Use Zed with open-source models hosted in the EU

Zed is a code editor written in Rust and built by Zed Industries, focused on speed and real-time collaboration. It includes an Agent Panel that can read your project, edit files and run commands, an inline assistant for targeted edits, and edit predictions while you type. It also hosts external agents through the Agent Client Protocol.

For its own AI features, Zed lets you bring your own provider. Besides first-class providers it supports any OpenAI-compatible endpoint, which is how you connect inference.eu without a plugin.

OpenAI-compatible providers live under language_models.openai_compatible in settings.json. Each provider has a name, an api_url and a list of available_models. For each model, set name to the exact inference.eu id and max_tokens to its context window, and declare capabilities such as tools.

Zed does not store keys in settings.json. Enter the key in the provider settings, where it goes to the system keychain, or set an environment variable. Zed builds the variable name from the provider id in upper snake case plus _API_KEY, so the provider inference-eu reads INFERENCE_EU_API_KEY. A non-empty environment variable takes precedence over the keychain.

The Agent Panel makes repeated tool calls to read files, edit code and run commands, so pick models with dependable tool calling. qwen3-coder-480b-a35b-instruct is a strong default for agent work. glm-5.3 and kimi-k3 are good for planning larger changes and explaining unfamiliar code; add them as further entries with their context sizes.

For the inline assistant and quick questions, gpt-oss-120b is fast. If a model supports reasoning effort, you can set reasoning_effort on its entry. Edit predictions are a separate feature with their own provider setting and are not served by these chat models.

If the provider shows as not authenticated, Zed found no key. Enter it in the provider settings, or start Zed from a shell where INFERENCE_EU_API_KEY is set, keeping in mind that the name follows the provider id. A 401 means the key is wrong. If you set the variable and later want to use the keychain instead, unset it and restart Zed.

"Model not found" means name differs from the inference.eu id. If the agent replies without calling tools, check that capabilities.tools is true and use a model trained for tool use. If the endpoint rejects the output limit, set capabilities.max_tokens_parameter to true.

Zed initialises language model providers in the local app and sends requests from your machine to the api_url you set, also for SSH and remote projects. With inference.eu, the code and context in your requests 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.

Other Zed features have their own settings. Edit predictions can use Zed's hosted service, and Zed collects telemetry that you can turn off. Review both if your policy requires all code to stay with inference.eu.