Use Cline with open-source models hosted in the EU

Cline is an open-source autonomous coding agent. It runs as an extension in VS Code and JetBrains IDEs and as a command-line tool. It reads files, writes and edits code, runs terminal commands and can use a browser, asking for approval at each step unless you allow actions automatically. Plan and Act modes separate thinking about a task from carrying it out.

Cline works with many model providers. Its OpenAI Compatible option accepts any endpoint that implements the OpenAI Chat Completions API, which is how you run it on open-weight models at inference.eu.

In the extension, choose OpenAI Compatible as the API provider and fill in three fields: Base URL, API Key and Model ID. The base URL is the inference.eu URL ending in /v1; Cline adds /chat/completions itself. The key is stored in your editor's secret storage.

The Model Configuration section lets you tell Cline the context window, maximum output tokens and whether the model accepts images. Cline uses these numbers to decide when to condense the conversation, so set them to the values listed for the model. In the CLI, cline auth with the -p openai, -k, -b and -m flags writes the same settings without the interactive wizard.

Cline sends long system prompts and relies on tool calls for every action, so model choice matters. qwen3-coder-480b-a35b-instruct is a strong default for Act mode: it is trained for agentic coding and follows tool formats closely. glm-5.3 and kimi-k3 are good Plan mode models when a task needs careful reasoning across many files.

Cline lets you set separate models for Plan and Act. A common pattern is GLM-5.3 or Kimi K3 to plan and Qwen3 Coder to act. gpt-oss-120b is fast and works well for small, well-defined tasks.

"Invalid API Key" or a 401 means the key is wrong or was pasted with spaces. "Model Not Found" means the Model ID differs from the inference.eu id; copy it exactly. Connection errors usually come from a base URL with a typo or an extra path such as /chat/completions.

If Cline reports that the model did not use a tool, or loops on the same step, the model is struggling with tool calls; switch to Qwen3 Coder, GLM-5.3 or Kimi K3. If long tasks stop abruptly, check the context window you entered: a value larger than the model supports leads to rejected requests.

Cline runs on your machine and sends model requests directly to the provider you configure. With inference.eu, your code, terminal output and task history 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.

Cline can collect anonymous telemetry, which you can turn off in its settings. Do not sign in to the Cline provider if you want all model traffic to go to inference.eu; the OpenAI Compatible provider uses only the base URL and key you enter.