An app calls a Foundry agent through the Responses API for a complex reasoning task that…

How it works

For long-running operations, the Responses API gives you two delivery modes:

• Streaming returns results incrementally while the connection stays open.
• Background mode runs completely asynchronously.

Background mode runs the agent asynchronously, which Microsoft says is useful for long-running tasks such as complex reasoning or image generation. You set background to true when you create the response, then poll the response status until it completes. In both modes you typically monitor the response until it finishes and then consume the final output items.

The other options, and when you'd use them

• tool_choice none: It only means the model doesn't call tools. It has nothing to do with how long the client must stay connected.
• Stream the response: Streaming delivers partial results incrementally. Fully asynchronous execution is what background mode is for.
• Create a conversation: Conversations are durable objects that store items such as messages, tool calls and tool outputs, and you can reuse them across sessions. They hold history for later turns but don't make one response run asynchronously.

Exam tip

Start, disconnect, come back = background: true + poll. Watch tokens arrive = stream: true. Multi-turn memory = conversation.

References

• Build with agents, conversations, and responses
• Tool best practices for Microsoft Foundry Agent Service