Skip to content
← Tools / Skills
ToolOpenAI/10 trustMetrics-checked
OpenAI Swarm logo

OpenAI Swarm

OpenAI's experimental multi-agent orchestrator — 22K stars, MIT, now in maintenance.

Visit official site →

About

Swarm (22,049 GitHub stars, MIT) is OpenAI's experimental multi-agent orchestration framework: lightweight routines and handoffs between agents, released as a cookbook-style exploration of agentic patterns.

Status: experimental and minimally maintained (last push April 2026). OpenAI positioned it as educational, not production — the patterns informed the Agents SDK. For learning multi-agent handoff patterns, Swarm is excellent; for production, use the OpenAI Agents SDK (already in the catalog as openai-agents) or a maintained framework.

What it can do

  • • Lightweight multi-agent routines\n• Agent handoffs and function calling\n• Cookbook-style educational examples\n• Minimal abstractions (easy to learn)\n• Informed the OpenAI Agents SDK

Setup

  1. Install: pip install git+https://github.com/openai/swarm.git.\n2. Set OPENAI_API_KEY.\n3. Define agents with functions and handoffs; run a Routine.\nSee the GitHub README (cookbook-style examples).

Prerequisites

Python ≥ 3.10; an OpenAI API key.

Use cases

agent-developmentmulti-agent-systemsai-agentseducation

Works with

Any agent (.md)Generic APIopenai-agents

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier). 22K stars and MIT from OpenAI, with genuinely educational handoff patterns. Docked three points: (1) experimental — not for production; (2) minimally maintained (last push Apr 2026); (3) superseded by the Agents SDK. For learning, not shipping.

★ 22,049 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.1/10
Maintenance · recency, releases, issue load
4.8/10
Openness · license + docs
9.1/10
Manual review · sanity read + red-flag check
6.0/10
  • · 22,049 GitHub stars (log-scaled vs catalog peers)
  • · last push 181d ago; 0 releases/12mo; 36 open issues
  • · manual review: experimental, educational; minimally maintained; superseded by Agents SDK; license: MIT

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.