Capability · Framework — agents

OpenAI Agents SDK

Released by OpenAI in March 2025, the Agents SDK is a lightweight but production-ready framework for orchestrating single- and multi-agent LLM apps. It provides agents with tools, handoffs between agents, guardrails for input/output validation, built-in tracing, sessions for memory, and both Python and TypeScript implementations. It works with the new OpenAI Responses API and third-party models via LiteLLM.

Framework facts

Category: agents
Language: Python / TypeScript
License: MIT
Repository: https://github.com/openai/openai-agents-python

Install

pip install openai-agents
# or
npm install @openai/agents

Quickstart

from agents import Agent, Runner

agent = Agent(
    name='Assistant',
    instructions='Answer concisely.'
)

result = Runner.run_sync(agent, 'What is the capital of India?')
print(result.final_output)

Alternatives

LangGraph — more flexible, not OpenAI-centric
CrewAI — role-based multi-agent
Pydantic AI — typed Pythonic alternative
AutoGen — Microsoft multi-agent equivalent

Frequently asked questions

Is this a replacement for Swarm?

Yes. OpenAI released the Agents SDK as the production-ready successor to the experimental Swarm library. It keeps the handoff/agent concept but adds guardrails, sessions, tracing, and first-class TypeScript support.

Can I use non-OpenAI models?

Yes. The SDK supports any model accessible via the OpenAI Responses API or via LiteLLM (Anthropic, Gemini, open models, etc.). Tracing still works across providers.

Sources

OpenAI Agents SDK — docs — accessed 2026-04-20
OpenAI Agents SDK — GitHub — accessed 2026-04-20