app.py

import gradio as gr
import ollama
import os

def chat(message, history):
    stream = ollama.chat(model='gemma4:e2b', messages=[{'role': 'user', 'content': message}], stream=True)
    reply = ""
    for chunk in stream:
        reply += chunk['message']['content']
        yield reply

demo = gr.ChatInterface(fn=chat, title="BrainMap OS on Space")
demo.launch(server_name="0.0.0.0", server_port=7860)