n8n shows how to run AI invoice workflows on an office computer
By Automate Basics, written with AI from n8n's original
n8n has published a guide to running its AI workflows on an office-based NVIDIA DGX Spark rather than calling a cloud model API. The setup is aimed at teams that need to keep documents and recordings under their control.
What changed
n8n says its workflows can send AI tasks to models running on a NVIDIA DGX Spark in the office, instead of sending the material to a cloud model API. Its published setup puts several models behind a single LiteLLM endpoint that n8n can use through its OpenAI credential. The guide covers tasks including reading invoices, transcribing calls and answering questions from internal documents.
The example workflow accepts an invoice image, reads its contents and extracts fields such as the supplier, date and total. This is a setup guide, not an announcement that n8n now includes a local model or a ready-made appliance. A team would need to run and maintain the computer, models and connecting services itself.
Who can use n8n workflows on DGX Spark?
Teams that already build workflows in n8n and need tighter control over where their documents go are the clearest audience. n8n points to work such as handling invoices, calls and internal files, where sending records to an outside model provider may not be acceptable.
The guide uses a DGX Spark, locally hosted models and LiteLLM. It explains how to give a team its own LiteLLM key and connect that endpoint to an n8n OpenAI credential. n8n does not specify which n8n plans support this setup, nor does it give a price for the hardware or the full setup. It also does not present the arrangement as a managed service: someone at the organization must prepare and operate the local models.
How do I try a local AI workflow at work?
An invoice is a sensible first test because n8n provides an example workflow that takes an uploaded image, reads it locally and returns extracted fields. Start with a document your team is permitted to process, then compare every returned field with the original before using the result in a filing or payment process.
n8n reports that its OCR test dropped a digit from an identification field, even though it read other parts of the sample correctly. That makes identifiers and totals worth checking closely. The example accepts PNG and JPEG images, so a PDF would need conversion first. Confirm that every service in the workflow points to the local setup before treating it as a workflow that avoids cloud model calls.
Frequently asked questions
Does n8n now include a local AI model?
No. n8n published instructions for connecting its workflows to models hosted on a NVIDIA DGX Spark. The setup also uses LiteLLM to provide an endpoint that n8n can call. A team must provide and maintain the hardware and model services.
Can the example process PDF invoices?
The invoice example accepts PNG and JPEG images, according to n8n. A PDF invoice must be converted to an image before that example can process it. Teams should check the extracted fields against their own documents, especially identifiers, before filing the results.
Does the setup send invoice data to a cloud model API?
n8n describes an invoice workflow that calls models on the DGX Spark rather than a cloud model API. That depends on configuring every model call to use the local services. Teams should inspect their own workflow connections before relying on it for sensitive records.
For you
Learn to do this with AI, free
Short courses for people who aren't engineers: ChatGPT, Claude, prompting and automation. Every lesson is free, with no card.
Start a free courseFor companies
Make an AI or automation tool?
List it in the directory for $9.99, refunded in full if it isn't approved, or claim its page if it is already here.
Related reading
Laplace uses AI agents to stress-test robots and self-driving carsLaplace Intelligence uses AI agents to create difficult situations for robots and self-driving vehicles inside interactive simulations. The product is aimed at teams that need to find failures before deploying physical AI.
Nvidia launches open models for workplace agents and local PCsNvidia has introduced an enterprise agent model and smaller models for AI PCs and workstations. The open-weight models give organizations options for running agents on their own infrastructure or locally, according to TechCrunch.
Google brings Gemini data analysis to Sheets on iOSGoogle says eligible users can now use Gemini in Google Sheets on iOS to ask questions about spreadsheet data, get analytical insights and create charts. The feature is rolling out gradually.
Tools in this piece
Written with AI from n8n's original and published after automatic checks: every figure here appears in the original, and no sentence is copied from it. The picture is AI-generated. The original is the authority.
Get the next one by email
One email, most weeks: what changed in AI for people who use it at work, and what to do about it.
Source: Run your n8n AI workflows on NVIDIA DGX Spark, with no cloud API, n8n, 19 Oct 2026.