diff --git a/README_TOOLS_EN.md b/README_TOOLS_EN.md new file mode 100644 index 0000000..cf6506f --- /dev/null +++ b/README_TOOLS_EN.md @@ -0,0 +1,81 @@ +# 🛠️Marsho +Marsho is a simple module loader. It allows to load kits and its function from `tools` in plugin directory, for AI to use. +More information for Function Call, please refr to [OpenAI Offical Documentation](https://platform.openai.com/docs/guides/function-calling) + +## ✍️ Development Kits +### 📁 Directory Structure +`tools` in plugin directory is caller **Toolset**. It contains many **Toolkit**, Toolkit is similar with **Packages** in Python in structure. It need `__init__.py` document and `tools.json` definition document in it. They are used to store and define functions. + +A directory structure of Toolkit: +``` +tools/ # Toolset Directory +└── marshoai-example/ # Toolkit Directory, Named as Packages' name + └── __init__.py # Tool Module + └── tools.json # Function Definition Document +``` +In this directory tree: +- **Toolset Directory** is named as `marshoat-xxxxx`, its name is the name of Toolset. Please follow this naming standard. +- ***Tool Module* could contain many callable **Sync** function, it could have parameters or be parameterless and the return value should be supported by AI model. Generally speaking, the `str` type is accepted to most model. +- **Function Definition Document** is for AI model to know how to call these function. +### Function Writing +Let's write a function for getting the weather and one for getting time. +###### **\_\_init\_\_.py** +```python +from datetime import datetime + +async def get_weather(location: str): + return f"The temperature of {location} is 114514℃。" #To simulate the return value of weather. + +async def get_current_time(): + current_time = datetime.now().strftime("%Y.%m.%d %H:%M:%S") + time_prompt = f"Now is {current_time}。" + return time_prompt +``` +In this example, we define two functions, `get_weather` and `get_current_time`. The former accepts a `str` typed parameter. Let AI to know the existence of these funxtions, you shuold write **Function Definition Document** +###### **tools.json** +```json +[ + { + "type": "function", + "function": { + "name": "marshoai-example__get_weather", # Name of this Function Call + "description": "Get the weather of a specified locatin.", # Description, it need to descripte the usage of this Functin + "parameters": { # Define the parameters + "type": "object", + "properties": { + "location": { # 'location' is the name that _init__.py had defined. + "type": "string", # the Type of patameters + "description": "City or district. Such as Beijing, Hangzhou, Yuhang District" # Description,it need to descripte the type or example of Actual Parameter + } + } + }, + "required": [ # Define the Required Parameters + "location" + ] + } + }, + { + "type": "function", + "function": { + "name": "marshoai-example__get_current_time", + "description": "Get time", + "parameters": {} # No parameter requried, so it is blanked + } + } +] +``` +In this document, we defined tow function. This Function Definition Document will be typed into AI model, for letting AI to know when and how to call these function. +The name of **Name of this Function Call** is specific required. Using weather-getting as an example, the Name of this Function Call, `marshoai-example__get_weather`, contain these information. +- **marshoai-example** is the name of its Toolkit. +- **get_weather** is the name of function. +- Two **underscores** are used as a sseparator. + +Using this Naming Standard, it could be compatible with more APIs in the standard format of OpenAI. So don't use two underscores as the name of Toolkit or Function. +### Function Testing +After developing the Tools, start the Bot. There loading information of Toolkit in Nonebot Logs. +This is the test ecample: +``` +> marsho What's the weather like in Shenzhen? +> marsho Please tell me the weather in Shimokitazawa, Hangzhou, and Suzhou separately. +> marsho What time is it now? +``` \ No newline at end of file