Swarm

Swarm is an open-source multi-agent orchestration framework released by OpenAI on GitHub, where it has earned more than 22,000 stars. Billed as an experimental, educational project, Swarm explores how networks of lightweight AI agents can coordinate and hand work off to each other using nothing more than the Chat Completions API. OpenAI notes that Swarm has since been superseded by the production-ready OpenAI Agents SDK, but the repository remains one of the clearest reference implementations of multi-agent patterns, and it was still being updated in 2026.

Overview

Swarm focuses on making agent coordination and execution lightweight, highly controllable, and easy to test. It is built on two simple primitives: Agents, which encapsulate instructions and tools, and handoffs, which let one agent transfer a conversation to another. Because Swarm runs almost entirely on the client side and stores no state between calls, developers can inspect every step of an interaction and compose agents into networks that mirror real workflows, such as triage desks, support bots, and multi-step customer service flows.

Key Features

• Two core primitives — Agents and handoffs — express rich multi-agent dynamics with minimal code
• Agents encapsulate instructions, functions, and model settings, and can switch context mid-conversation
• Automatic conversion of Python functions into JSON Schema tool definitions
• Context variables let functions share and update state across agents and turns
• Streaming support and a built-in demo loop for interactive experimentation
• Example projects include airline customer service, weather, support bot, and personal shopper agents
• Stateless design powered entirely by the Chat Completions API

💳 Pricing

Swarm is completely free and open source under the MIT license. You only pay for your own OpenAI API usage when running agents.

Who Is It For

Swarm is ideal for developers and students who want to learn multi-agent orchestration patterns from a minimal, readable codebase, and for teams prototyping agent handoff designs before adopting a production framework.