How to beginner · 3 min read

How to use AgentOps with AutoGen

Q: How to use AgentOps with AutoGen

Use agentops.init() to automatically instrument your AI agents built with AutoGen. Wrap your agent calls with agentops.start_session() and agentops.end_session() to track sessions and capture detailed telemetry with minimal code changes.

Quick answer

Use agentops.init() to automatically instrument your AI agents built with AutoGen. Wrap your agent calls with agentops.start_session() and agentops.end_session() to track sessions and capture detailed telemetry with minimal code changes.

PREREQUISITES

Python 3.8+
AgentOps API key (set in AGENTOPS_API_KEY environment variable)
pip install agentops autogen openai>=1.0
OpenAI API key (set in OPENAI_API_KEY environment variable)

Setup

Install the required packages and set environment variables for AgentOps and OpenAI API keys.

Install packages: pip install agentops autogen openai
Set environment variables in your shell or .env file:
export AGENTOPS_API_KEY=<your_agentops_key>
export OPENAI_API_KEY=<your_openai_key>

bash

pip install agentops autogen openai

output

Collecting agentops
Collecting autogen
Collecting openai
Successfully installed agentops autogen openai

Step by step

This example shows how to initialize AgentOps for automatic tracing and integrate it with an AutoGen agent using the OpenAI gpt-4o model. It demonstrates starting and ending a session to capture observability data.

python

import os
import agentops
from autogen import AssistantAgent, UserProxy
from openai import OpenAI

# Initialize AgentOps for automatic instrumentation
agentops.init(api_key=os.environ["AGENTOPS_API_KEY"])

# Create OpenAI client
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Define a simple AutoGen assistant agent
assistant = AssistantAgent(
    name="assistant",
    llm=client,
    model="gpt-4o"
)

# Define a user proxy to interact with the assistant
user = UserProxy(name="user")

# Start an AgentOps session to track this interaction
session = agentops.start_session(tags=["autogen-example"])

# Run a simple conversation
response = assistant.chat([user.message("Hello, how can you help me today?")])
print("Assistant reply:", response[0].content)

# End the session with success status
agentops.end_session("Success")

output

Assistant reply: Hello! I can help you with a variety of tasks such as answering questions, generating text, or assisting with coding. What would you like to do today?

Common variations

You can use AgentOps with asynchronous AutoGen agents by awaiting calls and using async session management. Also, you can customize tags and metadata in start_session() for richer observability.

Example async usage:

python

import asyncio
import agentops
from autogen import AssistantAgent, UserProxy
from openai import OpenAI

async def main():
    agentops.init(api_key=os.environ["AGENTOPS_API_KEY"])
    client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
    assistant = AssistantAgent(name="assistant", llm=client, model="gpt-4o")
    user = UserProxy(name="user")

    session = agentops.start_session(tags=["autogen-async"])

    response = await assistant.chat_async([user.message("Async hello")])
    print("Assistant reply:", response[0].content)

    agentops.end_session("Success")

asyncio.run(main())

output

Assistant reply: Hello! How can I assist you asynchronously today?

Troubleshooting

If you see no telemetry in AgentOps dashboard, verify AGENTOPS_API_KEY is set correctly and agentops.init() is called before agent usage.
Ensure your AutoGen agent uses the OpenAI client initialized with OPENAI_API_KEY.
If sessions do not end properly, always call agentops.end_session() to flush data.

✅

Key Takeaways

Initialize AgentOps early with agentops.init() to enable automatic instrumentation.
Wrap AutoGen agent interactions with agentops.start_session() and agentops.end_session() for detailed observability.
Use environment variables for all API keys to keep credentials secure and configurable.
AgentOps supports both synchronous and asynchronous AutoGen workflows with minimal code changes.

Verified 2026-04 · gpt-4o

Verify ↗