app.py

import gradio as gr
from huggingface_hub import InferenceClient
import os
HF_TOKEN = os.getenv("HF_TOKEN")
client = InferenceClient("deepcogito/cogito-v2-preview-llama-109B-MoE")

def chat(prompt):
    try:
        output = client.chat_completion(prompt)
        return output
    except Exception as e:
        return f"Error: {e}"

gr.Interface(
    fn=chat,
    inputs="text",
    outputs="text",
    title="💬 Cogito Chat Demo",
    description="Runs Cogito remotely via Hugging Face's free API (no download)."
).launch()