Microsoft Agent Framework Integration Guide

This document describes how to integrate and use the Microsoft Agent Framework within AgentScope Runtime to build agents that support multi-turn conversations, conversation memory, and streaming responses.

📦 Example Overview

The following example demonstrates how to use the Microsoft Agent Framework inside AgentScope Runtime:

  • Uses the Qwen-Plus model from DashScope

  • Supports multi-turn conversation and session memory

  • Employs streaming output (SSE) to return responses in real-time

  • Implements session history storage via an in-memory database (InMemoryDb)

  • Can be accessed through an OpenAI-compatible API mode

Here’s the core code:

# ms_agent.py
# -*- coding: utf-8 -*-
import os
from contextlib import asynccontextmanager

from fastapi import FastAPI
from agent_framework.openai import OpenAIChatClient

from agentscope_runtime.engine import AgentApp
from agentscope_runtime.engine.schemas.agent_schemas import AgentRequest

PORT = 8090


def run_app():
    """Start AgentApp and enable streaming output."""
    @asynccontextmanager
    async def lifespan(app: FastAPI):
        app.state.thread_storage = {}  # Only for testing
        yield

    agent_app = AgentApp(
        app_name="Friday",
        app_description="A helpful assistant",
        lifespan=lifespan,
    )

    @agent_app.query(framework="ms_agent_framework")
    async def query_func(
        self,
        msgs,
        request: AgentRequest = None,
        **kwargs,
    ):
        """Handle agent queries."""
        session_id = request.session_id
        user_id = request.user_id

        # Export historical context
        id_key = f"{user_id}_{session_id}"
        thread = agent_app.state.thread_storage.get(id_key)

        # Create agent
        agent = OpenAIChatClient(
            model_id="qwen-plus",
            api_key=os.environ["DASHSCOPE_API_KEY"],
            base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
        ).create_agent(
            instructions="You're a helpful assistant named Friday",
            name="Friday",
        )

        # Restore or create conversation thread
        if thread:
            thread = await agent.deserialize_thread(thread)
        else:
            thread = agent.get_new_thread()

        # Streaming responses
        async for event in agent.run_stream(
            msgs,
            thread=thread,
        ):
            yield event

        # Save session state
        serialized_thread = await thread.serialize()
        agent_app.state.thread_storage[id_key] = serialized_thread

    agent_app.run(host="127.0.0.1", port=PORT)


if __name__ == "__main__":
    run_app()

⚙️ Prerequisites

Note

Before starting, make sure you have installed AgentScope Runtime and Microsoft Agent Framework, and configured the required API keys.

  1. Install dependencies:

    pip install "agentscope-runtime[ext]"

  2. Set environment variables (DashScope provides the Qwen model API Key):

    export DASHSCOPE_API_KEY="your-dashscope-api-key"

▶️ Run the Example

To run the example:

python ms_agent.py

🌐 API Interaction

1. Ask the Agent (/process)

You can send HTTP POST requests to interact with the agent, with support for SSE streaming responses:

curl -N \
  -X POST "http://localhost:8090/process" \
  -H "Content-Type: application/json" \
  -d '{
    "input": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "What is the capital of France?" }
        ]
      }
    ],
    "session_id": "session_1"
  }'

2. OpenAI-Compatible Mode

This example also supports the OpenAI Compatible API:

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8090/compatible-mode/v1")
resp = client.responses.create(
    model="any_model",
    input="Who are you?",
)
print(resp.response["output"][0]["content"][0]["text"])

🔧 Customization

You can extend the example in the following ways:

  1. Change the model — Replace the model_id in OpenAIChatClient with another model, or use a client from another model provider.

  2. Add system prompts — Modify the instructions field to create a different persona for the agent.

📚 相关文档