sparkrun Plugin for OpenClaw

AI-assisted inference on NVIDIA DGX Spark -- run, manage, and stop LLM workloads with OpenClaw.

What It Does

This plugin teaches OpenClaw how to use sparkrun to manage LLM inference workloads on NVIDIA DGX Spark systems. It provides:

• Skills -- Detailed reference that OpenClaw uses automatically when working with sparkrun
• sparkrun_exec Tool -- A dedicated tool for executing sparkrun CLI commands

Installation

From npm

openclaw plugins install @sparkarena/sparkrun-openclaw

From Local Directory

# Clone the repo
git clone https://github.com/spark-arena/sparkrun.git
cd sparkrun

# Install from local path
openclaw plugins install ./sparkrun-openclaw-plugin

# Or link for development
openclaw plugins install ./sparkrun-openclaw-plugin --link

Prerequisites

sparkrun CLI

The plugin requires sparkrun to be installed:

# Install via uvx (recommended)
uvx sparkrun setup install

# Or via uv
uv tool install sparkrun

DGX Spark Cluster

You need SSH access to one or more NVIDIA DGX Spark systems. The fastest way to get started:

# Interactive setup wizard (handles everything)
sparkrun setup wizard

# Or manual cluster creation
sparkrun cluster create mylab --hosts 192.168.11.13,192.168.11.14 -d "My DGX Spark lab"
sparkrun cluster set-default mylab
sparkrun setup ssh --cluster mylab

Skills (Automatic)

OpenClaw automatically uses these skills when the task context matches:

| Skill | Activates When | |------------|-----------------------------------------------------------------------------------------------------------| | run | Running, monitoring, stopping, benchmarking, tuning, or managing inference workloads and proxy | | setup | Installing sparkrun, configuring clusters, SSH setup, CX7 networking, Docker group, permissions, earlyoom | | registry | Managing recipe registries, browsing benchmark profiles, creating/editing recipes |

Usage Examples

Describe what you want in natural language -- OpenClaw will use the skills automatically:

• "Run the Qwen3 1.7B model on my cluster"
• "What inference jobs are running?"
• "Stop all inference jobs on my cluster"
• "Show me available recipes for llama models"
• "Benchmark the sglang recipe on a single node"
• "Set up sparkrun on my DGX Spark cluster"
• "Configure CX7 networking on my cluster"
• "Create a recipe for Mistral 7B on vLLM"
• "Monitor my cluster's GPU usage"
• "Start the inference proxy and load a model"
• "Check if my job is healthy"

Key Concepts

• Recipes are YAML files describing an inference workload (model, runtime, container, defaults)
• Runtimes are inference engines: vLLM, SGLang, llama.cpp, TensorRT-LLM
• Clusters are named groups of DGX Spark hosts
• Registries are git-based collections of recipes and benchmark profiles
• Benchmark profiles define standardized benchmark configurations from registries
• Proxy is a unified OpenAI-compatible gateway in front of multiple inference endpoints
• Each DGX Spark has 1 GPU, so --tp N (tensor parallelism) = N hosts
• sparkrun launches detached containers -- Ctrl+C detaches from logs, never kills the job
• Recipe names support @registry/name syntax for explicit registry selection

Links

• sparkrun Documentation
• sparkrun github repo
• Recipe Format Specification

License

Apache 2.0 License -- see LICENSE for details.