Custom engine agent
Concept
A custom engine agent is a Microsoft 365 Copilot agent type in which the developer chooses the model and orchestrator, for example built with the Microsoft 365 Agents Toolkit, as opposed to declarative agents that run on Copilot's own orchestrator.
함께 다룬 주제
Custom engine agent 관련 에피소드에서는 Microsoft 365 Copilot Chat, HIPAA, Agent Builder (Microsoft 365 Copilot), Semantic Index for Microsoft 365 Copilot, SharePoint agents 및 Predictive analytics도 다룹니다.
Description
A custom engine agent is a type of Microsoft 365 Copilot agent in which the developer selects the AI model and the orchestrator. This sets it apart from declarative agents, which run on Copilot's own orchestrator. In practice, a custom engine agent is used when a team wants full control over how requests are interpreted, which model produces answers, and how external data or actions are handled, while still making the result available inside the Microsoft 365 Copilot experience.
According to the episode "Extend Microsoft 365 Copilot with Agents, Plugins, and Connectors", a custom engine or custom engine agent plugs into the Copilot extensibility landscape as an external or hosted processing component. It can provide grounding, domain-specific reasoning, or actions. Developers can register a custom engine to surface specialized knowledge or run workflows, so that Copilot can call the engine and return its responses to the user. The episode places this within enterprise AI scenarios in which developers extend Microsoft 365 Copilot's capabilities, ground responses with organizational data, and let users take actions directly from Copilot chat or other Copilot experiences.
The tooling differs by skill level and agent type. The episode "Connect External Business Data with Microsoft 365 Copilot Connectors" lays out a comparison: for declarative agents, low-code options are Microsoft 365 Copilot and Copilot Studio, while the pro-code option is Visual Studio Code with the Microsoft 365 Agents Toolkit. For custom engine agents, the low-code option listed is Copilot Studio, and the pro-code options are Visual Studio Code with the Teams SDK, or Visual Studio 2022. The same episode notes that plugins can be added to give Copilot new abilities or to connect to unique data, for instance through Microsoft Graph Connectors, so that Copilot can be tailored to tasks such as analyzing sales data, managing customer support, or accessing custom databases. The short description for this concept also mentions that a custom engine agent can be built with the Microsoft 365 Agents Toolkit.
The episode "Build Custom Microsoft 365 Copilot Plugins" gives example scenarios that show where each approach fits. A Copilot connector suits customer information integration, for example a Salesforce connector for quick access to customer data. A declarative agent with an API plugin suits a case such as a healthcare compliance assistant. A custom engine agent is listed for a legal research AI that integrates legal databases for case law analysis. The same episode describes project templates as a way to start quickly and avoid setup mistakes, naming the Yeoman generator, Visual Studio, and the Microsoft 365 Agents Toolkit as tools that create the basic files and folders needed.
The term also appears in the context of Microsoft Teams. In the episode "Build Microsoft Teams Meeting Apps and Bots", a custom engine agent is listed among the main bot types for Teams meetings. It is described as using AI to perform automated tasks, understanding natural language, and being able to engage in conversations. The episode contrasts it with a conversational bot that simulates interactions through text and interactive cards, a notification bot that sends alerts in channels or chats, and a workflow bot that automates business processes by managing tasks.
When deciding between agent types, the sources point to a few considerations. A declarative agent relies on Copilot's own orchestrator and is configured rather than coded around a chosen model. A custom engine agent is the option when the developer needs to bring their own model and orchestration logic, for example for domain-specific reasoning over specialized data. The choice of low-code or pro-code tooling then follows the team's skills: Copilot Studio on the low-code side, and Visual Studio Code with the Teams SDK or Visual Studio 2022 on the pro-code side, with the Microsoft 365 Agents Toolkit also named as a build option.
Overview compiled with AI from M365.FM episodes and the facts on this page. You can improve it with the edit button below.
| Type | Concept |
|---|
Community wiki
Short info above, community knowledge below: add practical tips, examples or corrections. New contributions are reviewed first; trusted members publish directly.
Experts and companies for this topic
Nobody has listed this topic yet. Add it to your profile.
Maintained by Mirko Peters (Microsoft MVP) · Last updated: