by default and does not require any configuration to enable.

Previously, makers had to create and configure various agent tools in a separate location (the Library page in Copilot Studio) from the agents where they wanted to use the tools.

With this update, the new Tools page streamlines how makers can view, create, and use tools.

Tools listed on the page can be added directly to any agent within the environment where the tool was created, and makers can craft reusable and valuable tools across multiple agents in the same environment.

Tools available on the page include:

• prompts
• agent flows
• custom connectors
• REST APIs
• computer using actions (CUA)
• model context protocol

Makers can create and view tools on the Tools homepage and also select Add to agent to directly add selected tools to their agents.

This feature allows makers and admins to unpublish a previously published Copilot connector.

Copilot connectors allow makers to connect actions and activities from other services to Microsoft 365 Copilot. However, connectors may need to be disabled in certain scenarios. For example, the content or data being connected to is out of date, your business is no longer using a specific service, or you need to take it offline to make updates to the connector's configuration.

A maker can select a connector and unpublish it to return it to its draft state and prevent it from being shown as ready and usable in Microsoft 365 Copilot. When a published connector is returned to a draft state, its associated solutions will also be updated. Draft solutions can either be edited and published or deleted.

Admins in the environment can unpublish any connectors present; makers can only unpublish the connectors they own.

This applies to all Microsoft 365 Copilot agents, such as Microsoft 365 Copilot for Sales. It doesn't apply to custom agents created in Microsoft Copilot Studio.

With enhanced visibility into how actions are being utilized within an agent, you can leverage analytics to drive actionable insights. This visibility allows you to understand usage patterns, identify areas for improvement, and optimize performance. By analyzing this data, you can make informed decisions, proactively address issues, and continuously refine your strategies to enhance user experience and operational efficiency.

The metrics provided will include answer rate, action usage rate, and error rate, giving you a comprehensive view of how users are interacting with actions.

Additionally, there will be a detailed breakdown of errors, allowing you to pinpoint specific issues and address them effectively. You can view how many times the extension was used by users in their organization over a period of time, providing valuable visibility as they iterate on the extension built. This experience will be available for connector, prompt, and REST API extensions.

By analyzing how often each action is used, which sources are utilized, and the error rates, you can improve the agent's effectiveness and reliability, leading to better user satisfaction and business outcomes.

This feature brings analytics specific to knowledge curation in Copilot Studio.

When working with knowledge sources in Copilot Studio, you can see valuable insights into the performance of your individual knowledge sources. You can view detailed metrics for the following, directly from the knowledge page in Copilot Studio:

Knowledge source usage: This metric shows the frequency of use for each knowledge source, helping you understand which sources are most popular or effective. Answer rate: This metric indicates the percentage of queries that were successfully answered using the knowledge sources, providing insight into their accuracy and reliability. Error rate: This metric tracks the number of errors encountered, allowing you to identify and address any issues with the knowledge sources.

By analyzing these metrics, you can make informed decisions to improve the performance and effectiveness of their knowledge sources.

The main features of advanced approvals:

• Multi-stage approvals: Users can configure complex approval processes with multiple stages. Each stage has its own set of stakeholders who make approval decisions, and each stage can be separately configured to meet the specific needs of those stakeholders. This feature reflects the real-world approval workflows of organizations.

• Conditions/Conditional approvals: Users can add conditions between stages to ensure flexible decision-making. Conditions allow approvals to be automatically approved, rejected, moved to another stage, or skipped based on whether specific conditions are met. This feature enhances the adaptability and efficiency of the approval process.

Power Platform connectors play a critical role in allowing makers to build low-code apps, workflow automation, and AI agents. They provide seamless connectivity to various data sources.

With the release of this feature, you can now build enhanced Power Platform connectors on top of structured data. These connectors allow makers to leverage PowerFx while building apps and use knowledge sources in agents built in Copilot Studio.

The Power Platform Connector SDK for structured data allows developers to build out a Web API that provides structured data. The Web API can then be registered as a connector in any Power platform environment as a connector with structured data. Doing so will automatically light up the enhanced experience for makers.

With this feature, you can use the simplified interface in Copilot Studio to quickly and easily build custom agents, either by using simple, natural language or manually.

You can author your custom agent in two ways:

• Via the Describe tab, using natural, conversational language.
• Via the Configure tab where you can build it manually.

Both tabs work seamlessly to provide a rich authoring experience.

The Describe tab allows you to create an agent using everyday natural language. As you provide information conversationally, the agent's name, description, instructions, and starter prompts update continuously to refine the agent's behavior.

This experience provides a rich yet simple natural language way to create a customized agent. You build agents in plain language by answering the questions asked in the Describe chat. This progressively updates the agent in each turn of the conversation, and changes are saved automatically.

You can make updates to the agent in any turn of the conversation. Changes to the agent's name, description, and instructions update automatically.

You can also add common knowledge sources, such as connecting to a public website or uploading a file by dragging and dropping into the field in the Describe tab.

In addition to using the Describe tab, you can also directly configure the agent's behavior using the Configure tab.

The Configure tab provides you with the ability to view and edit information about the agent, giving you more control and precision.

The Describe and Configure tabs stay in sync between your changes, so the fields in the Configure tab update to reflect the latest changes from the Describe tab. You can switch between the tabs to use the experience that is most comfortable to you to author the agent.

Agent Fields

The following table describes the fields that make up the agent:

Test Your Agent

The agent in the right pane is an ephemeral instance of the agent that shows up as a side-by-side screen within the authoring experience. You can use it to test the agent while you're creating or making updates to it. You can test your agent in the test pane to experience it in a similar way as end users would. The test pane is enabled as soon as the agent has a name, description, and instructions. It behaves according to those instructions like a fully-fledged agent, including the ability to respond to complex queries within its realm of specified knowledge. The agent updates during each turn of the conversation as new information is added in the Describe or Configure tab.

Building agents is an iterative process that goes through an author-edit-test cycle. Once you are satisfied with the initial setup of your agent, select Continue to complete the initial creation phase, beyond which you can deeper into the world of tools, triggers, and topics.

Connecting agents with the end user's work data with knowledge sources enhances the accuracy of the agent's responses through domain knowledge. The main challenge in knowledge source selection is identifying relevant data and its location.

With this update, we're rebuilding the experience in Copilot Studio when you select the data sources they want to use in their agent.

Previously, selecting data sources required either existing knowledge of the precise name and location of the data source, or dedicated time to locate and select the source. Now, when you configure your agent's data sources, you'll have a search-first experience that responds to queries and returns applicable data sources.

The new selection experience is backwards compatible with existing knowledge sources and allows for easily integrating new sources.

You will see the new experience when selecting data sources while configuring an agent.

Customers often face challenges when the original owner of an agent leaves the organization or changes roles, leading to orphaned agents that lack proper management. This can disrupt business processes and create security risks, as there may be no responsible party to manage the agent's settings and compliance requirements.

With this feature, admins (with Tenant admin, Power Platform administrator, or Dynamics 365 Service Administrator roles) can use PowerShell commands to assign new owners to agents.

The new owner must have a Copilot Studio license and must be a member of the environment where the agent is located.

After an admin changes the owner of an agent, the new owner is:

1. Granted access to all the agent's components with standard owner permissions (including delete permissions).
2. Assigned the Environment Maker security role for the environment where the agent exists.

The previous owner will lose any access to the agent. All other co-owners won't be affected by the change.

The new Search within your Agent feature makes it easy for makers to instantly search and navigate to elements across their entire agent, without switching tabs or manually scanning through content.

With a simple keyboard shortcut (Ctrl + J on PC, Command ⌘ + J on Mac) or by selecting the top-level search icon, you can search across Knowledge, Topics, Tools, Skills, and Entities defined in your agent, all from one place.

Search is fast, intuitive, and built to keep you focused. Filters help narrow results and surface the most relevant content. With deep linking and keyboard navigation, you can jump straight to what you need in one click, no interruptions, no wasted time.

Streamline your workflow, stay in control, and build with confidence. Search is always available in the top navigation bar or via shortcut, ready whenever you are.

This feature improves how you can test and improve you actions when building agents:

Enhanced testing experience: By providing clear and actionable testing information, you can easily understand and implement the necessary steps without confusion. This ensures you can maintain the context of your ongoing tasks in Copilot Studio or Microsoft 365, leading to a more seamless workflow.

Debugging and recommendations: Features such as plugin run history, in-line recommendations, and workflow visibility are invaluable. These tools help you trace issues, understand exceptions, and make effective improvements to your actions. This comprehensive approach to debugging ensures that any problems are quickly identified and resolved, enhancing the overall efficiency of the development process.

Simulated test data: Creating a set of test data allows you to verify the functionality of your actions in a controlled environment. This step is crucial for ensuring correctness and reliability before live deployment. By testing in a simulated environment, you can confidently deploy your solutions, knowing they have been thoroughly vetted for any potential issues.

This feature provides a critical layer of security that addresses a significant element of security enforcement and remediation: quarantine.

When an agent is potentially compromised (for example, if it violates data loss prevention policies), it can pose a risk to your entire system. By quarantining the agent, admins can isolate the threat, preventing it from spreading and causing further damage, while they investigate and remediate the issue.

Admins (with the Tenant admin, Power Platform administrator, or Dynamics 365 Service Administrator roles) can quarantine agents with PowerShell, by blocking and un-blocking an agent.

Once an agent is blocked by an admin, makers will see an indication of the agent's blocked status in Copilot Studio. Makers can edit the blocked agent but not test it.

Agent users can't interact with the blocked agent and will see an error message if they try to interact with it.

The integration of single sign-on (SSO) for connectors in Copilot Studio ensures that users of your agents are authenticated for the connectors used in the agent instantly.

By handling authentication for connectors through SSO, you agent users don't need to manually enter login credentials or navigate through multiple screens whenever the agent tries to retrieve information through a connector to a secured service.

When a maker creates an agent, they can configure SSO authentication for their agent's connections. Then, when the user of your agent initiates the connection, all authentication and connection details are automatically handled in the background, using the pre-configured data set by the maker.

Microsoft Teams is a comprehensive collaboration platform within the Microsoft 365 suite, designed to enhance workplace communication and productivity. The data in Teams, such as chat history, meeting recordings/transcripts, shared files, and channel communications, is the fuel that empowers an organization's effectiveness.

With this update, makers can add Teams knowledge sources as a URL endpoint or through a data source selector. This includes messages shared or sent in channels, meetings, and group chats.

When asked a question by the agent user, the agent will respond with information gathered from message threads in Teams.

The availability of the chats and information available depends on the maker and end user's access to the chats, channels, or meetings that have been specified as the source.

This means makers can't connect their agents to Teams chats if the maker doesn't have the meeting invite or isn't a member of the team or channel. Agent users will also only receive responses generated from chats they currently have access to. In the case of mismatching permissions, either the maker or the agent user will be notified that they don't have access to the data source and given options to resolve the problem.

With this capability, makers can now add structured business data from Dynamics 365 Finance and Operations directly into Copilot Studio.

This enables agents to provide high-quality answers based on operational data, such as financials, inventory, or customer records.

Makers can select specific Dynamics 365 Finance and Operations tables to use as knowledge sources, and agents can answer grounded questions over enterprise data with accuracy and context without manual integration or scripting to surface data.

Makers can add the knowledge sources within Copilot Studio, either when creating a new agent or when configuring or updating an existing agent.

Previously, makers had limited visibility into why their files weren’t loading or being used effectively in runtime. With this update, they can now proactively address ingestion issues, reducing dependency on engineering support and speeding up deployment cycles.

This feature enhances the knowledge experience in Copilot Studio by surfacing real-time status updates and error diagnostics for each knowledge source file added.

When a knowledge file fails to load or has issues during processing, makers now see detailed status indicators and contextual troubleshooting guidance directly within the experience.

How to use:

1. Go to the Knowledge Management page in Copilot Studio.
2. Upload or view existing knowledge files.
3. Check the new Status column to verify success or detect issues.
4. Select any flagged status to view tailored troubleshooting instructions.

The feature is automatically turned on for every agent created in Copilot Studio. This feature includes:

• Protection shield across views: Views for makers indicating their agent's security throughout Copilot Studio, giving makers confidence that their agents are covered and secure.
• Agent-specific insights: Provide a detailed view of an agent's security status, so that makers and coders can easily monitor and assess threats and configurations for that agent.
• Actionable suggestions: Explain security information to makers clearly, with straightforward calls to action. This includes guidance on actions makers can take (such as requiring authentication) and other actions that affect the agent (such as an admin policy).
• Session and activity drill-down: Help makers investigate specific sessions, activities, or prompts in test mode with detailed security events and actions. The drill-down also includes the scope of prompt attacks prevented, content moderation issues, and governance enforcement.