app.py

import gradio as gr
from google import genai
import os
import time

# Initialize Client
client = genai.Client(api_key=os.environ.get("GEMINI_API_KEY"), http_options={'api_version': 'v1'})

# Priority list of models (Flash-Lite is the most 'generous' in 2026)
MODEL_PRIORITY = ["gemini-2.0-flash", "gemini-2.5-flash-lite", "gemini-1.5-flash"]

def bot(message, history):
    prompt_parts = [message["text"]]
    for file_path in message["files"]:
        with open(file_path, "rb") as f:
            prompt_parts.append(genai.types.Part.from_bytes(data=f.read(), mime_type=None))

    # Loop through models if one is exhausted
    for model_id in MODEL_PRIORITY:
        try:
            response = client.models.generate_content_stream(
                model=model_id,
                contents=prompt_parts
            )
            
            partial_message = ""
            for chunk in response:
                if chunk.text:
                    partial_message += chunk.text
                    yield partial_message
            return  # Success! Exit the function.

        except Exception as e:
            if "429" in str(e):
                print(f"🔄 {model_id} exhausted, trying next model...")
                continue 
            else:
                raise gr.Error(f"API Error: {str(e)}")

    raise gr.Error("❌ All free models are currently exhausted. Please wait a few minutes.")

demo = gr.ChatInterface(fn=bot, multimodal=True, title="Resilient Gemini Assistant")
demo.launch()