AutoGen Review: Maintenance-Mode Status & AG2 Fork

Who Built AutoGen, and What Happened to It?

What Is AutoGen?

• AgentChat: the high-level API for defining agents with roles (assistant, user proxy, critic) and running multi-agent conversations or group chats
• Core: the lower-level, event-driven runtime AgentChat is built on, for teams that need custom orchestration logic
• Code-execution agents: a signature AutoGen pattern — an agent can write Python code, hand it to a code-executor agent (commonly sandboxed in Docker), and use the output to continue the conversation
• Group chat orchestration: multiple agents participate in a shared conversation with a manager agent deciding turn order, useful for tasks needing several specialist roles (e.g., planner, coder, reviewer)
• Extensions: first- and third-party integrations for model providers, tools, and memory beyond the Core/AgentChat layers

from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import TextMentionTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient

model_client = OpenAIChatCompletionClient(model="gpt-4o-mini")
assistant = AssistantAgent("assistant", model_client=model_client)
critic = AssistantAgent(
    "critic",
    model_client=model_client,
    system_message="Review the answer and reply APPROVE if correct.",
)

team = RoundRobinGroupChat(
    [assistant, critic],
    termination_condition=TextMentionTermination("APPROVE"),
)

await team.run(task="Summarize what AutoGen does in one sentence.")

How Much Does AutoGen Cost?

• AutoGen (the framework): free forever under the MIT license (code) / CC BY 4.0 (documentation), for any use including commercial products
• No AutoGen-branded paid tier, hosted service, or enterprise SKU exists as of this review — Microsoft's commercial agent offering going forward is the Microsoft Agent Framework, which is itself open source, with paid cost coming from whichever Azure services you choose to attach (e.g., Azure AI Foundry hosting)
• AG2, the community fork, is likewise free under the Apache 2.0 license, with commercial support offered separately by AG2AI rather than bundled into the framework
• Model API usage is billed by the model provider you connect (OpenAI, Anthropic, Azure OpenAI, or your own self-hosted model), not by AutoGen

How Do You Install and Get Started With AutoGen?

1. 1
   Install the AgentChat package and a model extension: `pip install -U "autogen-agentchat" "autogen-ext[openai]"` (requires Python 3.10+).
2. 2
   Set your model provider's API key as an environment variable (for example OPENAI_API_KEY) — AutoGen does not provide or proxy model access itself.
3. 3
   Define one or more AssistantAgent instances with a model client, and combine them into a team (e.g., RoundRobinGroupChat) with a termination condition.
4. 4
   For code-execution workflows, add a CodeExecutorAgent backed by a sandboxed executor (Docker is the documented default) so generated code runs in isolation, not on the host machine.
5. 5
   Read the official AutoGen documentation for the current API reference, and review the migration guide if starting a new project, since Microsoft now points new users toward the Microsoft Agent Framework.

Who Should Use AutoGen?

When Should You NOT Use AutoGen?

• A new project with no existing AutoGen code — avoid legacy AutoGen and start with the Microsoft Agent Framework (Microsoft's actively supported successor) or AG2 (the actively developed community fork) instead
• A team that needs Microsoft enterprise support commitments or long-term API stability guarantees on a Microsoft-maintained product — Microsoft Agent Framework carries those, AutoGen does not
• A team that wants fast-moving open-source feature development — that happens at AG2 now, led by AutoGen's original authors, not at microsoft/autogen
• A project that wants a structured, role-based agent abstraction out of the box rather than assembling conversation patterns manually — CrewAI or LangGraph fit that better
• Use AutoGen only when you are maintaining an existing AutoGen application and a migration has not yet been scoped

AutoGen vs. Alternatives

Compare in directory→

Common Mistakes When Evaluating AutoGen

Frequently Asked Questions

Sources

Related Reading

• LangChain Review: Features, Pricing, Alternatives — the general-purpose framework most often compared against AutoGen, including its LangGraph agent runtime.
• Semantic Kernel Review: Features, Pricing, Alternatives — the other Microsoft SDK that Microsoft Agent Framework converges with AutoGen, with native .NET/Python/Java support.
• Letta Review: Features, Pricing, Alternatives — a framework built specifically for single-agent long-term memory rather than multi-agent conversation, useful when AutoGen's conversational agents need to persist state across sessions.
• Local LLM Software Directory — the full catalog of local AI tools this article belongs to, including where multi-agent frameworks fit.
• MetaGPT vs Agnai vs RisuAI: What Each AI Agent Tool Is Actually For — clears up the difference between MetaGPT (a role-based multi-agent framework in the same category as AutoGen) and two unrelated AI roleplay chat tools it often gets confused with.