n8n is an open-source workflow builder that lets a small business wire AI into real processes - leads, invoices, documents - without a custom backend, and it is the usual pick when customer data cannot leave the country. The news: n8n now runs those AI workflows on an NVIDIA DGX Spark in the office, with no cloud API call.
The n8n team published a full build: one DGX Spark, 128GB of unified memory, and a stack of local models behind a single OpenAI-compatible endpoint that n8n's OpenAI nodes already speak. Qwen3.6-35B in NVFP4 handles chat, extraction and vision at around 120 tokens per second per user. Whisper transcribes calls, GLM-OCR reads invoices, Qwen embeddings and reranker drive document search. Every model in the stack has a commercial license.
What it means for small business automation
This closes the main blocker for clinics, accountants and law firms: rules that forbid sending client data to a foreign cloud API. One box covers call transcription, invoice OCR and Q&A over internal archives - the exact jobs that keep teams in copy-paste. If you were weighing a legal archive audit flow like the one in this legal archive AI assistant, the DGX Spark path removes the data-cannot-leave-our-server objection. The hardware is not cheap, but it replaces an open-ended API bill.
Why it matters: on-prem AI for a whole team is now a real option on one appliance, which changes the answer when a small business asks whether privacy-sensitive automation is feasible at all.
Source: n8n Blog