Skip to main content
Automatically capture agent loops, message histories, and tool executions within the AutoGen framework. Note: This integration supports ag2 (v0.11.5+), the community-maintained continuation of the classic AutoGen framework.

​
Setup

To capture the complete picture—including both the agent orchestrations and the underlying token costs—we highly recommend initializing both the AutoGen integration and your specific LLM provider (e.g., Google GenAI, OpenAI, Anthropic). 
import traceroot
from traceroot import Integration

# Initialize AutoGen alongside your LLM provider to capture tokens and costs
traceroot.initialize(integrations=[
    Integration.AUTOGEN,
    Integration.GOOGLE_GENAI  # Or OPENAI, ANTHROPIC, etc.
])

​
Usage

Once initialized, the standard initiate_chat loop and internal agent reasoning steps are captured automatically: 
import os
import autogen

llm_config = {
    "config_list": [{
        "model": "gemini-2.5-flash", 
        "api_key": os.environ["GEMINI_API_KEY"],
        "api_type": "google"
    }]
}

assistant = autogen.AssistantAgent(
    name="assistant", 
    llm_config=llm_config
)
user_proxy = autogen.UserProxyAgent(
    name="user_proxy", 
    human_input_mode="NEVER", 
    max_consecutive_auto_reply=2
)

# The entire conversation hierarchy is automatically traced
user_proxy.initiate_chat(
    assistant,
    message="Explain the difference between Python lists and tuples.",
)

​
Usage with Tools

Function registrations and local tool executions are automatically traced as nested child spans within the executing agent’s workflow: 
import os
import autogen
from typing import Annotated
from autogen import register_function

# Define the LLM and Agents
llm_config = {
    "config_list": [{
        "model": "gemini-2.5-flash", 
        "api_key": os.environ["GEMINI_API_KEY"],
        "api_type": "google"
    }]
}

assistant = autogen.AssistantAgent(
    name="assistant", 
    llm_config=llm_config
)
user_proxy = autogen.UserProxyAgent(
    name="user_proxy", 
    human_input_mode="NEVER", 
    max_consecutive_auto_reply=2
)

def get_weather(city: Annotated[str, "City name"]) -> str:
    """Get current weather for a city."""
    return f"The weather in {city} is 72 degrees and sunny."

# Register the tool
register_function(
    get_weather,
    caller=assistant,     # The agent that decides to use the tool
    executor=user_proxy,  # The agent that actually runs the Python function
    name="get_weather",
    description="Get current weather for a city",
)

# Tool executions appear as 'tool' spans under the UserProxyAgent
user_proxy.initiate_chat(
    assistant,
    message="What is the weather in Tokyo?",
    max_turns=5,  # Add a guard to ensure an LLM response cap
)

​
What Gets Captured

AttributeDescription
Conversation LoopThe overarching initiate_chat session
Agent StepsIndividual spans for each AssistantAgent or UserProxyAgent turn
MessagesFull chat history, input messages, and agent replies
Tool callsFunction names, input arguments, and execution outputs
LLM calls*Raw completion requests to the provider
Tokens & Cost*Aggregated usage and pricing for the chat session
**Requires initializing the corresponding LLM integration (e.g., Integration.GOOGLE_GENAI) alongside Integration.AUTOGEN