Example demonstrating how to use the reasoning content feature with models that support it, running in the context of Temporal's durable execution.

Adapted from OpenAI Agents SDK reasoning content

Some models, like deepseek-reasoner, provide a reasoning_content field in addition to the regular content. This example shows how to access and use this reasoning content within Temporal workflows. The reasoning content contains the model's step-by-step thinking process before providing the final answer.

This example uses an activity to handle the OpenAI model calls. The workflow orchestrates the process by calling the get_reasoning_response activity, which uses the OpenAI provider to get a response from a reasoning-capable model and extracts both reasoning content and regular content.

The model calls are run in an activity rather than directly in the workflow because Temporal's the involve I/O.

First, start the worker:

uv run openai_agents/reasoning_content/run_worker.py

Then run the reasoning content workflow:

uv run openai_agents/reasoning_content/run_reasoning_content_workflow.py

• Set your OPENAI_API_KEY environment variable
• Use a model that supports reasoning content (e.g., deepseek-reasoner)
• Optionally set EXAMPLE_MODEL_NAME environment variable to specify the model

The original OpenAI Agents SDK example includes streaming capabilities, but since Temporal workflows do not support streaming yet, this example contains only the non-streaming approach.