AutoGen

Install:

shell

pip install "nullrun[autogen]"

autogen_assistant.py

import nullrun
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient

nullrun.init(api_key="nr_live_...")

model = OpenAIChatCompletionClient(model="gpt-4o-mini")
assistant = AssistantAgent("assistant", model_client=model)
result = await assistant.run(task="Hello")

Patched via nullrun.instrumentation.autogen.patch_autogen, which hooks the AutoGen model-client path so every agent run emits a track_llm event. The vendor imports (autogen-agentchat, autogen-ext[openai]) are wrapped in try/except ImportError, so installing only this extra group does not crash on init().

See also

• Auto-instrumentation overview
• Quickstart