DGX Spark Playbooks

Collection of step-by-step playbooks for setting up AI/ML workloads on NVIDIA DGX Spark devices with Blackwell architecture.

About

These playbooks provide detailed instructions for:

• Installing and configuring popular AI frameworks
• Running inference with optimized models
• Setting up development environments
• Connecting and managing your DGX Spark device

Each playbook includes prerequisites, step-by-step instructions, troubleshooting guidance, and example code.

Use as Claude Code Skills

Every playbook in this repo is also exposed as a Claude Code skill, so Claude can install and configure them for you interactively. An index skill (dgx-spark) routes broad questions to the right specific playbook, and each leaf skill carries cross-references to related playbooks (prerequisites, alternatives, what to try next).

Option A: install via Claude Code plugin marketplace (recommended)

In Claude Code, run:

/plugin marketplace add jkneen/dgx-spark-playbooks
/plugin install dgx-spark-playbooks@dgx-spark-playbooks

Option B: install locally via script

Clone and run the install script. This symlinks each skill into ~/.claude/skills/, so git pull updates them in place.

git clone https://github.com/jkneen/dgx-spark-playbooks
cd dgx-spark-playbooks
./scripts/install.sh                  # individual skills → ~/.claude/skills/
./scripts/install.sh --plugin         # or: whole repo as a plugin → ~/.claude/plugins/
./scripts/uninstall.sh                # remove

Requires Node 18+ if you want to regenerate skills from overrides; otherwise the committed skills/ directory is used as-is.

Customizing a skill

Hand-curate any skill by creating or editing overrides/<playbook-name>.md (see overrides/ollama.md as a worked example), then run node scripts/generate.mjs to rebuild. Overrides contribute the frontmatter description (which controls when the skill triggers) and any extra sections like "Related skills" or gotchas. Generated content between <!-- GENERATED:BEGIN --> and <!-- GENERATED:END --> markers is rewritten from the upstream README on every regeneration; override content outside those markers is preserved.

A GitHub Action auto-regenerates skills/ whenever a playbook README or override changes.

Available Playbooks

NVIDIA

• Comfy UI
• Connect Three DGX Spark in a Ring Topology
• Set Up Local Network Access
• Connect Two Sparks
• CUDA-X Data Science
• DGX Dashboard
• FLUX.1 Dreambooth LoRA Fine-tuning
• Install and Use Isaac Sim and Isaac Lab
• Optimized JAX
• Live VLM WebUI
• Run models with llama.cpp on DGX Spark
• LLaMA Factory
• LM Studio on DGX Spark
• Build and Deploy a Multi-Agent Chatbot
• Multi-modal Inference
• Connect Multiple DGX Spark through a Switch
• NCCL for Two Sparks
• Fine-tune with NeMo
• NemoClaw with Nemotron 3 Super and Telegram on DGX Spark
• Nemotron-3-Nano with llama.cpp
• NIM on Spark
• NVFP4 Quantization
• Ollama
• Open WebUI with Ollama
• OpenClaw 🦞
• Secure Long Running AI Agents with OpenShell on DGX Spark
• Portfolio Optimization
• Fine-tune with Pytorch
• RAG Application in AI Workbench
• SGLang for Inference
• Single-cell RNA Sequencing
• Spark & Reachy Photo Booth
• Speculative Decoding
• Set up Tailscale on Your Spark
• TRT LLM for Inference
• Text to Knowledge Graph on DGX Spark
• Unsloth on DGX Spark
• Vibe Coding in VS Code
• vLLM for Inference
• VS Code
• Build a Video Search and Summarization (VSS) Agent

Resources

• Documentation: https://www.nvidia.com/en-us/products/workstations/dgx-spark/
• Developer Forum: https://forums.developer.nvidia.com/c/accelerated-computing/dgx-spark-gb10
• Terms of Service: https://assets.ngc.nvidia.com/products/api-catalog/legal/NVIDIA%20API%20Trial%20Terms%20of%20Service.pdf

License

See:

• LICENSE for licensing information.
• LICENSE-3rd-party for third-party licensing information.