NVIDIA DGX SPARK

Cluster your Spark.
Skip the science project.

Orion turns your Spark into shared compute that looks after itself. One command to install, one click to launch a workload, and agents that keep it running.

Orion Apps

×

+

←

→

↻

spark.your-lab.edu/appstore

☆

⋮

Spark Kit in the Orion Apps store on a DGX Spark with model servers and interfaces

Orion Apps

×

+

spark.your-lab.edu/appstore

⋮

Spark Kit in the Orion Apps store on a DGX Spark with model servers and interfaces

Orion Apps

×

+

←

→

↻

spark.your-lab.edu/appstore

☆

⋮

Spark Kit in the Orion Apps store on a DGX Spark with model servers and interfaces

ONE-COMMAND INSTALL

Open the terminal once.

Paste one command on a fresh Spark. When it finishes you get a login page, and nobody on your team touches a shell again.

bash
$curl -sL "$(curl -s https://api.github.com/repos/juno-fx/Juno-Bootstrap/releases/latest | grep browser_download_url | grep orion-install-helper | cut -d '"' -f 4)" | bash -

The Spark Kit puts eleven workloads on your desk.

Model servers, interfaces, and pipelines pre-configured for Spark’s unified memory. Currently, the kit is privately maintained, outside Juno’s official Terra plugins.

Orion Apps×
+
←→↻
spark.your-lab.edu/appstore☆
⋮
Spark Kit in the Orion Apps store on a DGX Spark with model servers and interfaces

Nemotron and Qwen on vLLM, Open WebUI, ComfyUI, and txt2kg.

Orion Apps×
+
←→↻
spark.your-lab.edu/appstore☆
⋮
Spark Kit in the Orion Apps store on a DGX Spark with model servers and interfaces

Nemotron and Qwen on vLLM, Open WebUI, ComfyUI, and txt2kg.

Orion Apps×
+
←→↻
spark.your-lab.edu/appstore☆
⋮
Spark Kit in the Orion Apps store on a DGX Spark with model servers and interfaces

Nemotron and Qwen on vLLM, Open WebUI, ComfyUI, and txt2kg.

What Orion adds to your Spark.

Agentic workloads, native

Hibernate a workload, swap in a different model, and resume where you left off.

Agentic workloads, native

Hibernate a workload, swap in a different model, and resume where you left off.

Agentic workloads, native

Hibernate a workload, swap in a different model, and resume where you left off.

Put the Spark’s hardware to work

Spark ships with ConnectX networking and unified memory. Orion uses both, and joins Sparks with your other machines.

Put the Spark’s hardware to work

Spark ships with ConnectX networking and unified memory. Orion uses both, and joins Sparks with your other machines.

Put the Spark’s hardware to work

Spark ships with ConnectX networking and unified memory. Orion uses both, and joins Sparks with your other machines.

Jupyter and VS Code, one click

A running notebook or IDE in 60 to 90 seconds.

Jupyter and VS Code, one click

A running notebook or IDE in 60 to 90 seconds.

Jupyter and VS Code, one click

A running notebook or IDE in 60 to 90 seconds.

One Spark, many users

Unified memory shared across concurrent workloads through time slicing.

One Spark, many users

Unified memory shared across concurrent workloads through time slicing.

One Spark, many users

Unified memory shared across concurrent workloads through time slicing.

Survive a node loss

Batch and service workloads reschedule across the rest of the pool.

Survive a node loss

Batch and service workloads reschedule across the rest of the pool.

Survive a node loss

Batch and service workloads reschedule across the rest of the pool.

Quotas and priority

Give each person or project a share of the Spark, and let priority decide who goes first.

Quotas and priority

Give each person or project a share of the Spark, and let priority decide who goes first.

Quotas and priority

Give each person or project a share of the Spark, and let priority decide who goes first.

AGENTS ON SPARK

Agents that act on infrastructure, safely.

An orchestrator agent takes the job, hands pieces to worker agents in their own workspaces, and collects their results. Agents call validated APIs, never the cluster.

Deploy

Agents launch workloads through the scheduling MCP.

Self-heal

Agents watch the cluster and repair drift.

Build

Agents create, test, and deploy new plugins.

hermes-agent

×

+

←

→

↻

spark.your-lab.edu/session/hermes-agent

☆

⋮

Hermes Agent session in a terminal: available tools and skills

hermes-agent

×

+

spark.your-lab.edu/session/hermes-agent

⋮

Hermes Agent session in a terminal: available tools and skills

hermes-agent

×

+

←

→

↻

spark.your-lab.edu/session/hermes-agent

☆

⋮

Hermes Agent session in a terminal: available tools and skills

Whether you run one Spark on a desk or cluster a dozen for a lab, Orion turns DGX Spark into shared compute your whole team can use.

ORION FOR NVIDIA DGX SPARK · NVIDIA INCEPTION MEMBER

2:1

GPU density from existing hardware, typically

60-90s

to a running notebook or IDE, from a pre‑pulled image

Questions about Spark

Cluster your Sparks.

Book a demo to see Orion on your Sparks, or start a Proof of Value on your own hardware.

Cluster your Sparks.

Book a demo to see Orion on your Sparks, or start a Proof of Value on your own hardware.

Cluster your Sparks.

Book a demo to see Orion on your Sparks, or start a Proof of Value on your own hardware.