AgenticWrapper 0.1.0

A lightweight Python framework for adding agent capabilities (tools calling, memory, structured outputs) to single custom LLM functions

AgenticWrapper

English | 简体中文

AgenticWrapper is an extremely lightweight Python Agent framework that wraps your custom LLM function, enabling it with Agent capabilities, including tool calling, memory management, and structured output.

AgenticWrapper has no runtime dependencies.

Quick Start

Installation

pip install AgenticWrapper

Basic Usage

from AgenticWrapper import Agent

async def llm_func(messages: list[dict[str, str]]) -> str:
    # Assume this is your custom LLM function
    return "Hello, I'm AgenticWrapper."

agent = Agent(llm_func)
response = await agent.query("Hello")
print(response)
agent.clear_memory()

Structured Output

Use Python dataclass to define the output structure:

@dataclass
class SearchResult:
    query: str
    results: List[str]
    total_count: int

response = await agent.query("Search for relevant content", structured_output_type=SearchResult)
assert isinstance(response, SearchResult)

Tool Calling

Define a tool function, preferably with function documentation:

async def get_weather(location: str) -> str:
    """Get weather information"""
    # Simulate an API call
    await asyncio.sleep(0.1)
    if location.lower() == "london":
        return "The weather is clear, with a temperature of 15°C."
    elif location.lower() == "paris":
        return "The weather is clear, with a temperature of 18°C."
    else:
        return f"I don't know the weather for {location}."

Use the tool in the Agent:

agent = Agent(llm_interaction_func=mock_llm_func, tools=[get_weather])
response = await agent.query("Check the weather in london")
print(response)

Tool function parameters and return types must be str. If your tool function has complex parameters, you need to define a new wrapper function to handle serialization/deserialization and communicate the format to the LLM.

More Examples

More examples can be found in example.py.