閱讀英文版本 編輯

分享方式：

Custom engine agents for Microsoft 365 overview

  • 文章

Custom engine agents are specialized chat experiences built on any large language model (LLM) and tailored for a specific domain or workflows. They allow organizations to create their own agents using tools like the Teams Toolkit, and Microsoft Copilot Studio. Custom engine agents allow you to use your custom orchestrators, foundation models, and custom logic.

An illustration that shows how a custom engine agent in Teams works

Custom engine agents offer advanced functionalities such as task execution and content creation. These capabilities make custom engine agents indispensable tools for streamlining workflows and boosting productivity. They were initially supported on Teams platform only, but now integrated on Microsoft 365 Copilot Chat, where the agents look and feel seamless to declarative agents to end-users.

重要

Custom engine agents for Microsoft 365 Copilot Chat are currently in public preview with limited functionality.

Custom engine agents don't require a Microsoft 365 Copilot license to build, debug, and deploy. They're available to users with Microsoft 365 Copilot licenses or users in tenants with metering enabled via Copilot Chat.

備註

The cost of a custom engine agent depends on the Azure services consumed in the app. To estimate the cost, visit the Azure Pricing Calculator.

Build custom engine agents

You can build custom engine agents in two ways:

Teams AI library and Teams Toolkit

For enterprises seeking solutions that are precisely customized to user demands and scenarios, using the Teams AI library is a great choice. Teams AI Library is based on Microsoft Bot Framework so it comes with all functions that conventional bots can do, plus, it utilizes LLMs that adhere to Microsoft's UX principles.

This option is ideal for developers looking to tailor model selection and orchestration, apply their prior experience with Teams bot development, or make their agents available to all Microsoft 365 users.

Developers can also take advantage of Teams Toolkit to construct their projects. The Teams Toolkit includes prebuilt templates for a seamless project launch, offering an ideal solution for quick starts, easy debugging, and deployment.

Explore templates to get started

Kickstart your custom engine agent development by using Teams AI library with the prebuilt templates that are available in Teams Toolkit.

Template Description
Basic AI ChatBot Build a basic AI chatbot for Teams using Teams AI library.
AI Agent Build an AI agent chatbot for Teams that can make decisions and perform actions based on LLM reasoning. AI Agent template has options to start from scratch or use OpenAI Assistants API.
Chat with your data Expand AI bot's knowledge with your content to get more accurate answers to your questions. Chat with your data supports Azure AI Search, Microsoft Graph Search with Microsoft 365 and SharePoint content, and custom OpenAPI and data ingestion options.

Take a tour of the sample scenarios

Explore sample scenarios available in Teams AI library GitHub repository to understand more about the capabilities you can add in your custom engine agents.

Sample Description C# JavaScript Python
List bot Harness the power of AI, simplify your workflow and bring order to your daily tasks with action chaining capabilities. C# JavaScript Python
DevOps bot Perform DevOps actions such as create, update, triage, and summarize work items. C# JavaScript Python
Twenty questions bot Use LLM capabilities and the concept of user intent by having users guess a secret within 20 questions, grounding the chatbot with some question and answer as a data source. C# JavaScript Python
Math tutor assistant Use OpenAI's Assistants API with the Code Interpreter tool and create an assistant that's an expert on math. C# JavaScript Python
Food ordering assistant Call actions to order food from a fictional restaurant called The Pub with complex interactions. C# JavaScript Python

Microsoft Copilot Studio

Microsoft Copilot Studio is a versatile platform that allows you to build and customize agents using low-code and pro-code tools. It integrates with Azure AI services and offers over 1,000 connectors, enabling the creation of sophisticated, multi-channel conversational agents.

For information about how to use Copilot Studio to build and publish custom engine agents, see the following articles:

Microsoft 365 Agents SDK

Microsoft 365 Agents SDK seamlessly integrates with AI services like Azure AI Foundry and orchestration layers such as Semantic Kernel. It offers interoperability with Copilot Studio, enabling you to enhance and extend existing agents with additional skills. This allows makers to delegate tasks to other agent functionalities. Additionally, you can connect to a Copilot Studio agent from code, gaining access to the full range of features within the Copilot Studio ecosystem.

For more information, see Microsoft 365 Agents SDK (preview).

Privacy and compliance

Custom engine agent prompts and responses in Copilot Chat and Teams are stored in compliance with Microsoft 365 product terms and conditions and are managed as per the customer's instructions. To view and manage this stored data, admins can use Content Search or Microsoft Purview.

Related content