AutoGen

Track every AutoGen team run with latency, tool call data, and failure details. AgentMetricsRunStream wraps AutoGen's run_stream(). Your event handling code stays the same.

Requirements

• Python 3.10 or later
• autogen-agentchat 0.4 or later

1. Install

pip install agentmetrics-autogen

2. Wrap run_stream

Replace team.run_stream(...) with tracker.run(team, ...). The async context manager and event loop stay the same.

import os
import asyncio
from agentmetrics_autogen import AgentMetricsRunStream

tracker = AgentMetricsRunStream(
    api_key=os.environ["AGENTMETRICS_API_KEY"],
    agent_id="my-autogen-team",
    base_url="http://localhost:8099",
)

async def main():
    async with tracker.run(team, task="summarize this document") as stream:
        async for event in stream:
            print(event)

asyncio.run(main())

3. Flush before exit (optional)

For short-lived scripts, call flush() to wait for all in-flight requests to complete before the process exits.

tracker.flush()

What gets tracked

AgentMetricsRunStream intercepts AutoGen's event stream and extracts:

A run summary is emitted when the async with block exits: on normal completion, on exception, or on cancellation.

Configuration