Zed Inline Assistant: provider not configured or model unsupported

If Zed shows LLM provider is not configured or does not support the configured model in Inline Assistant, first check whether Zed sent any HTTP request. When there is no request, the failure is local model selection; changing a key or base URL cannot fix a provider/model pair Zed did not resolve.

Use the configured provider ID

Zed documents custom OpenAI-compatible providers under language_models.openai_compatible. The object key you create there is the provider ID. In agent.inline_assistant_model.provider, use that key itself, without an openai_compatible: prefix.

Check the saved JSON shape too: agent.inline_assistant_model is an object with separate provider and model fields. In issue #63878, a community commenter says a slash-delimited string can fail to deserialize and leave the setting unset. That diagnosis is not maintainer-confirmed.

{
  "language_models": {
    "openai_compatible": {
      "my-gateway": {
        "api_url": "https://api.example.com/v1",
        "available_models": [
          {
            "name": "provider/model-id",
            "display_name": "My model",
            "max_tokens": 65536
          }
        ]
      }
    }
  },
  "agent": {
    "inline_assistant_model": {
      "provider": "my-gateway",
      "model": "provider/model-id"
    }
  }
}

Replace the example URL, provider ID and model with values from your setup. The model must exactly match one of that provider's available_models[].name values. Keep API keys in Zed's provider settings or the documented environment variable; do not put a key in settings.json.

Reselect and verify the Inline Assistant model

1. Run agent: open settings and open the LLM Providers page. Confirm your custom provider appears under the same ID used by the JSON key.
2. Open the Inline Assistant model selector and choose the model from that provider. If you edit settings.json manually, use the object shape shown above: separate provider and model fields.
3. Reopen the settings file and compare the saved provider ID and model name with the provider object and its available_models entry, including capitalization and punctuation.

Do not copy a provider selector string from an error message into the settings object. A provider namespace used elsewhere in the UI is not necessarily the ID expected by the settings field.

What the public report does and does not show

Zed issue #63878 reports that an OpenAI-compatible Groq provider worked in the Agent Panel but Inline Assistant showed this error before any HTTP request appeared in the log. Its posted settings used openai_compatible:Groq in the model selection even though the provider was registered under the key Groq. A later community comment identified that provider-ID mismatch as a likely local cause and showed the object form above. The report is still open; no maintainer-confirmed fix or resolved status is shown. Treat this as the first configuration check, not proof that every occurrence has the same cause.

Zed's current Inline Assistant docs say a provider already configured for the Agent Panel should also work for Inline Assistant. If the saved provider ID and model match exactly but Zed still rejects the selection before sending a request, record your Zed version and sanitized settings shape and compare with the open issue. If a request does reach the endpoint, switch to HTTP status, response body and API-surface diagnosis instead; capture the response before changing credentials.

Sources and scope

Zed: OpenAI-compatible endpoint settings and API keys
Zed: Agent and feature-specific model settings
Zed: Inline Assistant provider setup
Zed issue #63878 and its configuration discussion

Checked 2 November 2026 against the linked Zed documentation and public issue. The issue describes one setup on Zed 1.18.1 Linux; no Zed account, API key or model request was used.