Use n8n with open-source models hosted in the EU

n8n is a workflow automation platform that you can self-host or use as a cloud service. Workflows connect triggers, hundreds of app integrations and code steps, and its AI nodes, built on LangChain, add agents with tools and memory, chains, document loaders and vector stores. Teams use it to automate support triage, document processing, reporting and internal assistants.

Language models are attached to AI nodes as Chat Model sub-nodes. The OpenAI credential accepts a custom Base URL, so the OpenAI Chat Model sub-node, and other nodes that use the same credential, can send their requests to inference.eu instead.

Create an OpenAI credential with your inference.eu key as API Key and the inference.eu base URL, ending in /v1, as Base URL. When n8n tests the credential, it calls /models at that address. Every node using the credential then sends requests to inference.eu.

In the OpenAI Chat Model sub-node, select the credential and pick a model from the list, or switch to By ID and type it. The node also has a Base URL option of its own, which overrides the credential for that node only. Keep Use Responses API off so the node uses Chat Completions. Self-hosted administrators can create the credential from the command line with n8n import:credentials.

AI Agent nodes call tools in a loop, so they need a model with dependable tool calling: kimi-k3 and glm-5.3 are strong defaults for agents that look up records, call APIs or update systems. For classification, extraction and short text generation inside a workflow, gpt-oss-120b keeps each execution quick.

Use glm-5.3 or kimi-k3 when a step summarises long documents or emails. For workflows that generate or review code, qwen3-coder-480b-a35b-instruct is a good fit. Each Chat Model sub-node has its own model setting, so one workflow can use a fast model for routing and a stronger one for the main agent.

If the credential test fails with 401, the key is wrong or has extra spaces. A 404 usually means Base URL is missing /v1 or has an extra path. If the model list is empty, set the model By ID instead.

n8n warns that models behind a custom Base URL may not support every feature. If the AI Agent returns tool calls as text or loops, choose a model with stronger tool calling. If you see errors about built-in tools or conversation IDs, Use Responses API is switched on; turn it off. Remember that sub-nodes resolve expressions only for the first input item.

With the credential pointed at inference.eu, the prompts, records and tool results your workflows send to the model are processed in EU data centres and never used for training. We sign a data processing agreement under Article 28 GDPR, publish our subprocessor list and bill in euros.

n8n stores workflow executions, including inputs and outputs, in its own database. Self-host n8n in the EU for full control over that data, or check the hosting region of your n8n plan. Self-hosted instances can turn off usage diagnostics with N8N_DIAGNOSTICS_ENABLED=false. Other nodes in a workflow send data to their own services.