OpenAI Agents SDK
What This Article Covers
- What the OpenAI Agents SDK is and why you’d use it.
- How to install it.
- How to build your first agent with a tool.
- How to work locally or with the OpenAI API.
- Where it excels and what its limitations are.
Introduction
The OpenAI Agents SDK is OpenAI’s official Python framework for building agents. It provides tool-calling capabilities, state management, and workflow orchestration. Installation is straightforward, and you can get started quickly. If you want to jump straight into working with powerful models, this is a solid choice. You can also run it locally via OpenAI-compatible endpoints like Ollama, though the framework is primarily designed around OpenAI’s APIs.
OpenAI Agents SDK at a Glance
The SDK works with agents, tools, and runners. An agent receives instructions and can invoke tools. A runner executes requests and manages the workflow. Tools are Python functions that the agent can call. This lets you build an agent that can research, perform calculations, or consume external APIs.
Installation
pip install openai-agentsTo use OpenAI, you’ll need an API key:
export OPENAI_API_KEY="your-key"If you want to use an OpenAI-compatible local model like Ollama, also set the base URL:
export OPENAI_BASE_URL="http://localhost:11434/v1"Getting Started: An Agent with a Tool
from agents import Agent, Runner
def current_year() -> str:
return "2026"
agent = Agent(
name="helpful_assistant",
instructions="You help answer simple questions.",
tools=[current_year]
)
result = Runner.run_sync(agent, "What year is it?")
print(result.final_output)In this example, you register a Python function as a tool. The agent decides on its own whether to call it.
What Is This SDK Good For?
| Use Case | Benefit |
|---|---|
| Quick prototypes | Simple tool-calling and runner API |
| Cloud agents | Direct integration with OpenAI models |
| API workflows | Agents access your own functions |
| Learning | Minimal boilerplate, fast results |
Common Pitfalls
- API costs - OpenAI models aren’t free. For local testing, integrate Ollama instead.
- No native multi-agent orchestration - For complex multi-agent teams, CrewAI or AutoGen are often more flexible.
- Base URL matters - With Ollama, the URL must end with
/v1and any API key will work.
Further Reading
FAQ - Common Questions
Is the OpenAI Agents SDK free?
The SDK itself is free. Costs only appear when you use the OpenAI API. Local models via Ollama are free.
Can I run it locally?
Yes, via OpenAI-compatible endpoints like Ollama. Set OPENAI_BASE_URL to http://localhost:11434/v1.
What do I need to get started?
Python, pip install openai-agents, and either an OpenAI API key or a local Ollama model.