RightNow AI logoRightNow AI<Preview>

Quick Start

Get productive with CUDA development in 3 simple steps

Step 1: Installation

Download RightNow AI

  1. Visit rightnowai.co and download for your platform
  2. Run the installer and follow setup prompts
  3. Launch RightNow AI

Install CUDA Toolkit

Required: CUDA Toolkit 11.0-12.5

  1. Download from NVIDIA Developer
  2. Install the toolkit for your platform (Windows/Linux)
  3. Verify installation: nvcc --version

Tip: RightNow AI will automatically detect your CUDA installation

Step 2: Open Your First CUDA Project

Create a new project

bash
# Create a new project folder
mkdir my-cuda-project
cd my-cuda-project

# Open in RightNow AI
rightnow .

Or use File → Open Folder in the UI.

Create your first CUDA file

Create a simple hello.cu file:

cuda
1#include <stdio.h>
2 
3__global__ void hello() {
4    printf("Hello from GPU thread %d!\n", threadIdx.x);
5}
6 
7int main() {
8    hello<<<1, 10>>>();
9    cudaDeviceSynchronize();
10    return 0;
11}

Step 3: Enable AI & Profile

Setup OpenRouter API Key (Free Tier)

Required for AI features - OpenRouter is the only supported BYOK provider

  1. Create account at openrouter.ai
  2. Get API key from openrouter.ai/settings/keys
  3. In RightNow AI: Settings → AI Providers → OpenRouter
  4. Enter your OpenRouter API key
  5. Test connection

Alternative: Upgrade to RightNow Pro for managed AI access

Profile your first kernel

  1. Click the ▶️ button next to your __global__ function
  2. View comprehensive metrics:
    • Execution time and SM efficiency
    • Memory throughput and occupancy
    • Cache hit rates and power consumption
  3. See color-coded performance indicators (🟢 🟡 🔴)
  4. Get AI-powered optimization suggestions

Tip: Try asking the AI: "How can I optimize this kernel for memory bandwidth?"

What's Next?

AI-Powered Coding

Use Ctrl+K editing and chat for CUDA development

Advanced Profiling

Explore comprehensive NCU metrics and AI optimization

Local Models

Set up Ollama, vLLM for privacy-first development

OpenRouter Models

Access 200+ models through OpenRouter integration

Need help? Try the chat sidebar and ask: "How do I get started with CUDA profiling?"