n8n on NVIDIA DGX Spark Runs AI Workflows Locally.

n8n on NVIDIA DGX Spark lets teams run AI workflows on hardware they own, without sending customer data to a cloud API. The setup can transcribe calls, read invoices, search internal documents, draft replies and generate images from one machine in the office.

The n8n blog post explains how to serve the models behind one OpenAI-compatible endpoint. Existing n8n OpenAI nodes can connect to that endpoint, so workflows do not need a different integration for each local model.

This is aimed at organisations that must keep records under their control because of contracts, regulation or professional duties. The examples include clinics, law firms and accountants. Keeping the models on owned hardware also gives teams a clear account of where sensitive data went.

DGX Spark has 128GB of unified system memory. n8n says that is enough to keep models for the listed jobs loaded at the same time, provided the models are chosen for the machine’s memory bandwidth.

The documented stack includes a 35B mixture-of-experts model served with vLLM in NVFP4. LiteLLM places one endpoint in front of the models and can issue separate keys for different teams. n8n workflows then call that local endpoint rather than a hosted model provider.

Unified memory

128GB

shared by the CPU and GPU

Language model

35B

mixture-of-experts model

Cloud API calls

0

for the documented local workflow

The guide also covers measuring what happens to chat speed when someone generates an image. That matters when one box is handling several types of work for a whole team, as image generation and chat share the same hardware.

This setup is the next step up from n8n’s self-hosted AI starter kit, which is designed to get one person running local models on a laptop. DGX Spark provides a shared office machine for a team instead. It also adds another option for regulated automation teams alongside n8n’s workflow security controls.

What is n8n on NVIDIA DGX Spark?

It is a setup for running n8n AI workflows against models hosted on an NVIDIA DGX Spark in your own premises. The workflows connect through an OpenAI-compatible endpoint instead of sending data to a cloud AI API.

Does n8n on NVIDIA DGX Spark require a cloud API?

No. The documented workflow can transcribe calls, read invoices, search documents and draft replies without sending a byte to a cloud API. The models and workflow data remain on hardware owned by the organisation.

How much does n8n on NVIDIA DGX Spark cost?

The post does not give pricing for DGX Spark or name an n8n plan requirement. It presents the setup as running on hardware the organisation owns.