ag2-imports

AG2 Imports

The package name is ag2, not autogen. All imports use the ag2 namespace.

Core Agents

from ag2.agentchat import AssistantAgent, UserProxyAgent, ConversableAgent

• AssistantAgent -- LLM-powered agent that generates replies.
• UserProxyAgent -- Proxy for human input; can also execute code.
• ConversableAgent -- Base class for all conversable agents.

LLM Configuration

from ag2 import LLMConfig

Use LLMConfig as a context manager to bind a model configuration to agents created within its scope:

with LLMConfig(api_type="openai", model="gpt-4o"):
    agent = AssistantAgent(name="assistant")

Group Chat

from ag2.agentchat.group import run_group_chat
from ag2.agentchat.group import DefaultPattern, AutoPattern, RoundRobinPattern

• run_group_chat -- Orchestrates a multi-agent conversation.
• DefaultPattern -- LLM-based speaker selection with optional handoffs.
• AutoPattern -- Automatic orchestration; the LLM decides flow.
• RoundRobinPattern -- Agents speak in fixed round-robin order.

Tools

from ag2.tools import tool

The @tool decorator converts a plain function into a tool that agents can call.

Nested Chats and Advanced Patterns

from ag2.agentchat import initiate_chats          # Sequential multi-conversation
from ag2.agentchat import register_hand_off        # Handoff registration
from ag2.agentchat.group import Handoff            # Explicit handoff target

Common Mistakes

• Do NOT use import autogen -- the package was renamed to ag2.
• Do NOT import GroupChat or GroupChatManager -- use run_group_chat with a pattern instead.
• Do NOT use config_list dicts -- use LLMConfig objects.