NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC

A major update is coming in August 2026.

Quick links: Overview · Installation · Production deployment · Local development · Planning · Contributing · Security · Community · Licensing

Your AI app store for NVIDIA GPU platforms. Browse, install, and launch AI apps with one click.

Overview

NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC provides a web UI and API for managing curated AI application recipes across NVIDIA GPU workstations, servers, and DGX-class systems.

The project includes:

• A FastAPI backend for recipe, system, and container management
• A React + Vite frontend served as static files by the backend
• An optional Electron desktop shell for packaged Windows, macOS, and Linux builds
• A Docker-based runtime model for AI applications in registry/recipes

Local development

For a complete local dev workflow (backend + frontend), production build validation, and white-page troubleshooting, see docs/local-development.md.

Desktop packaging

The repository now includes optional Electron packaging assets under frontend/electron/.

Desktop packaging currently follows this model:

• the web UI is still built from source with Vite
• the desktop app launches a bundled local FastAPI backend on 127.0.0.1
• packaged builds target Windows, macOS, and Linux through electron-builder
• Linux artifacts are intended to include the desktop app bundle rather than rely on the shell installer alone

Frontend desktop commands:

• npm run desktop:dev
• npm run desktop:pack
• npm run desktop:dist

These commands are run from frontend/ and require local Python plus repository backend dependencies to package successfully.

Desktop packaging notes:

• app icons are generated from frontend/public/brand/spark-ai-hub-mark.svg during desktop packaging
• frontend production builds now split large vendor dependencies into dedicated chunks to reduce the main bundle size for web and desktop output
• desktop pack/dist commands now clear frontend/release/ before packaging to avoid stale locked output directories on Windows
• npm install in frontend/ now re-checks and repairs an incomplete Electron binary install automatically
• Windows signing is designed to be driven by CSC_LINK and CSC_KEY_PASSWORD
• macOS signing and notarization additionally expect CSC_NAME, APPLE_ID, APPLE_APP_SPECIFIC_PASSWORD, and APPLE_TEAM_ID
• transient Electron download failures from upstream mirrors can surface as HTTP 504 during npm run desktop:dist; retrying the build is the expected first mitigation

Installation and deployment

For Linux deployment, Windows and macOS setup boundaries, Docker and NVIDIA runtime prerequisites, and optional PM2 process management, see docs/installation.md.

For repository automation, release packaging, and installer builds, see docs/github-actions.md.

For production-style Linux service management, reverse proxy setup, TLS, and network exposure guidance, see docs/deployment-production.md.

For tracked deployment example files, see deploy/systemd/[email protected], deploy/nginx/nvidia-ai-hub.conf, deploy/caddy/Caddyfile, and deploy/pm2/ecosystem.config.cjs.

Contributing

See CONTRIBUTING.md for contribution workflow, development setup, recipe guidance, and pull request expectations.

Security and Conduct

• Security policy: SECURITY.md
• Community and collaboration rules: CODE_OF_CONDUCT.md

Community

• Contribution guide: CONTRIBUTING.md
• Security policy: SECURITY.md
• Code of conduct: CODE_OF_CONDUCT.md
• Support guide: SUPPORT.md
• Governance index: docs/community.md
• GitHub Actions rollout guide: docs/github-actions.md
• Repository maintenance guide: docs/maintenance.md
• Pull request process: docs/pull-request-process.md
• Licensing guide: docs/licensing.md
• Discussions: enable GitHub Discussions in repository settings for support questions, ideas, and roadmap conversations
• Sponsorships and commercial licensing: COMMERCIAL-LICENSE.md

Repository Operations

For repository automation, governance maintenance, and review routing, use:

• GitHub Actions rollout guide: docs/github-actions.md
• Repository maintenance guide: docs/maintenance.md
• Pull request process: docs/pull-request-process.md

Planning and roadmap

The repository keeps product direction, implementation sequencing, and catalog growth planning in tracked planning documents.

• Product blueprint: planning/sparkdeck-project-blueprint.md
• Execution backlog and phased delivery plan: planning/development-execution-plan.md
• Registry sourcing and batch expansion plan: planning/registry-expansion-roadmap.md

Use these documents together:

• sparkdeck-project-blueprint.md defines the long-range product model and feature architecture
• development-execution-plan.md converts that model into bounded implementation workstreams and phases
• registry-expansion-roadmap.md governs recipe expansion batches, banner coverage, and category balance

When product scope, legal positioning, or rollout priorities change, keep the planning files synchronized with README.md, docs/community.md, and the relevant governance documents.

Legal Notice and Trademark Attribution

NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC is a software solution developed and distributed by Pho Tue SoftWare And Technology Solutions Joint Stock Company (including the HiTechCloud brand).

Company information:

• Legal entity: CÔNG TY CỔ PHẦN GIẢI PHÁP CÔNG NGHỆ VÀ PHẦN MỀM PHỔ TUỆ
• English name: Pho Tue SoftWare And Technology Solutions Joint Stock Company
• Tax code: 0318222903
• D-U-N-S Number: 557339920
• Address: 128 Binh My Street, Binh My Commune, Ho Chi Minh City

NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC, related repository branding, and associated product presentation in this repository are proprietary identifiers used for this solution.

NVIDIA, the NVIDIA logo, DGX, CUDA, and other NVIDIA product or program names are trademarks and/or registered trademarks of NVIDIA Corporation and its affiliates in the United States and other countries.

Any reference to NVIDIA hardware, software, platforms, runtimes, or ecosystem technologies in this repository is provided solely to describe compatibility, deployment requirements, or integration context.

No statement in this repository should be interpreted as:

• granting any license to use NVIDIA trademarks except for lawful nominative reference;
• implying sponsorship, endorsement, certification, partnership, or approval by NVIDIA Corporation, unless such relationship is expressly stated in writing; or
• transferring any ownership in the names, logos, trade dress, or brand assets of Pho Tue SoftWare And Technology Solutions Joint Stock Company, NVIDIA Corporation, or any other third party.

The short product name NVIDIA AI Hub and the detailed product name NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC both refer to the same software solution described in this repository.

All other trade names, trademarks, service marks, logos, and brand features mentioned in this repository remain the property of their respective owners.

For controlling legal terms, review LICENSE, NOTICE, COMMERCIAL-LICENSE.md, docs/licensing.md, docs/legal-notice.md, and the Vietnamese reference notice docs/legal-vi.md.

Quick install

curl -fsSL https://raw.githubusercontent.com/hitechcloud-vietnam/nvidia-ai-hub/main/install.sh | bash

Install without starting the server:

curl -fsSL https://raw.githubusercontent.com/hitechcloud-vietnam/nvidia-ai-hub/main/install.sh | bash -s -- --no-start

Install on a custom port:

curl -fsSL https://raw.githubusercontent.com/hitechcloud-vietnam/nvidia-ai-hub/main/install.sh | bash -s -- --port 9010

Install on a custom host and port:

curl -fsSL https://raw.githubusercontent.com/hitechcloud-vietnam/nvidia-ai-hub/main/install.sh | bash -s -- --host 127.0.0.1 --port 9010

Windows local setup

Use the local development guide for Windows-compatible commands:

• docs/local-development.md

For supported platform boundaries and deployment guidance, see docs/installation.md.

After installation, open:

• http://localhost:9000
• or http://<your-host-ip>:9000 from another device on the same network

Run the same command again to update.

What the installer does

The installer is designed to provision both backend and frontend automatically.

It will:

1. Install git if missing
2. Install python3, python3-venv, and pip if missing
3. Install Docker Engine if missing
4. Install Node.js 22.x if the system version is not suitable for the frontend build
5. Clone or update the nvidia-ai-hub repository in $HOME/nvidia-ai-hub
6. Create a Python virtual environment in .venv
7. Install backend dependencies from requirements.txt
8. Install frontend dependencies from frontend/package.json
9. Build the production frontend into frontend/dist
10. Create .env from .env.example when needed
11. Persist NVIDIA_AI_HUB_HOST and NVIDIA_AI_HUB_PORT in the root .env
12. Start the backend with ./run.sh using the configured host and port

Because the backend serves the built frontend from frontend/dist, the UI is available immediately after install.

If --no-start or -NoStart is used, the installer completes all setup steps but skips launching the API server.

If --port, --host, -Port, or -Host is used during install, the chosen values are written into the shared root .env file.

run.sh also supports --host and --port for one-off overrides and will otherwise read the persisted values from .env.

Features

• Browse a catalog of AI apps ready for NVIDIA GPU platforms
• Install any app with one click — no terminal needed
• Launch, stop, and monitor running apps from the dashboard
• Track GPU, RAM, disk, and temperature in real time

Available apps

App	What it does	GPU
Open WebUI + Ollama	Chat with local LLMs	Yes
vLLM (Qwen 3.5)	High-performance LLM inference (8 model sizes)	Yes
ComfyUI	Image & video generation workflows	Yes
FaceFusion	Face swap & enhancement	Yes
Hunyuan3D 2.1	Image to 3D model generation	Yes
TRELLIS 2	Text/image to 3D generation	Yes
LocalAI	OpenAI-compatible API server	Yes
AnythingLLM	RAG & AI agents	No
Flowise	Drag-and-drop LLM workflows	No
Langflow	Visual LLM app builder	No

Apps are delivered as Docker-based workloads with NVIDIA runtime integration. Architecture and GPU requirements vary by recipe.

Requirements

Minimum

• Linux host with a supported NVIDIA GPU, Docker Engine, and NVIDIA Container Toolkit or equivalent NVIDIA runtime integration
• Ubuntu/Debian-based Linux environment with apt-get
• Internet access during installation
• Permission to use sudo for package installation

Installed automatically

• Git
• Python 3 + venv
• Docker Engine
• Node.js 22.x

Manual operation

Update an existing installation

Run the installer again:

curl -fsSL https://raw.githubusercontent.com/hitechcloud-vietnam/nvidia-ai-hub/main/install.sh | bash

Start manually from an existing clone

If the repository is already available locally:

./run.sh

Before starting, you can validate the machine state with:

./check.sh

run.sh now checks whether frontend/dist is missing or outdated. If needed, it rebuilds the UI automatically before starting the backend.

You can also start on a custom port for a single run:

./run.sh --port 9010

You can also override host and port for a single run:

./run.sh --host 127.0.0.1 --port 9010

If .env is present, run.sh uses NVIDIA_AI_HUB_HOST and NVIDIA_AI_HUB_PORT as defaults.

If the frontend must be rebuilt, ensure the machine has:

• node >= 22
• npm

Default service URL

• UI: http://localhost:9000
• API root: http://localhost:9000

Windows notes

Local development on Windows is supported through standard Python and npm commands. Use docs/local-development.md for setup and run instructions.

Windows is documented as a development environment rather than the primary local GPU deployment target. See docs/installation.md for platform boundaries.

Shared configuration

The repository now includes a shared root .env local file format, with .env.example checked in as the template, used by:

• daemon/config.py
• install.sh
• run.sh
• check.sh

Default values include:

• NVIDIA_AI_HUB_HOST
• NVIDIA_AI_HUB_PORT
• NVIDIA_AI_HUB_NODE_MAJOR
• NVIDIA_AI_HUB_REGISTRY_PATH
• NVIDIA_AI_HUB_DATA_DIR
• NVIDIA_AI_HUB_DB_PATH

To create a local configuration manually:

cp .env.example .env

or in PowerShell:

Copy-Item .env.example .env

install.sh creates .env from .env.example automatically when needed.

Update .env if you want to keep a custom default host, port, or path layout across runs.

Troubleshooting

The page opens but has no styling or JavaScript

This usually means the frontend build was not generated. Re-run the installer so it rebuilds frontend/dist.

You can also run ./check.sh to confirm whether the frontend bundle is missing or stale.

Docker works only with sudo

The installer adds the current user to the docker group. Log out and log back in, or run:

newgrp docker

python3 -m venv .venv fails

Ensure python3-venv is installed. The installer attempts to install it automatically.

Frontend build fails because of Node.js version

The installer installs Node.js 22.x when the detected version is too old. Re-run the installer if the system Node version changed unexpectedly.

run.sh exits with a frontend build requirement message

This means the checked-in or generated UI bundle is missing or stale, and the current machine does not have a compatible Node.js toolchain. Run install.sh to provision Node.js and rebuild the frontend.

check.sh reports Docker daemon is not reachable

Start Docker Desktop or the Docker service, then re-run ./check.sh. NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC can start without Docker only in a limited UI/API state.

Windows dependency installation fails

Install Git, Python 3.11+, Node.js 22+, and Docker Desktop manually, then follow docs/local-development.md.

Uninstall

curl -fsSL https://raw.githubusercontent.com/hitechcloud-vietnam/nvidia-ai-hub/main/uninstall.sh | bash

Preserve local runtime data during uninstall:

./uninstall.sh --keep-data

The uninstaller now removes, in order:

• NVIDIA AI Hub by Pho Tue SoftWare Solutions JSC recipe containers, images, and volumes
• Backend cache/runtime paths such as .venv and data/
• Frontend cache/build paths such as frontend/node_modules and frontend/dist
• Generated recipe .env files
• Python cache directories such as __pycache__
• The installation directory itself

It does not uninstall Docker itself.

For repository contribution standards and templates, see CONTRIBUTING.md.

License

This repository is licensed for strictly non-commercial use under the terms of LICENSE.

Copyright (c) 2026 HiTechCloud by Pho Tue SoftWare Solutions JSC.

Commercial use, client delivery, paid services, SaaS distribution, marketplace redistribution, and other revenue-generating usage require separate written permission from the copyright holder. See COMMERCIAL-LICENSE.md and docs/licensing.md.