Swarm

Swarm is an experimental educational framework from OpenAI designed to explore lightweight and highly controllable multi-agent orchestration using the Chat Completions API. It provides developers with hands-on experience in building scalable solutions by focusing on two core concepts: Agents, encapsulating instructions and tools, and handoffs, enabling agents to seamlessly pass conversations between each other. This approach allows for the creation of complex dynamics between agents and tools, proving ideal for scenarios with numerous independent capabilities that are difficult to encode into a single prompt.

Swarm prioritizes a lightweight and highly customizable experience, operating stateless on the client without persistent memory between calls, much like the underlying Chat Completions API. It introduces a run() function which manages agent function execution, hand-offs and context variables, and takes multiple turns before returning to the user. This allows developers to focus on composing diverse agents, workflows and tasks into a cohesive network. Swarm isn’t a fully-featured library, but a tool for developers curious about multi-agent orchestration patterns to experiment and learn.