Use Continue with open-source models hosted in the EU

Continue is an open-source extension for VS Code and the JetBrains IDEs, with a command-line agent as well. It offers chat, inline edits, an agent mode that uses tools, and tab autocomplete. Continue does not bundle a model of its own; you decide which provider answers each kind of request.

That design suits teams that want to keep their editor and choose where models run. Because Continue supports any OpenAI-compatible API, inference.eu works with the built-in openai provider and a custom apiBase.

Models are listed in config.yaml, either locally in ~/.continue/config.yaml or through a shared configuration. Each entry names a provider, the model id, an apiBase and an apiKey. With provider: openai and apiBase set to inference.eu, Continue calls our Chat Completions endpoint.

The ${{ secrets.INFERENCE_EU_API_KEY }} syntax reads the key from ~/.continue/.env, keeping it out of the config file. Roles decide what each model does: chat for the chat panel and agent, edit for inline changes, apply for applying code blocks to files. You can give different roles to different models.

Give the chat, edit and apply roles to qwen3-coder-480b-a35b-instruct; it is trained for code and handles agent mode's tool calls well. Add glm-5.3 or kimi-k3 as further chat models for design questions, reviews and work with longer documents, and switch between them in the model selector.

gpt-oss-120b is a good choice when you want fast answers to short questions. Tab autocomplete is a different job: it needs a small, fast model trained for fill-in-the-middle completion. Leave the autocomplete role unassigned until you have chosen such a model, rather than giving it to a large chat model.

A 401 error usually means the secret was not resolved. Check that ~/.continue/.env contains INFERENCE_EU_API_KEY without quotes or spaces, then reload the window. "Model not found" means the model value differs from the inference.eu id; copy it exactly.

If Continue tries the Responses API and fails, add useResponsesApi: false to the model entry so it uses Chat Completions. If agent mode answers in prose instead of calling tools, choose a model with reliable tool calling, such as Qwen3 Coder or GLM-5.3. Make sure apiBase ends in /v1 without a trailing path.

With a local config, Continue sends model requests directly from your IDE to the apiBase you set. With inference.eu, the code, files and context you include are processed in EU data centres and never used for training. We sign a data processing agreement under Article 28 GDPR, share our subprocessor list and bill in euros.

Continue can collect anonymous usage telemetry; you can turn it off in the extension settings. If you use a configuration stored on Continue's hub, review how secrets are managed there. For the strictest setups, keep config.yaml and the .env file local and distribute them through your own tooling.