Microsoft AutoGen

AgentV provides native support for AutoGen agent workflows, allowing you to measure the performance of complex multi-agent conversations.

1. Setup Your Agent API

Expose your AutoGen agent via a standard HTTP endpoint.

import ag2
from fastapi import FastAPI

# ... setup your ag2 config ...

app = FastAPI()


@app.post("/execute_task")
async def execute(request: dict):
    # assistant = ag2.AssistantAgent(...)
    # user_proxy = ag2.UserProxyAgent(...)
    # user_proxy.initiate_chat(assistant, message=request["task_description"])
    # result = user_proxy.last_message()["content"]
    return {"action": "final_answer", "summary": result}

2. Run Evaluation

Use the ag2 protocol to connect the harness to your agent service.

agentv run \
  --path scenarios/loan_scenario.json \
  --protocol ag2 \
  --agent ag2://localhost:8000/execute_task \
  --agent-name "AG2-Finance-Expert"

3. Review DNA Trajectories

Launch the console to see the internal transitions between your AutoGen agents visualized in a high-fidelity trajectory map.

agentv console