Install Guide

Deploy NGARi on NVIDIA Jetson AGX Orin — bare metal to sovereign AI.

This guide walks through the full deployment of NGARi on an NVIDIA Jetson AGX Orin 64GB. These are the exact steps used to deploy the reference instance running at ngari.com. Total time: approximately 2 hours depending on download speeds.

1. Prerequisites

NVIDIA Jetson AGX Orin Developer Kit (64GB recommended)
NVMe SSD (512GB+) or high-endurance SD card (128GB+)
USB-C power supply (supplied with the dev kit)
Ethernet cable or WiFi (dual-band preferred)
Another computer for flashing (Ubuntu or macOS)
USB flash drive (16GB+) for flashing

2. Flash Ubuntu 24.04 LTS

NVIDIA provides JetPack and the L4T BSP through the SDK Manager tool. Flash the Jetson with Ubuntu 24.04 LTS (or 22.04 LTS if 24.04 is not yet available for your Jetson revision).

# Download and install NVIDIA SDK Manager
# Select: Jetson AGX Orin → JetPack 6.x → Ubuntu 24.04
# Follow the flashing wizard — this takes 15-30 minutes

# After flashing, boot the Jetson and complete first-run setup:
#   - Create a user account
#   - Connect to WiFi or Ethernet
#   - Run all system updates

Note: If you already have Ubuntu running on your Jetson, skip to verify the NVIDIA drivers are installed.

3. Install NVIDIA Drivers

Verify the NVIDIA drivers and CUDA are properly installed:

# Check NVIDIA driver
nvidia-smi

# Check Jetson-specific info
cat /proc/device-tree/model

# Expected output:
#   NVIDIA AGX Orin Developer Kit
#   CUDA Version: 12.x

# If nvidia-smi is not found, re-run the SDK Manager
# or install via apt:
sudo apt update
sudo apt install nvidia-jetpack

4. Install Ollama

NGARi uses Ollama for local model inference. Install the official build for Linux (aarch64):

curl -fsSL https://ollama.com/install.sh | sh

# Verify installation
ollama --version

# Configure Ollama to accept connections (needed by NGARi web UI)
sudo systemctl edit ollama.service

# Add this line:
# Environment="OLLAMA_HOST=0.0.0.0:11434"

# Restart Ollama
sudo systemctl daemon-reload
sudo systemctl restart ollama

Storage tip: Ollama stores models under ~/.ollama/models. If your home partition is small, symlink it to a larger drive: mkdir -p /mnt/ssd/ollama-models && \ ln -s /mnt/ssd/ollama-models ~/.ollama/models

5. Pull Models

Pull the models used by NGARi. These are the models deployed on the reference Orin at ngari.com:

# Core chat models
ollama pull qwen2.5:7b
ollama pull qwen2.5:32b
ollama pull qwen2.5-coder:14b
ollama pull glm4:9b
ollama pull llama3.2:1b

# Embedding model (required for agent memory)
ollama pull nomic-embed-text

# Verify all models are present
ollama list

Space: These models total approximately 35GB. Ensure you have sufficient free space before pulling. The 32B model alone is 18GB.

6. Install Docker

NGARi runs as a Docker container for easy deployment and updates.

# Install Docker
curl -fsSL https://get.docker.com | sh

# Add your user to the docker group
sudo usermod -aG docker $USER

# Log out and back in, or run:
newgrp docker

# Verify
docker --version
docker run hello-world

7. Clone and Build NGARi

# Clone the repository
git clone https://github.com/ngariglobal/ngari-site.git
cd ngari-site

# Install dependencies
npm install

# Configure environment
cp .env.example .env
# Edit .env with your settings:
#   DATABASE_URL=postgresql://... (if using shared DB)
#   OLLAMA_HOST=http://localhost:11434
#   NEXTAUTH_SECRET=<generate a random secret>

# Build for production
npm run build

# Run
npm start

Alternatively, use Docker Compose for a containerized deployment. See the docker-compose.yml in the repository.

8. Start Agent Services

NGARi includes 6 autonomous agents and a Control Tower API. Start them using the deployment scripts in the repository:

# Clone the NGARi agent platform
git clone https://github.com/ngariglobal/ns-bos.git
cd ns-bos

# Deploy all agents (CEO, CRO, Engineer, Worker, Email, Communications)
bash core/agents/deploy_agents.sh

# Start the Agent API (port 8895)
bash start_agent_api.sh

# Start the Control Tower (port 8888)
cd apps/control-tower
npm install
npm start

# Verify all services
curl http://localhost:8888/health
curl http://localhost:8895/health

9. Access the System

Once everything is running, access NGARi at these endpoints:

Chat UI: http://localhost:3001— private AI chat with all models
Control Tower: http://localhost:8888— agent monitoring, revenue, system health (JWT auth required)
Sandbox: http://localhost:3001/sandbox— browser sessions and code execution
API: http://localhost:3001/api/v1/chat/completions— OpenAI-compatible API with Bearer auth
BTCPay Server: http://localhost:3003— sovereign payment processing

10. Optional: Remote Access

To make your NGARi instance accessible from the internet (e.g., to share with your team), set up a Cloudflare tunnel:

# Install cloudflared
curl -L https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-arm64.deb
sudo dpkg -i cloudflared-linux-arm64.deb

# Authenticate and create a tunnel
cloudflared tunnel login
cloudflared tunnel create ngari

# Configure the tunnel to point to your local NGARi instance
# See docs/cloudflare-tunnel.md for the full config

Remote access requires a domain name and a Cloudflare account. The free tier of Cloudflare Tunnel is sufficient for most deployments.

Troubleshooting

Ollama not reachable. Ensure the service is running and the OLLAMA_HOST environment variable is set correctly in your Docker/NGARi configuration.

Disk space low. Ollama models and Docker images consume significant space. Consider moving Ollama models to a secondary drive or NVMe.

Web UI not loading. Check that port 3001 is not blocked by a firewall. Run ss -tlnp | grep 3001 to verify the process is listening.

Questions? [email protected] · [email protected]