OpenAI & Gemini Function Decorator

This package provides a decorator @openai_function that automatically creates function tools from Python functions and executes them using either:

• Gemini API (default)
• OpenAI API
• Any OpenAI-compatible API

Key Features

• 🚀 Uses Gemini 2.5 Flash by default (powerful, fast, efficient)
• 🤖 Supports both Gemini and OpenAI through same interface
• 🔄 Automatic tool schema generation from function docstrings
• 🧠 AI-powered parameter generation when called without arguments
• ⚙️ Full configuration of API keys, models, and endpoints
• 📄 Supports complex types and default values

Installation

pip install openaiWrapper

Quick Start

1. Get Gemini API Key

Create an API key from Google AI Studio

2. Import and Configure

from openAiWrapper import openai_function, OpenAIWrapper

# Set your Gemini API key
OpenAIWrapper.set_api_key("your-gemini-api-key")

# Optional: Explicitly set Gemini endpoint
OpenAIWrapper.set_base_url("https://generativelanguage.googleapis.com/v1beta/openai/")

3. Define Your Function

@openai_function(temperature=0.7)
def calculate_sum(a: int, b: int) -> int:
    """Sum two integers and return the result"""
    return a + b

4. Use Your Function

# Direct call (bypasses Gemini)
print(calculate_sum(5, 3))  # Output: 8

# Gemini-powered call (parameters generated by AI)
print(calculate_sum())  # Uses AI to determine parameters

# Custom prompt query
print(calculate_sum.query("What is 17 plus 24?"))  # Output: 41

Advanced Usage

Different Models

# Use gemini-1.5-flash-latest
@openai_function(model="gemini-1.5-flash-latest")
def analyze_text(text: str) -> dict:
    """Analyze text and return sentiment and key entities"""
    # Implementation would call external NLP service...
    return {"sentiment": 0.8, "entities": ["Google", "AI"]}

# Use OpenAI models (if configured)
@openai_function(
    model="gpt-4o", 
    base_url="https://api.openai.com/v1/",
    api_key="your-openai-key"
)
def openai_function(text: str) -> str:
    """Process text using OpenAI"""
    return text.upper()

Complex Parameters and Defaults

@openai_function(max_tokens=300)
def generate_email(recipient: str, subject: str, 
                  tone: str = "formal", urgency: int = 3) -> str:
    """
    Generate an email with given parameters
    
    Parameters:
    recipient: Email recipient name
    subject: Email subject
    tone: Writing tone (friendly, formal, urgent) - default: formal
    urgency: Importance level (1-5) - default: 3
    """
    # Implementation would generate email content...
    return f"Dear {recipient},\n\nAbout: {subject}\n\n[Email content]"

Summarization Example

@openai_function()
def summarize_document(text: str, length: str = "medium") -> str:
    """
    Summarize text to specified length
    
    Parameters:
    text: Text to summarize
    length: Summary length ('short', 'medium', 'long' - default: medium)
    """
    if length == "short":
        return text[:100] + "..."  # Simplified example
    elif length == "medium":
        return text[:250] + "..."
    else:
        return text[:500] + "..."
    
# Usage
sample_text = ("Google's Gemini is a powerful AI model " * 20)
print(summarize_document.query(f"Summarize this: {sample_text}"))

Error Handling

Clear error messages for:

• Missing API keys
• Gemini/API errors
• Parameter validation issues
• Unexpected responses

Switching Providers

To Use OpenAI Instead:

# Configure OpenAI
OpenAIWrapper.set_api_key("your-openai-key")
OpenAIWrapper.set_base_url("https://api.openai.com/v1/")

@openai_function(model="gpt-4o")
def openai_function(text: str) -> str:
    """Process text using OpenAI"""
    return f"Processed: {text}"

To Use Local Endpoints:

# Local AI service (e.g., LocalAI, ollama)
OpenAIWrapper.set_base_url("http://localhost:8080/v1")

@openai_function(model="llama3")
def local_ai_function(query: str) -> str:
    """Answer questions using local AI"""
    return "Response from local AI"

Development

Install Locally

git clone https://github.com/pilot4u/openaiWrapper.git
cd openaiWrapper
pip install -e .

Run Tests

python test/test.py

License

MIT License