Connected agents overview (GitHub Copilot)

Connected agents are agents your primary agent can invoke during a conversation. When a user's request falls within a connected agent’s domain, the orchestration runtime can route the request to that agent. This approach lets you build modular solutions where each agent handles a specific domain or task.

Note

Features in this article are used by agents or workflows powered by the GitHub Copilot harness.

Usage-based billing applies to using, building, testing, and evaluating agents. These actions might consume Copilot Credits. Learn more in Manage costs for agents powered by the GitHub Copilot harness.

In the new agent experience, you can connect other agents built in Copilot Studio, agents hosted in Microsoft Foundry, remote agents available over the Agent2Agent (A2A) protocol, and agents built with the Microsoft 365 Agents SDK. You add every connected agent from the same Add a connected agent dialog on the Build tab.

Why use connected agents?

Connected agents are useful when you have multiple agents, each specialized in a different area, and you want a single front-door agent that routes users to the right specialist. Connected agents provide:

  • Specialization: Each agent focuses on a specific domain, making it easier to build and maintain.
  • Reusability: A specialized agent can be connected to multiple primary agents.
  • Separation of concerns: Different teams can own and manage different agents independently.
  • Scalability: Add new capabilities by connecting new agents rather than expanding a single agent's instructions.

How connected agents work

  • The primary agent's orchestration runtime evaluates each user message.
  • If the request aligns with a connected agent's domain, the primary agent delegates the request.
  • The connected agent runs in its own orchestration context, with its own instructions, knowledge, and tools. For runtime details, see How the orchestrator routes to connected agents.
  • The primary agent sends the relevant conversation history and user message to the connected agent.
  • The connected agent processes the request and returns a response to the primary agent.
  • The primary agent presents the response to the user.

Note

  • When work is delegated to a connected agent, the orchestrating model creates the task description and can rephrase or expand the user's request. Don't design connected-agent workflows that depend on the original prompt being forwarded verbatim.

  • Connected-agent handoffs are subject to a messaging payload limit, so an oversized file handoff might not reach the primary agent. Return compact structured data when possible, and create or process the file in the primary agent.

  • Last updated on