Summary

In this chapter, we set up the development environment required to use OpenAI Agents SDK. We installed the SDK in an isolated Python environment, configured our OpenAI API key, and verified our setup by running a minimal agent.

We also covered key Python constructs that the SDK leans on heavily. We discussed how type hints and docstrings provide metadata that the LLM uses to interpret your tools and how decorators mark functions as callable tools. We then learned about asynchronous execution in Python and why that is relevant for agents. Finally, we saw how Pydantic models validate structured inputs and outputs.

We then built our first real agent, which was a simple customer service assistant, and then extended its capabilities by adding a tool and handoff. We added a new tool that the agent could execute when the customer wanted the status of an order. We did this by creating a custom Python function (get_order_status) that was well-typed and documented, and then...