I wrote a minimalist library for dealing with tools and function calls: https://github.com/zby/LLMToolBox:
LLMToolBox is a minimal Python package designed for seamless interaction with OpenAI Python API library. It focuses on ‘tools’ and ‘function calls’, offering a minimalist approach that doesn’t get in your way.
One of the key advantages of LLMToolBox is its non-intrusive design. It doesn’t take over the interaction with the OpenAI client, making it easier for developers to debug and see exactly what values are being passed.
By integrating Pydantic, LLMToolBox ensures robust data validation and schema generation.
Features
- Schema Generation: Effortlessly create JSON schemas for tools using Pydantic models.
- Function Name Mapping: Flexibly map JSON schema names to Python code.
- Dispatching Function Calls: Directly invoke functions based on LLM response structures.