OpenAI Swarm
multi-agentOpenAI Swarm was an experimental, educational multi-agent framework released by OpenAI in late 2024. It has been superseded by the OpenAI Agents SDK, which provides the same handoff-based patterns with production-grade tracing, guardrails, and sessions.
Why It's on Hold
Swarm was explicitly labelled by OpenAI as an "educational framework exploring ergonomic, lightweight multi-agent orchestration" — not a production tool. In March 2025, OpenAI replaced it with the Agents SDK, which builds on the same conceptual primitives (agents + handoffs) but adds everything Swarm deliberately omitted:
| Feature | Swarm | Agents SDK |
|---|---|---|
| Built-in tracing | ❌ | ✅ |
| Guardrails | ❌ | ✅ |
| Sessions | ❌ | ✅ |
| Human-in-the-loop | ❌ | ✅ |
| Production support | ❌ | ✅ |
| Stateless design | ✅ (intentional) | Optional |
When to Use Swarm
The only reason to use Swarm today is to read its source code as a learning exercise — it's deliberately minimal (~300 lines) and clearly explains the agent-handoff pattern. Once you understand the concepts, move to the Agents SDK for anything real.
Migration Path
Switch to the OpenAI Agents SDK:
pip uninstall swarm
pip install openai-agents
The concepts map directly: Swarm's Agent and handoff become Agents SDK's Agent and handoffs.