Run Local AI Agents with Faster Models and Multi-Node Clustering on NVIDIA DGX Spark

2026-06-01 · Source: NVIDIA Technical Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Cloud Computing & IT Infrastructure · Depth: Intermediate, medium

Summary

NVIDIA announced significant updates for its DGX Spark system at Computex 2026, aiming to simplify local AI agent development and deployment. These enhancements include a streamlined NemoClaw installation path, enabling developers to run autonomous agents in minutes after initial model download. NemoClaw, an open-source blueprint, integrates open models, an agent harness like OpenClaw, and the NVIDIA OpenShell runtime for secure, sandboxed execution. Furthermore, performance improvements for agentic models like Qwen3.6-35B on vLLM, utilizing NVIDIA's NVFP4 quantized checkpoint and MTP optimizations, deliver up to 2.6x faster inference. For scaling beyond a single device, NVIDIA Sync now features a cluster assistant that automates connecting two to four DGX Spark units, providing up to 512 GB of unified memory for large MoE models or multi-agent pipelines.

Key takeaway

For AI Engineers and MLOps teams building autonomous agents with strict privacy or performance needs, NVIDIA's DGX Spark updates significantly streamline your workflow. You can now deploy local AI agents in minutes using the simplified NemoClaw installation, benefiting from up to 2.6x faster inference with models like Qwen3.6-35B. If your projects demand greater compute, utilize NVIDIA Sync's cluster assistant to easily scale your DGX Spark units, providing ample memory for larger models and multi-agent pipelines.

Key insights

Integrated hardware and software solutions significantly accelerate local AI agent development and deployment, enhancing security and scalability.

Principles

• Local agent execution improves security and privacy by keeping context on-device.
• Simplified installation workflows drastically cut setup time for AI agent systems.
• Automated multi-node clustering enables scaling local compute for large models.

Method

Install NemoClaw on DGX Spark via a single `curl` command, followed by an interactive wizard to accept licenses, perform express install, and automatically download Qwen3.6-35B and set up Ollama.

In practice

• Access agent WebUI using `nemoclaw gateway-token --quiet` for interaction.
• Customize agent behavior by editing system prompts and OpenShell network policies.
• Use `nemoclaw onboard --fresh --gpu` to create new sandboxes with different models.

Topics

• AI Agents
• NVIDIA DGX Spark
• NemoClaw
• Local AI
• Multi-node Clustering
• Inference Optimization
• Qwen3.6-35B

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• NVIDIA DGX Spark — The NVIDIA DGX Spark offers 1 petaFLOP performance for local LLM prototyping via Ollama.
• Deploy Self-Evolving Agents for Faster, More Secure Research with a Hermes Agent and NVIDIA NemoClaw — Self-evolving AI agents can securely synthesize public and private data for research, learning new skills that persist across deployments.
• Are massive LLM API costs crippling your OpenClaw? The new shift is toward local, agentic AI, and the combination of Google Gemma 4 and NVIDIA GPUs is changing the economics and performance of AI development. — Local, agentic AI powered by Google Gemma 4 and NVIDIA GPUs offers cost-free, high-performance, and secure generative AI.
• Hermes Unlocks Self-Improving AI Agents, Powered by NVIDIA RTX PCs and DGX Spark — Hermes Agent and Qwen 3.6 LLMs offer a powerful, reliable, and efficient local AI agent solution on NVIDIA hardware.
• MiniMax M2.7 Advances Scalable Agentic Workflows on NVIDIA Platforms for Complex AI Applications — MiniMax M2.7, a 230B-parameter MoE model, offers high efficiency and performance for agentic tasks via NVIDIA optimizations.
• Build Your Own AI Assistant with Hugging Face on NVIDIA DGX Spark — A personal AI assistant can integrate local and cloud models for privacy and physical interaction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA Technical Blog.