Use Cursor with open-source models hosted in the EU

Cursor is an AI code editor built on VS Code, made by Anysphere. It combines Tab completion, a chat panel, inline edits and an Agent mode that plans changes across files and runs terminal commands. By default it uses models that Cursor provides under its own subscription.

Cursor also accepts your own API keys, and for OpenAI-format providers it offers a base URL override. That is how you connect a custom OpenAI-compatible endpoint such as inference.eu and use open-weight models in Chat and Agent.

In Settings, Models, the OpenAI API Key field takes your inference.eu key and Override OpenAI Base URL takes our base URL. Then add custom models by id. Cursor sends requests for those models in the OpenAI Chat Completions format to the overridden URL.

Two details matter. First, the override is global: while it is on, OpenAI-family requests go to your endpoint, so turn it off when you want to use Cursor's own models. Second, on Business and Enterprise teams an admin must allow personal API keys under Team Settings, Models, or custom models are refused. Avoid model names that start with claude-, which Cursor treats as Anthropic models.

Agent mode makes many tool calls in a row, so use models that are reliable with tools. qwen3-coder-480b-a35b-instruct is a strong default for edits, refactors and tests. glm-5.3 and kimi-k3 handle longer plans and explanations of unfamiliar code well.

For quick questions in Chat, gpt-oss-120b responds fast. You can add several custom models and switch between them in the model picker. Cursor's Tab completion is not affected by any of this: it always runs on Cursor's own completion model and cannot be pointed at a custom endpoint.

"Invalid API key" or a 401 means the key is wrong or Cursor is still sending your inference.eu key to OpenAI because the override is off. "Model not found" or "This model does not support custom API keys" usually means the custom model name differs from the inference.eu id, or your team admin has not enabled personal OpenAI keys.

If Cursor's own models stop working after setup, that is the global override at work; switch it off to use them. If the Settings fields do not take focus in some 3.15 builds, update Cursor. When Agent describes changes without applying them, the model is not returning tool calls; switch to one of the coding models above.

Be clear about the path. Cursor routes every request through its own servers for final prompt building, including requests that use your key and custom base URL. Your code therefore passes through Cursor's backend before it reaches inference.eu. Cursor's zero data retention policy does not apply to requests made with your own keys. Tab completion and other features keep using Cursor's own models.

What inference.eu covers is the model step: requests we receive are processed in EU data centres, never used for training, and covered by our Article 28 GDPR data processing agreement and subprocessor list, billed in euros. If no traffic may pass a US service, choose an editor that calls the endpoint directly.