Quickstart
Run the agent loop in the browser, expose your tools via WebMCP
Last verified:
What is Quickstart?
AG2 is an open-source AgentOS for building AI agents and multi-agent systems. It is designed to help developers create production-ready workflows in minutes by combining LLM-powered agents, orchestration patterns, tools, and human-in-the-loop control.
The framework centers on multi-agent collaboration. It provides a high-level conversation framework for building agent workflows such as group chats, swarms, nested conversations, and sequential processes, letting specialized agents coordinate on complex tasks.
AG2 also supports practical agent features that make it useful beyond simple chatbots. It includes tool registration for external functions and APIs, structured output, retrieval-augmented generation, and code execution, so agents can work with real data and perform task-specific operations.
It is aimed at developers, researchers, and teams building agentic AI applications. The quickstart emphasizes fast setup, Python-based development, and examples that range from simple single-agent use to more advanced production-style orchestration and human oversight.
Quickstart pricing
Pricing model: Freemium
The quickstart page does not list product pricing. It presents AG2 as open source and shows installation with pip install "ag2[openai]", with examples that rely on model-provider API keys such as OpenAI's key. The website content visible here does not describe free tiers, paid plans, or bundled subscription features.
Quickstart pros
- Open-source framework
- AgentOS positioning
- Fast 3-minute quickstart
- Simple pip installation
- Python-first API
- ConversableAgent core abstraction
- Supports multi-agent workflows
- Built-in GroupChat orchestration
- Built-in Swarm pattern
- Human-in-the-loop support
- Tool registration support
- External API integration
- Structured output support
- RAG support
- Code execution support
- Works with multiple LLM providers
- Supports autonomous workflows
- Supports human-guided workflows
- Helpful for production-ready systems
- Includes real-world example repository
Quickstart cons
- Requires Python setup
- Needs API keys for hosted models
- Quickstart example is OpenAI-based
- Some features need advanced configuration
- Multi-agent design adds complexity
- Tooling requires custom function wiring
- Human-in-the-loop can slow automation
- Code execution may need safety controls
- Documentation spans multiple sections
- Not a no-code product
Frequently asked questions about Quickstart
What is AG2?
AG2 is an open-source AgentOS for AI agents. It provides the building blocks for creating agent workflows, especially multi-agent systems that can communicate, coordinate, and use tools to solve tasks.
Who is AG2 for?
AG2 is aimed at developers and researchers who want to build agentic AI applications. The quickstart and examples are geared toward people who are comfortable working in Python and connecting models, tools, and workflows.
What is the main agent type in AG2?
The core building block is the ConversableAgent. It can send and receive messages, generate responses with LLMs, use tools, and participate in agent-to-agent conversations.
How does AG2 support multiple agents?
AG2 supports multi-agent collaboration through patterns like GroupChat, Swarm, nested conversations, and sequential workflows. It also lets you create custom orchestration by registering specialized reply methods.
Does AG2 support human input?
Yes. AG2 includes human-in-the-loop support through human_input_mode settings such as ALWAYS, NEVER, and TERMINATE. It also provides UserProxyAgent for workflows where a human participates directly.
Can AG2 use external tools?
Yes. AG2 lets you register external functions so agents can call Python code, APIs, or other services. The quickstart shows a date-calculation example where one agent suggests a function and another executes it.
What advanced patterns does AG2 support?
The quickstart lists structured output, retrieval-augmented generation, and code execution as advanced agentic design patterns. These features help AG2 handle more structured, data-driven, and operational workflows.
How quickly can I get started?
The quickstart says you can get started in about 3 minutes. It shows a simple installation command and a short Python example that creates and runs an agent.
What LLM providers work with AG2?
The documentation shown mentions support for multiple providers, including OpenAI, Anthropic, Google Gemini, and Ollama for structured output, and the quickstart example itself uses OpenAI configuration.
What kinds of example use cases does AG2 provide?
The site points to use cases such as customer service, travel planning, and game design. It also highlights interactive notebooks, real-world application examples, and API reference material for deeper learning.