How to run n8n AI agents on EU-hosted models (2026)

If you build AI agents in n8n, the model choice is a sub-node: OpenAI Chat Model, Anthropic Chat Model, Mistral, Gemini, or the OpenRouter Chat Model node when you want many models behind one key. That works until someone asks where the inference is running, what all the workflows on the instance are spending, or why trying a different model means swapping nodes and credentials. The n8n community has asked for an EU AI gateway node for exactly this reason.

Opper ships a verified community node for n8n. It is a native chat-model sub-node, so it plugs into the AI Agent and LLM Chain nodes like any built-in provider. Behind it sits the Opper gateway with the whole catalog: Claude, GPT, Gemini, Mistral, and EU-hosted open-weight models, all on one OpenAI-compatible credential.

Here is the setup in under three minutes, including how to restrict the model dropdown to EU-hosted models:

Why route n8n AI agents through a gateway

Pointing your n8n instance at Opper instead of at each provider buys you a few things:

• One credential, every model. Add the Opper credential once, and every workflow on the instance can pick from the full catalog. Switching models is a dropdown change, not a rebuild.
• EU-hosted inference. The catalog tells you where each model is served. You can pick EU routes for Claude, GPT, Mistral, or open-weight models, and you can lock the whole instance to EU so nobody picks anything else by mistake.
• Spend and visibility. Every call is traced with token usage and cost, and you can set spend limits per project. n8n instances tend to run many workflows against one key, so this is where you find out which one is expensive.

One thing worth separating: where n8n runs and where the model runs are two different hops. Self-hosting n8n in Europe, or using n8n Cloud, which is hosted in Frankfurt, covers the workflow engine. The prompt still leaves for whichever model endpoint the chat-model node points at. This post is about that second hop.

Install the Opper Chat Model node in n8n

You need n8n 1.94.0 or later. Only instance owners and admins can install community nodes; once installed, every member can use it.

1. Install the node

Open the nodes panel on the canvas with + or by pressing n, and search for Opper. Verified community nodes appear under More from the community at the bottom of the panel. Select Opper Chat Model, then Install.

If you prefer installing by package name, go to Settings → Community Nodes and add @opperai/n8n-nodes-opper. n8n loads community nodes at startup, so restart the instance if the node does not appear right away. On n8n Cloud, verified community nodes can be hidden per instance, and the owner toggles them in the Cloud Admin Panel.

2. Add the credential

Create an Opper API credential with two fields:

The credential test calls Opper's OpenAI-compatible GET /models endpoint. If it fails, the key is wrong or the base URL has an extra path segment.

3. Attach it to the AI Agent node

Add an AI Agent node, a Basic LLM Chain, or anything else that accepts a language-model connection, and attach Opper Chat Model as its model sub-node.

Which models the n8n AI Agent node can use

The Opper Chat Model node fetches the catalog live with your credential, so the dropdown always shows what your account can use. The AI Agent node needs tool calling, so pick a model the catalog marks with the tools capability. Some EU-hosted options that work well with agents:

The full list is in the model directory. Each model page shows where it is served, who runs the datacenter, and what the provider's data-handling terms say.

Under Options you can set temperature, top-p, penalties, max tokens, timeout, retries, and a JSON response format. Changing model later is the same dropdown, and the rest of the workflow stays as it is.

Lock every n8n workflow to EU-hosted models

Because the node reads the catalog through your credential, a model access rule in Opper changes what n8n users see. Set an organization allowlist with Regions set to EU, and the dropdown in every workflow shrinks to EU-hosted models. You can also allow specific providers, block others, or narrow access further for one project.

The rule is enforced at the gateway, not just in the UI. A call to a model outside the allowlist returns an error that names the model and the rule, rather than silently switching to something else. That is the behaviour you want when the reason for the rule is a data-residency commitment.

Opper vs the OpenRouter Chat Model node

The built-in OpenRouter node is the closest comparison, since both give you many models behind one key. The difference is where the traffic goes. OpenRouter's EU in-region routing is enterprise-only, arranged by request, so on the standard product the request path is not kept within Europe, and each upstream provider keeps its own logging and retention policy. Opper's control plane and traces run in AWS Stockholm, the catalog shows the serving region and data-handling terms per route, and a model access rule pins the whole instance to EU routes. We compare the options in more depth in OpenRouter alternatives and the best European AI gateways.

Test the agent

Send a message through the chat trigger or run the workflow. The response comes back through the Opper node, and the call shows up in the Opper dashboard with model, latency, tokens, and cost. If you have a lot of workflows on one instance, tag them so you can tell them apart: the usage attribution guide shows how to put tags in the base URL, which is the one field the n8n credential gives you.

n8n and Opper FAQ

Get started

Sign up, install the node, and add the credential. The integration docs cover the troubleshooting cases, including hidden community nodes on n8n Cloud. Questions? Find us on Discord or drop us a line at hello@opper.ai.