🧠 Arabic-Only AI Assistant (via OpenRouter + Deepseek) This project sets up an AI Assistant that only responds in Arabic, powered by the deepseek/deepseek-prover-v2:free model through OpenRouter API. It is designed to run in Google Colab using openai-agents and AsyncOpenAI.

🚀 Features Uses OpenRouter to access third-party AI models

Sets strict instruction boundaries (Arabic-only replies)

Asynchronous execution with asyncio

Simple and minimal example using the Agent SDK

🧩 Tech Stack Python (async/await)

OpenRouter

Deepseek Prover Model (deepseek/deepseek-prover-v2:free)

openai-agents

Google Colab integration

🔧 Setup (Google Colab) python Copy Edit

Install the agent SDK

!pip install -Uq openai-agents

Apply async loop patch for Colab

import nest_asyncio nest_asyncio.apply()

Load API key securely from Colab

from google.colab import userdata OPENROUTER_API_KEY = userdata.get("DEEPSEEK_KEY")

OpenRouter API setup

from openai import AsyncOpenAI BASE_URL = "https://openrouter.ai/api/v1" MODEL = "deepseek/deepseek-prover-v2:free" client = AsyncOpenAI(api_key=OPENROUTER_API_KEY, base_url=BASE_URL) 📜 Run the Assistant python Copy Edit from agents import Agent, OpenAIChatCompletionsModel, Runner, set_tracing_disabled import asyncio

set_tracing_disabled(disabled=True)

async def main(): agent = Agent( name="Assistant", instructions="You can only respond in arabic", model=OpenAIChatCompletionsModel(model=MODEL, openai_client=client) ) result = await Runner.run(agent, "what is your name?") print(result.final_output)

if name == "main": asyncio.run(main()) 🔐 Notes Replace DEEPSEEK_KEY in userdata.get() with your actual variable name.

Make sure you’ve set up OpenRouter API access and your model is authorized for use.