Ten n8n workflows that run on a local Ollama model. No cloud service, no API keys, no bill after setup. Import a JSON file, point it at your own machine, done.
MIT licensed. Nothing here is for sale.
git clone https://github.com/dankofly/selfhosted-ai-kit.git
cd selfhosted-ai-kit
docker compose up -d
docker exec -it ollama ollama pull llama3.1Then open n8n on localhost:5678 and import any file from workflows/.
You are not the first to do this. A free bundle of 11 n8n + Ollama templates went up on the n8n community forum on 2026-03-24, and n8n ships its own self-hosted-ai-starter-kit. Look at both. If they fit you better, use them.
What this repo adds is business tasks instead of demos (invoice extraction, BANT lead scoring, CRM enrichment, competitor price monitoring) and a GDPR checklist that is actually written down.
These files were built on 2026-03-26 and have not been retested since. If a newer n8n breaks an import, open an issue and I will fix it. That is more useful than a compatibility guarantee I cannot back up today.
Nobody has downloaded this yet. No star count to point at, no testimonials, no logos. There is nothing to show and I am not going to invent it.
| File | Nodes | What it does |
|---|---|---|
01-email-classifier.json |
8 | Sorts inbound mail into 6 buckets, answers the urgent ones |
02-document-summarizer.json |
6 | PDF or text in, structured summary and action items out |
03-sentiment-analyzer.json |
7 | Flags angry and urgent feedback, pings the team |
04-content-generator.json |
5 | Posts, threads, captions, drafted on your own model |
05-meeting-notes-extractor.json |
5 | Raw notes in, decisions owners deadlines out |
06-lead-qualifier.json |
6 | BANT scoring, hot-lead alert, talking points |
07-invoice-extractor.json |
6 | Invoice PDF to vendor, items, totals, tax, IBAN |
08-internal-faq-chatbot.json |
6 | Staff FAQ bot with no seat fees attached |
09-competitor-monitor.json |
6 | Scheduled price checks, analysis written locally |
10-crm-enrichment.json |
5 | Persona, seniority and industry from thin contact data |
Three containers in one docker-compose.yml:
- postgres
:16-alpine— where n8n keeps its state - ollama
:latest— where the model runs, port11434 - n8n
:latest— where the workflows run, port5678
A GPU is optional. The NVIDIA block ships commented out in the compose file; uncomment it if you have the card.
| Cloud AI | This kit | |
|---|---|---|
| API keys required | yes | 0 |
| Calls to a model vendor | every run | 0 |
| Recurring cost after setup | per token | 0 |
| Where your invoices and emails go | their servers | your box |
guides/SETUP.md— step by step, written for a first installguides/MODELS.md— which Ollama model belongs on which workflowguides/GDPR-CHECKLIST.md— for operators inside the EU
Each also ships as a PDF in the same folder.
Issues and pull requests are welcome, particularly compatibility fixes and new workflows in the same spirit: a real task, no external API, importable in one click.
I write about AI compliance and marketing. This kit is how I meet people who care where their data runs. Take the files and go, or leave an address at kit.hypeakz.io and you get new workflows when they land. Both are fine.
MIT licence. Daniel Kofler, HYPEAKZ.IO, Austria.