---
title: Sentinel - ROCm
license: mit
sdk: gradio
emoji: 🚀
colorFrom: green
colorTo: gray
pinned: true
short_description: 'ROCm-Sentinel, a system designed to redefine how developers '
---
<div align="center">
  <img src="https://img.shields.io/badge/AMD-ROCm-ed1c24?style=for-the-badge&logo=amd&logoColor=white" alt="AMD ROCm" />
  <img src="https://img.shields.io/badge/AI_Agent-LangGraph-000000?style=for-the-badge&logo=openai&logoColor=white" alt="LangGraph" />
  <img src="https://img.shields.io/badge/UI-Gradio-ff7c00?style=for-the-badge" alt="Gradio" />
  
  <h1>🚀 ROCm-Sentinel</h1>
  <p><strong>Secure AI Agent for Automated Migration from NVIDIA CUDA to AMD ROCm (HIP)</strong></p>
</div>

---

## 📖 About the Project

**ROCm-Sentinel** is an AI-assisted tool designed to break vendor lock-in in hardware-accelerated software development. Using an Agentic Workflow, the application takes native NVIDIA CUDA code and automatically translates it to AMD HIP (Heterogeneous-Compute Interface for Portability), ensuring a smooth transition into the AMD ROCm ecosystem.

It acts as more than just a syntactic translator; it incorporates an autonomous **Security Auditor** that reviews the generated code for memory leaks, parallelism issues, and architecture-specific optimizations for AMD.

---

## 🎯 Goals and Vision

1. **Democratize Hardware:** Facilitate developers and companies in migrating their workloads from closed ecosystems to open and competitive platforms like AMD.
2. **Secure Migration (Zero-Leak):** Go beyond simple "find and replace". Ensure the resulting code correctly manages unified memory and threads on AMD GPUs.
3. **Interactive Workflow:** Provide a user-friendly interface where the developer acts as a supervisor (Human-in-the-Loop), guiding the agent through real-time chat.

---

## 🧠 Language Models (AI Agents)

Our multi-agent architecture (Translator + Auditor) is powered by state-of-the-art LLMs via the Fireworks AI API.

* 🟢 **Current Model in Use: `DeepSeek V3.1`**
    * The project is currently configured to run on DeepSeek V3.1, a highly capable model with exceptional logical reasoning and lightning-fast response times for interactive demonstration.
* ⭐ **Recommended Model: `Qwen2.5-Coder 32B Instruct`**
    * For production environments and complex code repositories, we **strongly recommend** switching the configuration to the Qwen 2.5 Coder model. Being trained specifically on deep programming syntax, it offers superior fidelity when translating pointers and complex C++/CUDA directives.

---

## 🛠️ Technologies Used

* **Python 3** - Base language.
* **Gradio** - Interactive web interface construction.
* **LangGraph & LangChain** - Agent workflow orchestration (nodes and edges).
* **Fireworks AI** - Serverless inference provider for open-source LLM models.

---

## ⚙️ Local Installation and Usage

Follow these steps to deploy ROCm-Sentinel on your local machine:

1. **Clone the repository:**
   ```bash
   git clone [https://github.com/MrZRo/ROCm-Sentinel.git](https://github.com/MrZRo/ROCm-Sentinel.git)
   cd ROCm-Sentinel

2. **Create and activate a virtual environment:**
  ```bash
  # En Windows
  python -m venv env
  source env/Scripts/activate
  
  # En Linux/Mac
  python3 -m venv env
  source env/bin/activate
  ```

3. **Install dependencies:**
   ```bash
   pip install -r requirements.txt
    ```

4. **Configure credentials:**
   ```bash
   FIREWORKS_API_KEY=tu_api_key_aqui
    ```

5. **Configurar las credenciales:**
   ```bash
   FIREWORKS_API_KEY=your_api_key_here
   ```

6. **Run the application:**
   ```bash
   Fpython app.py
   ```