Ag2
AG2 (formerly AutoGen): The Open-Source AgentOS.Join us at: https://discord.gg/sNGSwQME3x
Last verified:
What is Ag2?
AG2 (formerly AutoGen) is an open-source programming framework for building AI agents and enabling multi-agent collaboration to solve tasks. It serves as an AI Agent Operating System (AgentOS) that allows developers to build, orchestrate, and evolve systems of AI agents as an AI workforce. The framework is built by the creators of AutoGen and is backed by award-winning research, used by millions of developers with over 50k+ GitHub stars and 20k+ developers in the community.
Key features include multi-agent conversation framework as a high-level abstraction for building LLM workflows, universal framework interoperability connecting agents from AG2, Google ADK, OpenAI, and LangChain into one team, cross-platform coordination for assembling dynamic teams of specialized personas, unified state management maintaining a shared brain across task lifecycles, and standardized protocols (A2A and MCPs) with enterprise security built-in. The platform supports enhanced LLM inference APIs for improved performance and reduced cost, Human in the Loop capabilities for human oversight at critical decision points, tools for connecting with external APIs and services, structured outputs using Pydantic models for consistent responses, and advanced patterns like RAG and code execution.
AG2 is designed for developers, data scientists, and enterprise teams building production-ready multi-agent AI systems. It targets organizations wanting to deploy AI-native operations, research institutions conducting agentic AI research, and companies seeking to achieve up to 70% faster workflows and 5x productivity gains. The framework supports various LLM providers including OpenAI, Google Gemini, and Azure OpenAI, and requires Python version 3.10 to 3.13.
Ag2 pricing
Pricing model: Freemium
AG2 is completely free as open-source software governed by Apache 2.0 license with no fees. The framework itself is free to use, install via pip install ag2, and modify. Users only pay for their own LLM API usage costs from providers like OpenAI, Google Gemini, or Azure OpenAI. The pricing model is usage-based where costs depend on AI model usage and API calls from the LLM provider, not from AG2 itself. Enterprise-ready features are included at no additional cost in the open-source framework.
Ag2 pros
- Open-source framework with Apache 2.0 license
- 50k+ GitHub stars and 20k+ developer community
- Multi-agent conversation framework with high-level abstraction
- Universal interoperability connects AG2, Google ADK, OpenAI, LangChain agents
- Cross-platform coordination for dynamic specialist teams
- Unified state management maintains shared brain across workflows
- Human in the Loop for human oversight at critical decisions
- Enhanced LLM inference APIs improve performance and reduce cost
- Structured outputs with Pydantic models ensure consistent responses
- Built-in tools for external API and service integration
- Supports multiple LLM providers (OpenAI, Gemini, Azure)
- Advanced patterns include RAG, code execution, nested workflows
- Enterprise-ready with production-grade security built-in
- 3-minute quickstart with pip install ag2
- Up to 70% faster workflows and 5x productivity gain proven
- Full decision auditability for compliance
- GroupChat and Swarm orchestration patterns built-in
- Interactive Jupyter Notebook tutorials for hands-on learning
Ag2 cons
- Requires Python 3.10-3.13 (version compatibility constraints)
- Users must provide and manage their own LLM API keys
- Learning curve for multi-agent orchestration concepts
- Minimal dependencies installed by default requires extra installs
- Docker may be needed for code execution in production
- Configuration files must be manually created for API keys
- Requires API key setup before running first agent
- Advanced features like RAG require additional configuration
Frequently asked questions about Ag2
How do I install AG2?
Install AG2 using pip with the command 'pip install ag2[openai]' for Windows/Linux or 'pip install ag2[openai]' for Mac. The framework requires Python version 3.10 to 3.13. Minimal dependencies are installed by default, and you can install extra options based on features needed like 'pip install ag2[gemini]' for Google Gemini support.
Is AG2 compatible with AutoGen code?
Yes, AG2 evolved from AutoGen and maintains compatibility. AG2 was formerly AutoGen, and the project maintains backward compatibility while evolving the framework. Existing AutoGen code can generally work with AG2 with minimal changes.
What LLM providers does AG2 support?
AG2 supports multiple LLM providers including OpenAI (gpt-5, gpt-5-nano), Google Gemini, and Azure OpenAI. You configure the LLM provider through LLMConfig specifying the api_type, model, and API key. Each provider requires setting the appropriate environment variable like OPENAI_API_KEY, GEMINI_API_KEY, or AZURE_OPENAI_API_KEY.
What agent communication protocols does AG2 use?
AG2 uses standardized protocols including A2A (Agent-to-Agent) and MCPs (Model Context Protocols) with enterprise security built-in. These protocols enable agents from different frameworks to communicate and work together in a unified runtime.
What is ConversableAgent?
ConversableAgent is the core building block of AG2 - a smart, interactive agent that uses your configured LLM to process information and interact with other agents or humans. It can send messages, receive messages, generate responses using GenAI models, external tools, or human inputs, make decisions based on its defined purpose, and execute tools and functions when needed.
How does Human in the Loop work in AG2?
Human in the Loop (HITL) brings human oversight into agent workflows, allowing agents to request human input for crucial decision points. You can configure human_input_mode as ALWAYS (requires human input for every response), NEVER (operates autonomously), or TERMINATE (only requests human input to end conversations). The UserProxyAgent class automatically sets human_input_mode to ALWAYS and supports code execution.
What orchestration patterns does AG2 offer?
AG2 offers multiple orchestration patterns including two-agent conversations, sequential conversations that chain multiple dialogues, group collaborations with many agents using GroupChat, nested workflows, and Swarm patterns. You can also create custom orchestration by registering specialized reply methods. The GroupChatManager uses an LLM to select the next agent to respond based on context.
How do tools work in AG2?
Tools extend an agent's capabilities beyond text conversations by connecting with external APIs and services, performing calculations and data processing, and accessing files, databases, or other systems. You register tools using register_function which makes them available to agents. The LLM then decides when to invoke the tool based on the task requirements.
What is the licensing for AG2?
AG2 uses the Apache License Version 2.0 (Apache-2.0) from v0.3, which enhances commitment to open-source collaboration while providing protections for contributors and users. The original AutoGen code from Microsoft is under MIT License, while modifications and additions in AG2 are under Apache-2.0.
What are the proven results from using AG2?
Customer results show up to 70% faster workflows, up to 5x productivity gain, and full decision auditability. The framework is trusted by enterprise teams and leading research institutions, with 50k+ GitHub stars and 20k+ developers using it. AG2 is used by millions of developers and backed by award-winning research including AutoGen and StateFlow.