GaasHub Documentation

Section 1 - Getting Started

1.1 What is GaasHub?

GaasHub is a revolutionary tool that turns any standard computer, laptop, or edge device into a high-powered GPU workstation instantly. Designed for ML engineers, robotics builders, and students, GaasHub eliminates the pain of complex cloud infrastructure setups, AWS billing surprises, and constant environment configuration. With a single click, your code is magically accelerated by remote RTX 3090/4090 and A6000 nodes - as if the hardware was sitting right on your desk.

1.2 System Requirements

Additional Requirements: Minimum 4GB RAM, 500MB free disk space, and a stable internet connection (>10 Mbps recommended for smooth output streaming).

1.3 Installation - Platform by Platform

Windows

1. Download the .exe installer from our homepage.
2. Run the installer (if Windows Defender flags it, select "More Info" → "Run Anyway").
3. Launch GaasHub from your Start Menu.
4. Log in or create a new account.

macOS

1. Download the .dmg file suitable for your chip (Apple Silicon or Intel).
2. Open the .dmg and drag the GaasHub application into your Applications folder.
3. Launch GaasHub and log in.

Linux

1. Run our one-liner install command to grab dependencies and the AppImage:

sudo apt update && sudo apt install -y fuse libfuse2 libgtk-3-0 libnss3 libdrm2 libgbm1 libasound2 build-essential && wget https://pub-aa7f5aa73b3d4a5c9075e918074b52aa.r2.dev/GaasHub-1.0.38.AppImage -O GaasHub.AppImage && chmod +x GaasHub.AppImage

2. Launch the AppImage using ./GaasHub.AppImage --no-sandbox.
3. Log in or create your account.

Raspberry Pi / ARM Boards

1. Determine your architecture by running uname -m in the terminal. (aarch64 = ARM64, armv7l = ARMv7).
2. Download the corresponding AppImage from the website.
3. Make it executable (chmod +x) and launch it. Log in to start your node.

1.4 Creating Your Account

Upon launching GaasHub for the first time, you will be prompted to log in. Click Create Account, enter your email and a secure password. Verify your email via the link sent to your inbox, then return to the app. You're ready to go!

1.5 Your First GPU Job - Hello World

This is the moment of truth. Let's verify you have real GPU access.

1. Open the GaasHub app, select a remote GPU, and click Start.
2. Open a standard, local terminal window.
3. Run exactly this command:

gaashub -c "import torch; print(torch.cuda.is_available())"

If it prints True - congratulations, you are successfully running code on a remote GPU!

1.6 How the gaashub Prefix Works

When you type gaashub script.py, the CLI intelligently intercepts the execution. It bundles your local environment context, securely transmits your code to the allocated remote GPU node over an encrypted channel, executes it natively on the NVIDIA hardware, and streams standard output, error, and relevant file artifacts back to your local terminal in real-time. To you, it feels completely local; under the hood, it's seamless cluster computing.

Section 2 - Supported Frameworks

2.1 PyTorch

• Supported Versions: 1.13.x, 2.0.x, 2.1.x, 2.2.x
• CUDA Compatibility: CUDA 11.8 and CUDA 12.1+

Working Example:

import torch
# Automatically uses remote GPU via GaasHub
x = torch.rand(5, 3)
print(x.cuda())

2.2 TensorFlow

• Supported Versions: TensorFlow 2.10 through 2.15
• Setup Differences: None. TF automatically detects the GaasHub GPU.

Working Example:

import tensorflow as tf
print("Physical Devices:", tf.config.list_physical_devices('GPU'))

2.3 JAX

• Supported Versions: Latest JAX & jaxlib (CUDA-enabled)
• Note: Ensure you install the jaxlib version matching the CUDA runtime of the selected GaasHub node.

2.4 Raw CUDA

You can compile and run native .cu C++ files directly using NVIDIA's nvcc compiler through GaasHub.

Example:

gaashub nvcc hello.cu -o hello && gaashub ./hello

2.5 Scikit-learn / NumPy / Pandas

Do these work? Yes.
Do they benefit from GPU? No, standard NumPy/Pandas run on the remote CPU of the GaasHub node. For GPU acceleration of dataframes, use RAPIDS cuDF.

2.6 Hugging Face Transformers

Hugging Face is natively supported. Large models will download directly to the GaasHub node, leveraging its high-speed internet backbone, saving you local bandwidth.

from transformers import pipeline
classifier = pipeline("sentiment-analysis", device=0) # device=0 points to GaasHub GPU
print(classifier("GaasHub makes GPU access effortless!"))

2.7 Coming Soon

We are actively working on first-class support for Mojo, distributed Ray clusters, and native Jupyter Notebook tunneling.

Section 4 - Troubleshooting

4.1 General Issues

4.2 Linux Specific

4.3 Windows Specific

4.4 macOS Specific

4.5 Raspberry Pi Specific

4.6 Where to Get More Help

We're here to help!
- Email us at info@gaashub.com (We usually respond within 60 minutes).

🔒 Section 5 - Security & Privacy

We treat your code and proprietary datasets with enterprise-grade paranoia. Here is exactly how we handle security:

• Data Transmission: All commands, code payloads, and output streams are encrypted instantly using TLS v1.3.
• Code Persistence: We do not store your code. Upon session termination, the container running on the remote hardware is immediately destroyed and the underlying disk is cryptographically wiped.
• Isolation: Every user gets an isolated Docker environment mapped exclusively to their allocated GPU hardware. There is zero cross-contamination.
• Local Access: The GaasHub CLI only reads the specific contextual directory you execute it from. It does not scan your local file system.

📋 Section 6 - FAQ