Deploy custom agents to Microsoft 365 Copilot

  • 4 minutes

While Microsoft 365 Copilot offers powerful built-in and extensible experiences, some organizations need even more flexibility—such as full control over orchestration, model selection, or deployment surfaces. In these cases, developers can build custom agents using tools like the Teams SDK, Copilot Studio, or the Microsoft 365 Agents SDK, and deploy them to Microsoft 365 Copilot.

Benefits of deploying custom agents to Microsoft 365 Copilot and Teams

Custom agents are ideal for organizations that need to go beyond the declarative extensibility model. These agents:

  • Can be built using your own orchestrator, foundation models, and logic.
  • Support advanced capabilities like retrieval-augmented generation (RAG), autonomous agents, and multi-agent coordination.
  • Can be surfaced in Microsoft 365 Copilot Chat and Teams, providing a seamless user experience for Microsoft 365 users.

Choose to build and deploy a custom agent if:

  • You need full control over orchestration, memory, and reasoning.
  • You want to use your own LLMs or integrate with third-party AI services.
  • You’re building a customer-facing or cross-platform AI experience.
  • You want to embed the agent in both Microsoft 365 and external applications.

Limitations of custom agents

  • Requires more development effort and infrastructure.
  • Not built on the Microsoft 365 Copilot orchestrator, so you must manage orchestration, logic, and governance.
  • Requires integration with cloud services. (e.g., Azure OpenAI, Azure AI Search.)
  • May involve additional cost and complexity depending on the services used.

Options for building custom agents that can be published to Microsoft 365

You can build custom agents and publish them to Microsoft 365 using one of the following approaches:

Teams SDK + Microsoft 365 Agents Toolkit

  • Build a custom engine agent using the Teams SDK with OpenAI or Azure OpenAI models.
  • Use Microsoft 365 Agents Toolkit to scaffold, test, and deploy the agent.
  • Provide a rich user experience in Teams using Teams SDK.
  • Publish to Microsoft 365 Copilot.

Copilot Studio

  • Use Copilot Studio to build a custom agent with your own logic, knowledge, and actions.
  • Publish the agent and connect it to the Teams + Microsoft 365 channel.

Microsoft 365 Agents SDK + Microsoft 365 Agents Toolkit

  • Use the Microsoft 365 Agents SDK to build sophisticated, custom AI agents with full control over orchestration, memory, and reasoning.
  • Integrate with Azure OpenAI, Microsoft Graph, external APIs, and other Azure AI services.
  • Publish agents to Microsoft 365 Copilot, making them available in Copilot Chat, Teams, and other Microsoft 365 experiences.
  • Support multi-channel deployment, allowing your agent to be embedded in both Microsoft 365 environments and external applications or websites.

Scenario spotlight: Intelligent travel planning assistant

As part of its travel and expense modernization initiative, the company wants to support executive assistants and travel coordinators who manage complex, multi-leg travel for senior leadership. These users need more than just policy guidance—they need an intelligent assistant that can:

  • Understand nuanced travel preferences. (e.g., preferred airlines, seating, layover limits.)
  • Coordinate across multiple calendars and time zones.
  • Integrate with internal approval workflows and external booking APIs.
  • Generate travel summaries and itineraries in real time.

To meet these needs, the IT team builds a custom engine agent using the Teams SDK and deploys it to Teams. This agent:

  • Uses retrieval-augmented generation (RAG) to pull from internal travel history and preferences.
  • Integrates with Microsoft Graph to check calendar availability.
  • Connects to external travel APIs to suggest optimal itineraries.
  • Generates a formatted itinerary summary that can be shared with executives.

This level of orchestration, personalization, and integration goes beyond what declarative agents can support. By building a custom agent, the team delivers a high-value, intelligent experience tailored to a specialized internal audience.

Analysis

A custom agent is the best fit because the solution needs to be:

  • Capable of advanced orchestration and multi-step reasoning.
  • Integrated with both internal systems (e.g., Microsoft Graph, travel approval workflows.) and external APIs. (e.g., airline booking platforms.)
  • Personalized based on user preferences and travel history.
  • Built and maintained by a development team with full control over logic, memory, and data flow.

A custom engine agent built with the Teams SDK provides the flexibility and control needed to deliver this experience, while still integrating seamlessly with Microsoft Teams.

Why are other options suboptimal?

  • Copilot connectors: Useful for surfacing travel policy documents, but not suitable for building interactive, personalized agents or orchestrating multi-step workflows.
  • Declarative agents in Copilot Studio or Microsoft 365 Agents Toolkit: Great for guided task automation and internal support, but lack the orchestration depth, memory handling, and fine-grained control required for this level of personalization and integration.