封装本地大模型

github:

https://github.com/auliwenjiang/agentcp/tree/master/samples/local_llm_agent

README.md

使用指南

1、环境要求

• Python 3.8+

• AgentCP SDK

• Ollama

2、环境准备

安装Ollama并下载模型:

Windows

macOS

Linux

curl -fsSL https://ollama.com/install.sh | sh

3、下载模型

ollama pull qwen3:0.6b

4、安装依赖

安装必要的Python库:

pip install agentcp requests

5、创建身份ID

请参考创建身份，读写公有私有数据文档

6、修改main.py文件

1. 将seed_password、agent_id修改为上一步创建的身份信息

2. 将model_name修改为你本地使用的model

7、执行main.py代码

python main.py

功能简介

该Agent基于agentcp库构建，实现与本地大模型交互，主要功能包括：

• 接收并处理用户的消息请求

• 转发消息到本地大模型

• 处理本地大模型的响应并返回给原始请求方

完整示例代码

import json
import requests
import agentcp
model_name = 'your_local_model'
def chatWithLLM(prompt):
    try:
        response = requests.post(
            "http://localhost:11434/api/chat",
            json={
                "model": model_name,
                "messages": [
                    {
                        "role": "user", 
                        "content": prompt
                    }],
                "stream": False
            })
        response.encoding = 'utf-8'
        response.raise_for_status()
        result = response.json()
        result = result.get('message')
        result = result.get('content')
        return result
    except requests.exceptions.RequestException as e:
        return "请求失败"
    except json.JSONDecodeError:
        return "响应解析失败"

if __name__ == '__main__':
    agent_id = 'your_agent_id'
    acp = agentcp.AgentCP('.', seed_password='')
    aid = acp.load_aid(agent_id)
    @aid.message_handler()
    async def sync_message_handler(msg):
        content = aid.get_content_from_message(msg)
        session_id = aid.get_session_id_from_message(msg)
        sender = aid.get_sender_from_message(msg)
        res = chatWithLLM(content)
        aid.send_message_content(session_id, [sender], res)
        return True
    aid.online()
    acp.serve_forever()