My Agent

English | 中文

---

English Version

✨ Features

• 🚀 Simple & Intuitive - Clean API design for rapid development
• 🔧 Function Calling - Full support for OpenAI-standard tool calls with automatic parsing and execution
• 📝 Streaming Output - Real-time streaming responses with tool call progress indicators
• 💾 History Management - Automatic conversation history with JSON save/load
• 🎯 JSON Mode - Enforce JSON output for structured data extraction
• 🔌 Flexible Configuration - Global defaults and per-instance settings
• 📦 Lightweight - Only depends on openai

📦 Installation

Follow these steps to install my_agent:

1. Clone the repository

   git clone https://github.com/Leabots/my_agent.git
   

2. Navigate into the project directory

   cd my_agent
   

3. Install in editable mode The -e flag installs the package in "editable" mode. This creates a symlink to your source code, so any changes you make to the source files are immediately reflected in your Python environment without needing to reinstall.

   pip install -e .
   

This command will also automatically install the library's only dependency, openai.

🚀 Quick Start

Basic Conversation

from my_agent import Agent

agent = Agent(
    base_url="https://api.openai.com/v1",
    model="gpt-3.5-turbo",
    api_key="your-api-key"
)

response = agent.complete("Hello, please introduce yourself")
print(response["content"])

Global Default Configuration

Agent.set_default(
    base_url="https://api.deepseek.com",
    model="deepseek-chat",
    api_key="your-api-key"
)

agent = Agent()  # uses the defaults
response = agent.complete("Hello!")

Streaming Output

for chunk in agent.stream_complete("Tell me a short story"):
    if chunk.get("delta"):
        print(chunk["delta"], end="", flush=True)

JSON Output

response = agent.json_output("List 3 fruits, return JSON format")
print(response["content"])  # Already a Python dict

🔧 Function Calling

Register a Tool (auto-extract from docstring)

def get_weather(city: str, unit: str = "celsius"):
    """Get current weather for a city.
    
    :param city: City name, e.g., Beijing, Shanghai
    :param unit: Temperature unit, celsius or fahrenheit
    :return: Dictionary with weather information
    """
    return {"city": city, "temperature": 24, "unit": unit, "condition": "Sunny"}

agent.register_tool("get_weather", "Get current weather for a city", get_weather)

response = agent.complete("What's the weather like in Beijing today?")
print(response["content"])

Manual Parameter Descriptions

agent.register_tool(
    "get_weather",
    "Get current weather for a city",
    get_weather,
    param_descriptions={
        "city": "City name, e.g., Beijing, Shanghai",
        "unit": "Temperature unit: celsius or fahrenheit"
    },
    return_description="Dictionary containing city, temperature, unit, and condition"
)

Streaming with Tool Call Messages

tool_messages = {
    "get_weather": "[🌤️ Fetching weather information...]",
    "get_location": "[📍 Getting location...]"
}

for chunk in agent.stream_complete("What's the weather like here?", tool_call_messages=tool_messages):
    if chunk.get("delta"):
        print(chunk["delta"], end="", flush=True)

💾 Conversation History

# Save history to file
agent.save_history("my_conversation")

# Load history from file
agent.load_history("my_conversation")

# List all saved histories
print(agent.list_saved_histories())

# Clear current history
agent.clear_history()

📖 API Reference

Agent Initialization

Parameter	Type	Default	Description
base_url	str	None	API base URL
model	str	None	Model name
api_key	str	None	API key
system_prompt	str	"You are a helpful assistant."	System prompt
temperature	float	0.5	Sampling temperature
max_tokens	int	4096	Max tokens to generate

Main Methods

• complete(user_input, record=True, max_iterations=10) - Send message, return full response (auto tool handling)
• stream_complete(user_input, record=True, tool_call_messages=None, max_iterations=5) - Streaming version with tool progress
• json_output(user_input, record=True) - Force JSON output
• register_tool(name, description, handler, param_descriptions=None, return_description=None) - Register a tool
• clear_history(), save_history(), load_history(), list_saved_histories() - History management

Response Format

Both complete() and json_output() return a dict with:

{
    "id": "chatcmpl-xxx",
    "content": "Response text",
    "model": "model-name",
    "created": 1234567890,
    "usage": {...},
    "completion_tokens": 150,
    "prompt_tokens": 50,
    "total_tokens": 200,
    "finish_reason": "stop",
    "system_fingerprint": "..."
}

🧪 Testing

from my_agent import test

# Run basic tests
test.test(base_url, model, api_key)

# Run JSON output test
test.json_output_test(base_url, model, api_key)

📄 License

MIT License - completely free to use, no restrictions.

🤝 Contributing

Issues and Pull Requests are welcome!

---

中文版本

✨ 特性

• 🚀 简单易用 - 简洁的 API 设计，快速上手
• 🔧 工具调用 - 支持 OpenAI 标准的 Function Calling，自动解析和执行工具
• 📝 流式输出 - 支持流式响应，实时显示生成内容，可显示工具调用进度
• 💾 历史管理 - 自动保存对话历史，支持保存/加载到 JSON 文件
• 🎯 JSON 输出 - 强制模型返回 JSON 格式，便于结构化数据提取
• 🔌 灵活配置 - 支持全局默认配置和实例级配置
• 📦 轻量依赖 - 仅依赖 openai

📦 安装

请按照以下步骤安装 my_agent：

1. 克隆仓库

   git clone https://github.com/Leabots/my_agent.git
   

2. 进入项目目录

   cd my_agent
   

3. 执行可编辑安装 -e 标志以“可编辑”模式安装包。这会在你的 Python 环境和项目源码之间创建一个链接，你对源码做的任何修改都会立刻生效，无需重新安装。

   pip install -e .
   

这个命令也会自动安装库的唯一依赖 openai。

🚀 快速开始

基础对话

from my_agent import Agent

agent = Agent(
    base_url="https://api.openai.com/v1",
    model="gpt-3.5-turbo",
    api_key="your-api-key"
)

response = agent.complete("你好，请介绍一下自己")
print(response["content"])

全局默认配置

Agent.set_default(
    base_url="https://api.deepseek.com",
    model="deepseek-chat",
    api_key="your-api-key"
)

agent = Agent()  # 使用默认配置
response = agent.complete("你好！")

流式输出

for chunk in agent.stream_complete("讲一个短故事"):
    if chunk.get("delta"):
        print(chunk["delta"], end="", flush=True)

JSON 输出

response = agent.json_output("列出3种水果，返回JSON格式")
print(response["content"])  # 已经是 Python 字典

🔧 工具调用

注册工具（自动从 docstring 提取描述）

def get_weather(city: str, unit: str = "celsius"):
    """获取指定城市的天气信息
    
    :param city: 城市名称，如北京、上海
    :param unit: 温度单位，celsius 或 fahrenheit
    :return: 天气信息字典
    """
    return {"city": city, "temperature": 24, "unit": unit, "condition": "晴"}

agent.register_tool("get_weather", "获取指定城市的天气信息", get_weather)

response = agent.complete("北京今天天气怎么样？")
print(response["content"])

手动指定参数描述

agent.register_tool(
    "get_weather",
    "获取指定城市的天气信息",
    get_weather,
    param_descriptions={
        "city": "城市名称，例如：北京、上海",
        "unit": "温度单位：celsius（摄氏度）或 fahrenheit（华氏度）"
    },
    return_description="包含城市、温度、单位和天气状况的字典"
)

带工具调用提示的流式输出

tool_messages = {
    "get_weather": "[🌤️ 正在获取天气信息...]",
    "get_location": "[📍 正在获取位置信息...]"
}

for chunk in agent.stream_complete("我这里天气怎么样？", tool_call_messages=tool_messages):
    if chunk.get("delta"):
        print(chunk["delta"], end="", flush=True)

💾 对话历史管理

# 保存历史到文件
agent.save_history("my_conversation")

# 从文件加载历史
agent.load_history("my_conversation")

# 列出所有保存的历史
print(agent.list_saved_histories())

# 清空当前历史
agent.clear_history()

📖 API 参考

Agent 初始化参数

参数	类型	默认值	说明
base_url	str	None	API 基础 URL
model	str	None	模型名称
api_key	str	None	API 密钥
system_prompt	str	"You are a helpful assistant."	系统提示词
temperature	float	0.5	采样温度
max_tokens	int	4096	最大生成 token 数

主要方法

• complete(user_input, record=True, max_iterations=10) - 发送消息并返回完整响应（自动处理工具调用）
• stream_complete(user_input, record=True, tool_call_messages=None, max_iterations=5) - 流式版本，支持工具调用提示
• json_output(user_input, record=True) - 强制返回 JSON 格式
• register_tool(name, description, handler, param_descriptions=None, return_description=None) - 注册工具
• clear_history() / save_history() / load_history() / list_saved_histories() - 历史管理

返回格式

complete() 和 json_output() 返回字典，包含以下字段：

{
    "id": "chatcmpl-xxx",
    "content": "回复内容",
    "model": "模型名称",
    "created": 1234567890,
    "usage": {...},
    "completion_tokens": 150,
    "prompt_tokens": 50,
    "total_tokens": 200,
    "finish_reason": "stop",
    "system_fingerprint": "..."
}

🧪 测试

from my_agent import test

# 运行基础测试
test.test(base_url, model, api_key)

# 运行 JSON 输出测试
test.json_output_test(base_url, model, api_key)

📄 许可证

MIT License - 完全自由使用，无任何限制。

🤝 贡献

欢迎提交 Issue 和 Pull Request！

---

Happy Coding! 🎉