**Disclaimer**\
\
This tutorial has been shared as a preview to the complete [Building a Custom Agent in Copilot Studio](https://damobird365.teachable.com/p/damobird365-membership) course. Please make sure you sign up to the new community which includes both this course (launching in January) and a Power Automate Cloud Flow course due to be launched in June 2025. You'll also get access to the community forum and fortnightly calls with Damien where you can learn more about the Power Platform and even share your own builds and learning. \
\
You can check out his [YouTube DamoBird365 (15,000+ subscribers) here](https://www.youtube.com/c/DamoBird365?sub_confirmation=1)

Navigate to [https://copilotstudio.microsoft.com](https://copilotstudio.microsoft.com/environments/Default-16c901d1-9763-49b1-961c-6cd701f5d0f7/bots/f57625fa-c5b8-ef11-a72f-000d3ad4aa08/overview)

Here, we have the custom Agent that you will learn to build throughout the course. For now, we'll take a tour and explore how to navigate around Copilot Studio

Instructions 
You are a friendly, helpful, polite HR Agent that will use knowledge to answer questions about HR Policies and Procedures, as well as helping Employees request annual leave.  Do not answer questions unrelated to this topic.

You are a friendly, helpful, polite HR Agent that will use knowledge to answer questions about HR Policies and Procedures, as well as helping Employees request annual leave.  Do not answer questions unrelated to this topic.

The overview tab, will summarize key information that can also be configured elsewhere.  It has a **description** for the maker, user and Agent understanding as well as some important **instructions** to guide the agent on its capability and behaviour as well as a toggle for Generative **Orchestration**

Analytics
Check your agent's key performance info from the last 7 days.
Open Analytics
Total sessions
3
Engagement
67%
Satisfaction score
--

There are details about the **analytics**, **total sessions**, **engagement** and **satisfaction score** of our Agent. Useful for driving user adoption and iterating on the Agents success to deliver return on investment

**Knowledge**, which relates to the data that our Agent is grounded on. It can also include **general knowledge** based on what it was trained on via the web, you can disable this.

**Actions**, these are native to the Power Platform and relate to a connector, for instance, **SharePoint**. Orchestration must be enabled to allow the Agent to called these dynamically using Generative AI.

**Triggers**, which relate to Power Automate triggers and **Autonomous Agents**. Ultimately a Cloud Flow will be triggered, for instance via email, and call our Agent to take appropriate action.

**Topics**, which relate to either system or custom Topics. When Orchestration is off, these are called based on Topic Trigger Words, otherwise Generative AI will reason over the Topic Description.

And finally the **Published Status**, if and when an Agent was last Published

Published agent status
Published by DamoBird365 12/30/2024, 9:03 PM

Along the top navigation panel there are a series of tabs.

The **Knowledge** tab lists all of the grounded knowledge that the agent has access to.  This can be queried via a Generative Answers node in a default System Topic or a custom Topic with granular access to content

By clicking on **Add knowledge**, you can attach various data sources that are natively available to our agent including public websites, SharePoint, Dataverse, Upload Files or use Advanced capabilites such as Azure AI Search

Under the **Topics** section, you'll find all available Topics, both custom and system-defined. Keep in mind that the Trigger method varies based on whether Orchestration is enabled or disabled. You can easily view the last update timestamp for each Topic and toggle them on or off as needed. Topics play a crucial role in providing a structured and organized approach to conversations, ensuring clarity and relevance in your agent's interactions.

If you click on and select a Topic, for example **Conversation Start**

Trigger
On Conversation Start
Edit
Message
Text

Speech variation available.
Hello, I'm 
Bot.Name
string
​, a custom agent built by Damo. Just so you are aware, I sometimes use AI to answer your questions. If you provided a website during creation, try asking me about it! Next try giving me some more knowledge by setting up generative AI.
Message
Text

Add

Enter a message


Quick replies
Can I request annual leave?

How can I become a digital champion?

Do I have any approved annual leave?

Tell me about the salary grades?


You will be presented with the **Authoring Canvas** which enables you to configure the topic workflow, adding **nodes** for questions, messages or logic and capturing variables from an Agent conversation

You can add nodes to your Topic by clicking on the **+** and then choosing from the following options:

- **Send a message** to the user
- **Ask a question** to the user and capture the response as a variable
- **Use an adaptive card**, which is a platform-agnostic JSON payload for creating custom cards that capture a response from the user
- **Add a condition** to determine an outcome based on branching logic

Via **Variable management** you can **set a variable value,** **parse a value** from custom data or **clear all variables**

From **Topic Management**, you can:

- Navigate to another Topic
- End the current Topic or all Topics
- Transfer to an Agent
- Go to a specific step in a Topic
- End the conversation

From **Call an Action**, you can:

- Call or create a Cloud **Flow** or AI Builder **Prompt**
- Call a Bot Framework **Skill** or specific Power Platform **Action** from a Standard/Premium or Custom Connector

Under **Advanced**, you can add:

- **Generative Answers**: Add this node to configure knowledge with granular knowledge sources
- **Send an HTTP Request**: Call an open API endpoint
- **Log Custom Telemetry**: For example, log a failure to Azure Application Insights
- **Send an Event**: Enables the Agent channel to intercept events
- **Send an Activity**: Enables the Agent to respond to various activity types
- **Authenticate**: Enable user authentication based on authentication provider settings

Copilot
Comments
Variables
Topic checker
Details
More
Save

At the top of each Topic, you can:

- Use **Copilot** to describe and implement changes.
- Add **Comments** for collaboration or documentation.
- Review **Variables**, including Topic, Global, and Environment variables.
- Check the Topic for warnings or errors to ensure accuracy.
- View **Details** such as the name, description, and input/output variables, which can be dynamically captured when Orchestration is enabled.

By clicking on the ellipses (**... More**), you can:

- Access analytics for this Topic to review engagement.
- Edit the Topic using a YAML **code editor**.
- Reset the Topic to its default settings if custom changes have been made, provided this is a system Topic.

Under the **Actions** section, you can see any of the Power Platform connector actions that have been added specifically to our Agent. In this case, an action to **Get Items** from **SharePoint**

With **Orchestration** enabled, actions can be Triggered using **Generative Actions** otherwise they must be called via a Topic

On the **Activity** tab (with **Orchestration** enabled), you can see the history of previous conversations either as specific users where they are authenticated or as an Autonomous Agent

You can see the history of previous interactions by clicking on one of these activities and explore the activity map.

From the **Activity map** you can explore a session and **Review** the Agents reasoning

You can see that the Agent used Knowledge sources on SharePoint and Files uploaded to retrieve details about salary grades

This action needs to be done to retrieve the salary range for grade 9 from the organization's resources.

You can also expand the **Activity map** to include a conversastion **Transcript**

The Agent was specifically asked the salary range for grade 9 and was able to respond, including a link to its source.

Under the **Analytics** tab, you can explore user engagement over a period of time defined by a drop down.  This includes **Total sessions**, **Engagement** rate and an overall **Satisfaction score**

You can check the effectiveness of your sessions by clicking **See details** and then based on a user Abandoning, Escalating or ending their session as Resolved, which is noted when the **End of Conversation** system Topic is called and a Satisfaction Score is requested

The success of your Knowledge source use can also be explored clicking **See details** enabling you to visualling check the relevance to the data you have grounded your Agent on

All sources
The percentage of engaged sessions that used each source. Sources with low use might need more specific instructions. Learn more
Pay Grades 2024 2025
100%
Human Resources HR
0%

Finally if a user has provided a feedback score upon completing a session, it will be noted here under **User feedback**

Not enough feedback yet
After users submit survey responses, you'll see them here.

On the **Channels** tab, you can make your Agent accessible to users, whether they are internal employees or external customers. The availability of specific channels may depend on the authentication type or your organization's Data Loss Prevention (DLP) policy, which might restrict access to certain channels. This is the central location where you configure your agent for end-user availability.

A common internal deployment involves making your Agent accessible directly within Microsoft Teams. Your Teams administrator can even pin it to the Teams navigation panel for easy access.  Perfect for an organisation wide HR Agent

The **Custom Website** channel allows you to embed your Agent directly into your organization's website, providing seamless interaction for visitors. You can customize the Agent's appearance and behavior to align with your brand, ensuring a consistent user experience. With straightforward setup instructions, you’ll be able to configure the Agent for various use cases, such as customer support, FAQs, or lead generation.

With Dynamics 365 Customer Service, your Agent can seamlessly hand off complex queries to a live representative, ensuring smooth transitions with full context preserved. This integration enhances customer satisfaction by combining efficient automation with personalized human support.

You can check out a demo of this process via my [YouTube video](https://youtu.be/99j-iGxJJ7c)

Over on the top right, there is a **Test** button which will enable you to interact with your Agent

Click the "Ask a question or describe what you need" field.

You can ask questions directly to your Agent before going live

From the test pane, clicking the ellipsis (**...**) provides access to various settings related to testing your Agent. These include options to test autonomous triggers, manage connections for different actions, flag issues with Copilot Studio, and enable debug mode to return additional information while testing your Agent.

At the very top of the pane, you'll find a **Publish** button to make changes made to your Agent live, along with a **Settings** button to access additional configuration options.

Published 12/30/2024
Publish
Settings
Test

From the top Navigation pane, by clicking on the ellipsis (**...**), you can export or import an Agent. You can also view the solution, which is a Power Platform concept for application lifecycle management, delete the Agent, or share it with others, allowing them to access or make changes to the Agent

Let's take a look at the additional **Settings**

Similar to the Overview tab, you can configure the **Agent name** and change the **icon**

Agent details
Your agent's name and appearance will be applied to your web chat controls and all configured channels. To customize the agent's look for specific integrations, go to Channels. Learn more
My agent's name
 
View Solution (Common Data Services Default Solution)

Icon
Your agent's name and icon appear on chat controls across all configured channels. Go to Channels to customize each individually.
Change icon
Used to represent the agent inside the chat canvas. Icon should be in PNG format and less than 30 KB in size.
Save

How should your agent interact with people?
Classic – Use the topics you build to respond to trigger phrases—actions can only be called from inside a topic.
Generative (preview) - Use generative AI to respond with the best combination of actions, topics, and knowledge.
Learn more about how to optimize your agent.
How strict should the content moderation be?
Low - More creative
Medium - More balanced
High - More precise
Image input
Users can upload images that agents can analyze and use to provide responses

On the **Generative AI** tab, you can enable generative capabilities. This setting is also available in the **Overview** panel, where you'll find the orchestration toggle. By adjusting this toggle, you can influence orchestration settings.

You can **moderate the content** to adjust the bot's responses, allowing it to either be more creative or more precise based on the type of answers you're looking to surface. By fine-tuning these settings, you can strike the perfect balance between imagination and accuracy, tailoring the bot’s tone and focus to better match your needs.

You can enable the ability to **upload images** into the chat, allowing the Agent to analyze and reason over the uploaded files.

Enhanced search results
Can provide improved search performance for Microsoft 365 Copilot tenants. Availability varies by data source. Learn more
Disabled

For those who already have Microsoft 365 Copilot licenses, you can enable **enhanced search results** to retrieve data from your knowledge sources and will improve upon the results being retrieved with the ability to reason over much larger files

Over on the **Security** tab you can change the default **Authentication** which is primarily configured for Teams integration with Entra ID to identify your users. There is advanced capability under **Web channel security** should you want to enable Secured access when the Agent is published via online **Channels**

By clicking on **Authentication**, you can view the default method using Entra ID, disable it entirely (which should be carefully considered), or integrate with third-party authentication

No authentication
Publicly available in any channel
Authenticate with Microsoft
Entra ID authentication in Microsoft Teams, Power Apps, or Microsoft 365 Copilot
Authenticate manually
Set up authentication for any channel

On the **Authoring Canvas** tab, you can adjust the behavior of how it renders while building

The **Entities** tab allows you to define data structures that your agent uses to understand user inputs. Entities can be either **Regex** patterns or predefined **Choices**, helping the agent recognize and process specific information like dates, locations, or products. You can create custom entities via **Add an entity** or use system-defined ones, allowing for more accurate, context-aware interactions based on user input.

Under **Skills,** an advanced feature, **i**f you have already built and deployed bots in your organization (using the Bot Framework SDK and pro-code tools) for specific scenarios, you can convert those bots into a skill and register that skill here

The **Voice** tab enables you to configure voice interaction settings, allowing your Agent to communicate with users via speech for a more dynamic conversational experience

Under **Languages**, your Agent is configured by default for English. While you can configure additional language support, it's worth referring to the current documentation for supported languages and their associated capabilities.

On the **Language Understanding** tab, an advanced capability, integrating a Conversational Language Understanding (CLU) model with Microsoft Copilot Studio enhances your Agent by enabling it to better understand and respond to user conversations. CLU is a custom version of Natural Language Understanding (NLU) designed specifically for conversational applications. This integration allows you to map CLU intents to trigger Agent dialogs, making interactions more natural and satisfying, and is especially useful for more complex conversational needs where specialized language or lingo is required.

Under **Component Collections**, reusable agent components such as topics, knowledge, actions, and entities can be shared across multiple agents within your environment. You can also export and import these collections to transfer content between different environments, making it easier to manage and reuse components across various applications.

Lastly, on the **Advanced** tab, you can connect your Agent to **Application Insights** in Azure to capture custom telemetry, access key **Metadata**, and even **view the solution** contained within your Power Platform environment. This is especially useful for those looking to migrate their Agent between Test, Dev, and Prod environments.

Agent app ID
7ad67c92-93b7-4306-acd5-179d64992c72

If you've enjoyed this tutorial, make sure you sign up to the new community which includes both this course (launching in January) and a Power Automate Cloud Flow course due to be launched in June 2025. You'll also get access to the community forum and fortnightly calls with Damien where you can learn more about the Power Platform and even share your own builds and learning.

[Sign up here](https://damobird365.community.teachable.com/)

Damobird365

This guide offers a comprehensive walkthrough of Copilot Studio, empowering users to build custom agents with advanced capabilities such as knowledge integration, variable management, and analytics tracking. By exploring features like conversation management, user authentication, and testing tools, users can optimize their agents for improved interaction and user satisfaction. Overall, this guide is essential for anyone looking to leverage Microsoft Copilot Studio effectively.

A Tour of Copilot Studio

Copilot Studio Walkthrough

Knowledge

Topics, Authoring Canvas and Nodes

Actions

Activity

Analytics

Channels

Test your Agent and additional controls

Additional Settings