Build AI agents with OpenAI Agents SDK

Prerequisites

Step-by-Step

1. 1

   Install the SDK

   The package is openai-agents. It depends on openai>=1.x and pydantic.

   pip install openai-agents

2. 2

   Define a basic agent

   An Agent is name + instructions + tools + model. Keep instructions short and operational.

   from agents import Agent, Runner
   
   triage = Agent(
     name='Triage',
     instructions='Route the user request to the right specialist agent.',
     model='gpt-4o-mini'
   )

3. 3

   Add tools

   Tools are typed Python functions. The SDK auto-generates JSON schemas from your signatures.

   from agents import function_tool
   
   @function_tool
   def get_weather(city: str) -> str:
       """Return current weather for a city."""
       return fetch_weather(city)

4. 4

   Set up handoffs

   Handoffs let one agent delegate to another. The SDK handles the transcript reset.

   weather_agent = Agent(name='Weather', instructions='You answer weather questions.', tools=[get_weather])
   triage.handoffs = [weather_agent]

5. 5

   Run with Runner

   Runner.run is the entry point. It returns a result object with output, messages, and trace metadata.

   result = Runner.run_sync(triage, 'What is the weather in Tokyo?')
   print(result.final_output)

6. 6

   Add tracing

   Set OPENAI_AGENTS_DISABLE_TRACING=0 and view runs in the OpenAI dashboard. Critical for debugging in prod.

Common Pitfalls

What's Next

• ->Add guardrails to validate inputs and outputs.
• ->Persist conversation state for multi-turn use.

More Build AI agents