Agency Swarm

Agency Swarm: Reliable Multi-Agent Orchestration Framework

Agency Swarm is an open-source framework designed to simplify the creation of AI agencies by leveraging and extending the OpenAI Agents SDK. It facilitates building, orchestrating, and managing collaborative swarms of AI agents, enabling true agent collaboration with user-defined communication flows.

What is Agency Swarm?

Agency Swarm enhances the OpenAI Agents SDK by introducing features like true agent collaboration with flexible communication flows, an Agency class for defining complex communication, flexible conversation persistence, and a specialized send_message tool for agent communication. It allows developers to think about automation in terms of real-world organizational structures, making it intuitive for both agents and users.

Key Features:

• Customizable Agent Roles: Define distinct agent roles (e.g., CEO, Virtual Assistant, Developer) with tailored instructions, tools, and capabilities.
• Full Control Over Prompts/Instructions: Maintain complete control over each agent’s guiding prompts for precise behavior customization.
• Type-Safe Tools: Develop tools using Pydantic models for automatic argument validation.
• Orchestrated Agent Communication: Agents communicate via a dedicated send_message tool, with interactions governed by explicit, directional communication flows.
• Flexible State Persistence: Manage conversation history by providing load_threads_callback and save_threads_callback to the Agency, enabling persistence across sessions (e.g., DB/file storage).
• Multi-Agent Orchestration: Build agent workflows on the OpenAI Agents SDK foundation, enhanced by Agency Swarm’s structured orchestration layer.
• Production-Ready Focus: Built for reliability and designed for easy deployment in real-world environments.

How does Agency Swarm work?

Agency Swarm builds upon the OpenAI Agents SDK by introducing the concept of an Agency which manages multiple agents and their communication flows. The key components include:

• Agency: An orchestrator that defines communication flows between agents using the > operator. It also handles conversation persistence through callback functions.
• Agent: An extension of the base SDK Agent with built-in file handling and sub-agent registration. Agents can be assigned specific roles, instructions, and tools.
• Communication Flows: Explicit, directional communication pathways defined on the Agency that govern how agents interact with each other.
• Tools: Custom tools developed using Pydantic models or converted from OpenAPI schemas, allowing agents to perform specific tasks.

How to use Agency Swarm?

1. Installation: Install the package using pip: pip install -U agency-swarm
2. Set OpenAI Key: Set your OpenAI API key in a .env file or export it in your shell.
3. Create Tools: Define custom tools using the @function_tool decorator or extend BaseTool.
4. Define Agent Roles: Define agent roles with specific instructions, tools, and capabilities.
5. Define Agency Communication Flows: Establish communication flows between agents using the Agency class and the > operator.
6. Run a Demo: Run a demo using the web UI (agency.copilot_demo()), terminal (agency.terminal_demo()), or programmatically.

Who is Agency Swarm for?

Agency Swarm is designed for developers, AI engineers, and researchers who want to build complex, multi-agent AI applications. It's particularly useful for:

• Automating complex tasks that require collaboration between multiple AI agents.
• Building AI-powered workflows that mimic real-world organizational structures.
• Creating production-ready AI applications that can be easily deployed and scaled.

Why choose Agency Swarm?

Agency Swarm offers several advantages over traditional AI development approaches:

• Simplified Multi-Agent Orchestration: Provides a structured framework for managing and orchestrating multiple AI agents.
• Enhanced Collaboration: Enables true agent collaboration with user-defined communication flows.
• Production-Ready: Built for reliability and designed for easy deployment in real-world environments.
• Extensible: Leverages and extends the OpenAI Agents SDK, allowing developers to customize and extend the framework to meet their specific needs.

Best way to build multi-agent AI applications?

Agency Swarm provides a robust and flexible framework for building multi-agent AI applications. By defining agent roles, communication flows, and custom tools, developers can create AI-powered workflows that can automate complex tasks and solve real-world problems. The framework's focus on production readiness and extensibility makes it an ideal choice for building scalable and reliable AI applications.

Compatibility:

• Python: 3.12+
• Model backends:
  • OpenAI (native): GPT-5 family, GPT-4o, etc.
  • Via LiteLLM (router): Anthropic (Claude), Google (Gemini), Azure OpenAI, OpenRouter (gateway), etc.
• OS: macOS, Linux, Windows

Conclusion:

Agency Swarm simplifies multi-agent AI application development by providing a structured framework for managing agent roles, communication flows, and tools. Its production-ready design and extensibility make it a powerful tool for building scalable and reliable AI solutions. If you're looking to build complex AI applications that require collaboration between multiple agents, Agency Swarm is an excellent choice.