ai-proxy-core 0.1.7

Minimal, reusable AI service handlers for Gemini and other LLMs

AI Proxy Core

A minimal Python package providing reusable AI service handlers for Gemini and other LLMs. No web framework dependencies - just the core logic.

Installation

pip install ai-proxy-core

Or install from source:

git clone https://github.com/ebowwa/ai-proxy-core.git
cd ai-proxy-core
pip install -e .

Usage

Completions Handler

from ai_proxy_core import CompletionsHandler

# Initialize handler
handler = CompletionsHandler(api_key="your-gemini-api-key")

# Create completion
response = await handler.create_completion(
    messages=[
        {"role": "user", "content": "Hello, how are you?"}
    ],
    model="gemini-1.5-flash",
    temperature=0.7
)

print(response["choices"][0]["message"]["content"])

Gemini Live Session

from ai_proxy_core import GeminiLiveSession

# Example 1: Basic session (no system prompt)
session = GeminiLiveSession(api_key="your-gemini-api-key")

# Example 2: Session with system prompt (simple string format)
session = GeminiLiveSession(
    api_key="your-gemini-api-key",
    system_instruction="You are a helpful voice assistant. Be concise and friendly."
)

# Example 3: Session with system prompt (Content object for more control)
from google.genai import types
session = GeminiLiveSession(
    api_key="your-gemini-api-key",
    system_instruction=types.Content(
        parts=[types.Part.from_text("You are a pirate. Speak like a pirate!")],
        role="user"
    )
)

# Set up callbacks
session.on_audio = lambda data: print(f"Received audio: {len(data)} bytes")
session.on_text = lambda text: print(f"Received text: {text}")  # Text transcription of audio

# Start session
await session.start()

# Note: When using audio responses, Gemini automatically provides both:
# - Audio data (PCM16 format) via on_audio callback
# - Text transcription of the audio via on_text callback

# Send audio/text
await session.send_audio(audio_data)
await session.send_text("Hello!")

# Stop when done
await session.stop()

Integration with FastAPI

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from ai_proxy_core import CompletionsHandler

app = FastAPI()
handler = CompletionsHandler()

class CompletionRequest(BaseModel):
    messages: list
    model: str = "gemini-1.5-flash"
    temperature: float = 0.7

@app.post("/api/chat/completions")
async def create_completion(request: CompletionRequest):
    try:
        response = await handler.create_completion(
            messages=request.messages,
            model=request.model,
            temperature=request.temperature
        )
        return response
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

Features

• No framework dependencies - Use with FastAPI, Flask, or any Python app
• Async/await support - Modern async Python
• Type hints - Full type annotations
• Minimal surface area - Just the core logic you need
• Easy testing - Mock the handlers in your tests

Development

Building the Package

When building the package for distribution, use setup.py directly instead of python -m build to avoid pip isolation issues:

python setup.py sdist bdist_wheel

This will create both source distribution and wheel files in the dist/ directory.

Publishing to PyPI

twine upload dist/*

License

MIT