setup-manual.md

Setup graph-app-kit

For quick launchers, see the AWS quick launch setup guide.

1. Launch and setup a server for Docker + extensions

Manual: Launch a Linux server and install, configure dependencies

• Open ports: 22, 80, 443, 8501 for all users (0.0.0.0/0) or for specific admin and user IPs

• Ubuntu 18.04 LTS is the most common choice for containerized GPU computing

• Install docker-ce and docker-compose

• Optional:

  • GPU: If you have a RAPIDS.ai-compatible GPU (see below), install the Nvidia docker runtime and set it as the default for Docker daemons

  • Extensions: Install Jupyter, a reverse proxy (ex: Caddy), and an authentication system

Note: GPU Instances: Cloud providers generally require you to request GPU capacity quota for your account, which may take 1 day. RAPIDS.ai-compatible GPU instance types include:

• AWS: g4, p3, p4
• Azure: NC6s_v2+, ND+, NCasT4

2. Download graph-app-kit

git clone https://github.com/graphistry/graph-app-kit.git

3. Build

cd graph-app-kit/src/docker
sudo docker-compose build

4. Set your Graphistry visualization credentials

Get a public or private Graphistry account:

• Graphistry Hub (public, free): Create a free Graphistry Hub account using the username/password option, which you will use for API access. Visualizations will default to pointing to the public Graphistry Hub GPU servers.

• Alternatively, launch a private Graphistry server, login, and use the username/password/URL for your configurtion.

Edit src/docker/.env with:

GRAPHISTRY_USERNAME=your_username
GRAPHISTRY_PASSWORD=your_password
### OPTIONAL: Add if a private/local Graphistry server
#GRAPHISTRY_PROTOCOL=http or https
#GRAPHISTRY_SERVER=your.private-server.net

5. Start & stop

cd src/docker and then:

• Start: sudo docker-compose up -d
• Use: Go to http://localhost:8501/dashboard (or whatever the public IP)
• Stop: sudo docker-compose down -v

Graph!

You are now ready to add custom views and add integrations.