n8n + Ollama: Private AI Automation in 5 Steps (2026)

If you want private AI automation without paying per prompt, n8n plus Ollama is one of the simplest stacks to start with in 2026. n8n handles triggers, branching, and app-to-app workflow logic. Ollama runs the model locally on your own machine or server. Put them together and you get automations that can summarize tickets, draft replies, classify leads, or answer internal questions without shipping every prompt to a cloud API.

The catch is pretty obvious: this is not a beginner-friendly one-click stack. n8n’s self-hosted Community edition is free, Ollama is free to run locally, but you still pay in hardware, setup time, and maintenance. If you want the fastest polished chat app experience, tools like Dify with Ollama or Flowise with Ollama can be a better fit. If your real job is automation across forms, email, webhooks, Slack, CRMs, and databases, n8n is usually the better bet.

Who this setup is for

This tutorial fits operators, founders, developers, and small internal tools teams who want to automate tasks around sensitive data or just cut API spend. It makes sense when you need workflow control first and AI second.

• Good fit: ticket triage, lead qualification, document summaries, internal assistants, webhook-based bots, scheduled AI jobs
• Bad fit: teams that want a polished end-user AI product with prompt management and knowledge-base UX out of the box
• Hardware reality: lightweight models are easy enough, but bigger local models will demand real RAM or GPU headroom

Bottom line: use n8n when AI is one step inside a larger business process. Use Dify or Flowise when the AI interface itself is the product.

What you need before you start

You need three things in place:

• n8n running in cloud or self-hosted mode. n8n’s docs say self-hosted installs run as the free Community edition unless you add a Business or Enterprise license key.
• Ollama installed on the same machine or on a reachable host. Ollama exposes its local API on port 11434 by default.
• A model already pulled into Ollama, such as a smaller Llama, Gemma, or DeepSeek variant that your hardware can actually run.

If you are self-hosting, Docker is the easiest path. n8n’s own documentation also pushes new users toward its self-hosted AI starter kit for proof-of-concept work, not full production hardening. That distinction matters more than most tutorials admit.

How n8n and Ollama fit together

Here is the simple version. n8n does not replace Ollama, and Ollama does not replace n8n.

• n8n: receives events, moves data, branches logic, retries failed steps, and sends results somewhere useful
• Ollama: runs the actual model locally and returns completions through an API

So a basic workflow looks like this: a webhook, chat message, schedule, or app event hits n8n; n8n sends the prompt to Ollama through the Ollama Chat Model node; the model responds; n8n routes that output to Slack, email, a CRM, a database, or another app.

That separation is why this stack is attractive. You keep workflow logic in one place and inference in another. It is also why the stack can feel brittle when networking is wrong.

Step 1: Make sure Ollama is reachable

Before you open n8n, check the boring part first: connectivity. Most failed setups are not model problems. They are network problems.

If Ollama and n8n are both running directly on the same host, the default local API address usually works. If one or both are inside Docker, you need to be more careful. n8n’s official troubleshooting docs call out a few common patterns:

• If only Ollama is in Docker, publish port 11434 and point n8n to the host address.
• If only n8n is in Docker, use host.docker.internal instead of localhost when pointing to Ollama.
• If both are in separate containers, put them on the same Docker network and use the Ollama container name as the host.
• If you hit ECONNREFUSED ::1:11434, switch from localhost to 127.0.0.1 to avoid an IPv6 mismatch.

This is one place where competing tutorials are a bit too cheerful. The stack works well once the address is right. Before that, it feels broken for no obvious reason.

Step 2: Add the Ollama Chat Model node in n8n

Inside n8n, create a new workflow and add the Ollama Chat Model node. The official n8n docs position this node for local conversational and agent-style use cases. You then create credentials with the base URL for your Ollama instance.

At this point, pick a model that matches your hardware. A smaller local model is the sane way to start. You can always move up later. If the model list is empty, that usually means one of two things: the connection is wrong, or the model is not installed in Ollama yet.

Keep temperature conservative for first tests. Fancy creativity settings are fun, but they make debugging harder. For classification, summarization, routing, and extraction, predictable beats clever.

Step 3: Build one useful workflow first

Do not begin with an “AI agent for everything” diagram. Start with a narrow workflow you can verify in a few minutes. A good first project is an inbound support triage flow:

• A form, webhook, or chat message triggers n8n
• n8n passes the message to Ollama with a short system prompt
• The model returns a category, summary, and urgency label
• n8n writes the result to your help desk, sends Slack alerts for urgent items, and stores the raw request in a sheet or database

That kind of workflow shows the real strength of n8n. The model does one bounded piece of reasoning. n8n handles the operational plumbing around it.

If you want a more UI-driven local setup for document chat, Aixcove already has guides for AnythingLLM with Ollama and Open WebUI with Ollama. Those are better picks when chat UX matters more than workflow orchestration.

Step 4: Write prompts for structure, not style

Here is the mistake people make: they treat local models like magic. Smaller Ollama-hosted models can work surprisingly well, but only if the task is tightly framed.

Ask for structured output. Tell the model exactly what fields to return. Keep the job narrow. For example, ask it to return category, priority, and summary instead of “analyze this message.” That single change makes downstream automation much more reliable.

The thing is, n8n shines when every later step has something clean to act on. If your model response rambles, the rest of the workflow becomes fragile fast.

Step 5: Know the limits before you scale it

This stack is useful, but it is not frictionless.

• Local model quality varies. If you pick a model that is too small, your classifications and summaries may drift.
• Throughput is limited by your hardware. One laptop-grade machine is fine for experiments, not for heavy concurrent workloads.
• Self-hosting adds ops work. n8n’s own docs warn that self-hosting assumes comfort with servers, containers, configuration, security, and scaling.
• n8n is not the best front-end AI app builder. If you want shareable chat apps, prompt versioning, or knowledge-base UX, compare it with Dify vs Flowise before you commit.

That is also why the search intent here matters. People looking up n8n with Ollama are usually not asking “what is AI automation.” They are asking whether this stack is practical, private, and cheap enough to justify the setup. The answer is yes, with a big asterisk around maintenance.

When n8n with Ollama is worth it

It is worth it when you need private inference inside real automations and you are comfortable owning the infrastructure. It is especially attractive for internal operations work: routing incoming requests, enriching records, summarizing documents, or powering assistants that touch company data.

It is less compelling if you mainly want a cleaner AI app interface. In that case, compare purpose-built tools instead of forcing n8n into that role. Aixcove’s Open WebUI vs AnythingLLM comparison is a better branch to explore.

Final verdict

n8n with Ollama is a smart stack for private AI automation, not a universal AI platform. That is exactly why it works. n8n handles the messy real-world logic. Ollama handles local inference. Keep those roles clean, start with one small workflow, and you can get real value without piling cloud model costs onto every task.

If you try to turn it into a full chat product, a production knowledge platform, and an integration hub all at once, you will spend more time debugging than automating. Start smaller. The stack rewards that approach.

Sources: official n8n hosting, pricing, and Ollama node documentation; official Ollama quickstart; competitor tutorial coverage from Hostinger, DEV Community, and n8n workflow templates.

When this n8n and Ollama setup makes sense

This workflow is best when you want automation that can touch private notes, internal documents, or local tools without sending every prompt to a cloud model. Use it for summarizing internal files, routing support notes, drafting routine updates, or adding a local AI step to an existing n8n automation.

It is not the fastest path for non-technical teams. If you want a visual AI app builder with hosted deployment, compare this setup with Dify. If your main goal is private document chat instead of automation, start with AnythingLLM.

Quick comparison: n8n plus Ollama vs alternatives

• Choose n8n plus Ollama if privacy, self-hosting, and workflow control matter more than a polished AI app interface.
• Choose Dify if you want to build a user-facing AI assistant or RAG app with less workflow plumbing.
• Choose Flowise if you prefer visual LLM chains and prototypes over general automation.
• Choose AnythingLLM if your core task is chatting with private files rather than triggering multi-step workflows.

FAQ

Can n8n use Ollama without an internet connection?

Yes, the model call can stay local if n8n and Ollama run on the same machine or private network. External app triggers, cloud storage, and third-party integrations may still require internet access.

Is this better than using OpenAI or Claude inside n8n?

It depends on the job. Local models reduce data exposure and recurring API costs, but cloud models are usually stronger for complex reasoning, long context, and production support.

What should I build first?

Start with a small workflow: receive text, send it to Ollama, return a summary, then log the result. Once that is stable, add branching, retries, and storage.