Agency Swarm is an open-source multi-agent orchestration framework by Arsenii Shatokhin (VRSEN), created to automate his own AI agency with collaborative teams of agents. Built on the OpenAI Agents SDK and Responses API, it models agent systems as real-world agencies with defined roles and communication flows. It has more than 4,600 stars on GitHub and is under active development.
Most agent frameworks leave structure to chance. Agency Swarm instead mirrors how a human agency works: a CEO agent receives the mission, specialized agents such as developers, designers and copywriters do the work, and explicit send-and-receive flows govern who can talk to whom. Each agent is defined in code with its own instructions, tools and settings, and agencies compose these roles into a chart that orchestrates the whole workflow. The framework adds practical touches for reliability, including persistent conversation state, custom tool creation from functions, and support for running complete agencies locally or behind an API endpoint.
The framework also includes tooling to convert existing code and API schemas into agent-ready tools, and agency definitions live in ordinary code files that can be versioned and reviewed like any other software. Community examples cover agencies for development, marketing and research tasks.
Agency Swarm is free and open source under the MIT license. Your only costs are the OpenAI API calls your agents make; the framework itself has no paid tier.
Agency Swarm is for developers and indie hackers building structured multi-agent products, especially those who think naturally in terms of teams and roles and want dependable orchestration instead of free-form agent chatter.