Orchestrate Microsoft AutoGen multi-agent workflows with nRouter. Connect agents via OpenAI-compatible endpoints with intelligent routing and budget controls.

Microsoft's AutoGen framework supports any OpenAI-compatible endpoint via its OpenAIChatCompletionClient. Point it at nRouter — guardrails, caching, and rate-limits auto-apply to every agent turn.

Installation

pip install autogen-agentchat autogen-ext[openai]

Setup

import os
from autogen_ext.models.openai import OpenAIChatCompletionClient

model_client = OpenAIChatCompletionClient(
    model="claude-sonnet-4-5-20250929",
    api_key=os.environ["NROUTER_API_KEY"],
    base_url="https://api.nrouter.ai/v1",
)

Single Agent

import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.messages import TextMessage
from autogen_core import CancellationToken

async def main():
    agent = AssistantAgent(
        name="assistant",
        model_client=model_client,
        system_message="You are a concise, technical assistant.",
    )

    response = await agent.on_messages(
        [TextMessage(content="What is an LLM gateway?", source="user")],
        cancellation_token=CancellationToken(),
    )
    print(response.chat_message.content)

asyncio.run(main())

Multi-Agent Conversation

from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import MaxMessageTermination

researcher = AssistantAgent(
    name="researcher",
    model_client=model_client,
    system_message="You research topics and surface the key facts.",
)

critic = AssistantAgent(
    name="critic",
    model_client=model_client,
    system_message="You critique research for accuracy and depth.",
)

team = RoundRobinGroupChat(
    [researcher, critic],
    termination_condition=MaxMessageTermination(max_messages=6),
)

result = await team.run(task="Evaluate the state of LLM observability in 2026.")
for msg in result.messages:
    print(f"{msg.source}: {msg.content}")

Per-Request Overrides

Pass nrouter_* fields through the client's extra_create_args:

model_client = OpenAIChatCompletionClient(
    model="gpt-5.5",
    api_key=os.environ["NROUTER_API_KEY"],
    base_url="https://api.nrouter.ai/v1",
    extra_create_args={
        "nrouter_prompt_template_id": "your-summarizer-id",
        "nrouter_prompt_variables": {"language": "Spanish"},
        "nrouter_cache": False,
    },
)

Guardrails are not passed here. You assign them in the dashboard at key, team, or organization scope — the narrowest scope that mentions a guardrail wins — and they run automatically on every agent turn that scope covers.

Tools

AutoGen tool wiring — guardrails check the user input before tools run:

async def get_weather(city: str) -> str:
    return f"72°F and sunny in {city}"

agent = AssistantAgent(
    name="weather_assistant",
    model_client=model_client,
    tools=[get_weather],
)

Next Steps

CrewAI — Alternative multi-agent framework
Python SDK — Without AutoGen

AutoGen