Universal Tool Schema Generation and Gemini Assistant Example

This repo attempts at exploring function calls and paralell tool usage for the new gemini-2.0-flash-exp model as well as automate tool creation and schema convertion for other llm apis This repository demonstrates how to:

1. Convert Python functions into schemas compatible with various AI APIs (e.g., OpenAI, Anthropic, and Gemini).
2. Use these schemas to initialize a Gemini assistant that can handle function calls in a conversation.
3. Showcase parallel function calling and schema transformations from a single list of Python functions.

Overview

tool_converters/tool_converter.py

Uses Google Gemini models to generate OpenAI-like schemas from Python functions.

tool_converters/schema_converter.py

Converts the OpenAI-like schemas generated by tool_converter.py into Anthropic and Gemini formats.
Also provides a single generate_all_schemas(functions) method that returns all formats at once:

{
  "openai": [...],
  "anthropic": [...],
  "gemini": [...]
}

tool_generators.py

Demonstrates generating all schemas from a list of Python functions.
Prints Anthropic, Gemini, and OpenAI schemas.

gemini_assistant.py

Provides GeminiAssistant, a class that:

• Accepts either a functions map, a list of functions, or a provided OpenAI schema.
• If given functions, it uses ToolConverter internally to create an OpenAI-like schema, then uses schema_converter to convert to Gemini tools.
• Initializes a Gemini model using the new Gemini 2.0 Flash experimental release.
• Handles parallel function calling automatically, making it simple to integrate and manage multiple tool invocations.
• Runs queries and automatically executes functions upon the model’s requests.

main.py

Example of using GeminiAssistant with a list of Python functions.
Runs a user query that triggers function calls and prints the final answer.

Requirements

• Python 3.9+
• google-generativeai Python package
• Set the GEMINI_API_KEY environment variable to your Google Gemini API key.

Usage

1. Install dependencies:

   pip install google-generativeai

2. Set your API key:

   export GEMINI_API_KEY=your_api_key_here

3. Run tool_generators.py to see the generated schemas:

   python tool_generators.py

4. Run main.py to see the Gemini assistant in action:

   python main.py

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Google Generative AI team for the Gemini API
• Contributors and maintainers