Get network access to your Spark

You want to drive the Spark from your normal workstation instead of sitting at it with a monitor and keyboard.

Local network (SSH)

DGX OS ships NVIDIA Sync, which configures SSH access over your local network. Follow the official Set Up Local Network Access playbook (about 5 minutes), then verify:

ssh your-user@your-spark-hostname.local

If .local resolution does not work on your network, use the Spark’s IP address. Find it on the Spark with:

ip -4 addr show | grep -v 127.0.0.1

Remote access (Tailscale)

To reach the Spark from anywhere without exposing it to the public internet, put it on your Tailscale tailnet. NVIDIA provides a Set up Tailscale on Your Spark playbook. Once joined, the Spark is reachable by its tailnet name from any other device you own, with no port forwarding.

Tip

Tailscale is the clean way to use the Spark as a personal inference backend from a laptop that moves between networks. Pair it with the run a local LLM how-to to expose an OpenAI-compatible endpoint only your devices can reach.

Reverse proxy (optional)

If you run a reverse proxy such as Caddy, you can front the Spark’s services with real hostnames and automatic TLS rather than remembering ports. Point the proxy at the Spark’s inference port (for example Ollama’s 11434) and let it handle the certificate.