Choose the right agent solution to support your use case (preview)
[This article is prerelease documentation and is subject to change.]
An agent is an application that takes instructions as input, and uses various layers of AI services and orchestration to carry out the instructions and provide a response.
The most common type of agent is conversational. A conversational agent has access to knowledge for grounding via retrieval augmented generation (RAG). The knowledge base improves the relevancy of the output. A conversational agent can also be connected to other services, providing extra functionality and capability.
This article provides an overview of some of the agent solutions Microsoft provides. If you're new to making agents, we recommend starting with Copilot Studio.
The following table lists Microsoft products and services for building agents, what audiences they support, and a brief description of each. The following sections describe each product in more detail:
| Product | Audience | Description |
|---|---|---|
| Copilot Studio | Fusion teams, citizen developers | Copilot Studio is an end-to-end agent-building tool, with built-in natural language understanding models, data connectivity through Power Automates, and support for multiple channels. |
| Microsoft 365 Agents SDK | Developers | Provides a framework for building agents, including tools, templates, and related AI services. The SDK is ideal for developers who want to build agents that are publicly available on the Microsoft Teams app store. |
| Azure AI Foundry SDK | Developers | Azure agents provide agent development capabilities under a unified Azure AI SDK. |
| Semantic Kernel SDK | Developers | The Semantic Kernel SDK allows developers to build agents and integrates various AI services to define orchestration chains for rapid development. |
Copilot Studio
Copilot Studio is a tool for agent development included in Microsoft Power Platform—a business-application platform that incorporates data analysis, solution building, and process automation. You don't need to write code or understand the details of the underlying AI technologies to build agents in Copilot Studio. Agents built in Copilot Studio can apply automation and other capabilities within the Power Platform, and you can rapidly develop sophisticated agent experiences.
You can connect virtual agents to various user platforms, such as Microsoft 365 and Microsoft Dynamics 365.
You can use over 1,000 prebuilt data connectors, available through Power Automate.
For more information about Copilot Studio, see the product overview page. For details about pricing, see Copilot Studio pricing.
Microsoft 365 Agents SDK
The Microsoft 365 Agents SDK is an SDK for building self-hosted agents. It's a collection of libraries and tools that allow you to build an agent in code, facilitating the communication between client and agents by handling the conversation between them. It provides an easy path to incorporate AI services that Microsoft provides, such as Graph, Azure OpenAI API, and also non-Microsoft AI services.
With these tools, developers can create agents that can be deployed to any channel of choice, with scaffolding to handle the required communication. Developers can also use numerous generative AI services that are secure and compliant, available on any channel you require, using the Azure Bot Service to facilitate the hosting and adapters.
For details about pricing, see Azure AI Bot Service pricing. Costs vary depending on what AI services are used for implementing and hosting your code.
Azure AI Foundry SDK
Azure AI Foundry includes many development capabilities under a unified Azure AI SDK tailored to suit specific needs or work with specific services. Azure Agents is the name of the feature within the Unified Azure AI SDK that also includes fine-tuning, evaluations, including agents.
The agent capabilities include API layers such as Multi-agent API, Assistants API, Responses API and Completion API. These API layer provide various amounts of capabilities to support multi-agent, RAG, and AI services.
Developers can add and use Azure Agents APIs from Assistants and broader APIs including Azure OpenAI Services and Azure Cognitive Services to add functionality and capabilities to the agent.
Semantic Kernel SDK
Semantic Kernel SDK allows developers to build agents and integrates various AI services to define orchestration chains for rapid development.
As with Azure Agents, developers can choose to use Semantic Kernel in the middleware or agent logic components of an Agents SDK agent to add capabilities such as the ability to easily orchestrate between APIs or plugins.
Agent skills in Copilot Studio (formerly Bot Framework skills)
Copilot Studio provides a range of functionality to create different types of agents, including copilot agents to extend Copilot for Microsoft 365. Copilot Studio lets you add hosted logic written in C#. These agent skills can be invoked within a topic, or used by the orchestrator with control returned to Copilot Studio. Skills allow you to extend Copilot Studio functionality even further.
An agent skill needs to be developed, and deployed and hosted (for example, on Azure), and then configured within Copilot Studio. Here is some additional information about agent skills: