The OpenAI Agents SDK is the official, minimal Python (and TypeScript) SDK for building agent workflows with the Responses API. It replaces the earlier experimental Swarm project.
Core primitives
- Agents — an LLM configured with instructions, tools, and guardrails.
- Handoffs — one agent can delegate to another as a first-class tool call.
- Guardrails — input/output validators that can abort a run.
- Tracing — every run is captured in the OpenAI dashboard for replay and debugging.
When to reach for it
- You're already committed to OpenAI models and want the least framework overhead.
- You need multi-agent handoffs (triage → specialist) without adopting a graph library.
- You care about first-party observability and evaluation tooling.
Learn more
- Python docs: <https://openai.github.io/openai-agents-python/>
- Repo: <https://github.com/openai/openai-agents-python>