OpenAI Agents SDK
OpenAI's 2025 production framework for building agents — successor to the older Assistants API, with first-class handoffs, guardrails, tracing, and tool use.
The Agents SDK is OpenAI's opinion on how production agents should be built. Core primitives: Agent (an LLM with instructions and tools), Handoff (delegate to another agent), Guardrail (input/output validation), and Runner (the loop). Available in Python and TypeScript.
It is intentionally minimal — fewer concepts than LangGraph, more opinionated than raw API calls. The trade is speed-to-prototype against ceiling: simple agents ship faster on Agents SDK; complex orchestration is more natural in LangGraph.
In 2026, the Agents SDK is the default choice for teams already running on OpenAI infra. It plugs into the broader OpenAI stack (Realtime, Files, Evals) cleanly.
Frequently asked
Agents SDK vs. LangGraph — which should I choose?+
Agents SDK if you are OpenAI-native and want fast prototyping. LangGraph if you need model portability, complex state machines, or human-in-the-loop interrupts.
Does the Agents SDK lock me into OpenAI models?+
Mostly yes — the SDK is designed around OpenAI's tool format and reasoning chain. You can swap models via the chat-completions endpoint, but you lose some SDK features.
