Creativity · Agent Protocol
OpenAI Swarm Framework
OpenAI Swarm is an educational open-source multi-agent framework, released in 2024, that modeled multi-agent systems around two primitives — agents and handoffs. Each agent had instructions and tools; a handoff tool transferred the conversation to another agent. Deliberately stateless and minimal, Swarm was not intended for production but seeded the concepts that became OpenAI's Agents SDK in 2025.
Protocol facts
- Sponsor
- OpenAI
- Status
- deprecated
- Spec
- https://github.com/openai/swarm
- Interop with
- OpenAI Agents SDK, function calling
Frequently asked questions
Is Swarm still the right thing to use?
No — OpenAI explicitly positions Swarm as educational and points production users at the Agents SDK, which covers the same patterns with proper tracing, guardrails, and session management.
What was Swarm good for?
Understanding the handoff pattern clearly. Because Swarm was so small, you could read the whole framework in an afternoon and see exactly how peer-to-peer agent systems work.
How does Swarm map to A2A?
Swarm is in-process — agents share a Python runtime. A2A is the over-the-wire version of the same ideas. The same handoff concept applies; A2A just adds a standardised network boundary.
Sources
- OpenAI Swarm — educational repo — accessed 2026-04-20