Zed + Luma Cloud

Before you start

• Create a Luma Cloud API key in Dashboard → API and copy a model ID from the authenticated model catalog.
• Use a Zed version with custom OpenAI-compatible providers.

Get an API key and a model ID →

Set up Zed

1. Add the provider

   Run agent: open settings. Under LLM Providers, choose Add Provider and OpenAI-compatible. Name it Luma Cloud.

2. Enter model details

   Set the API URL and model ID below. Enter the model's supported context window. Use the provider settings key field; Zed stores the key in the system keychain.

3. Select capabilities

   Select the new model. Match tool and image settings to its actual capabilities. For a Responses-only model, set capabilities.chat_completions to false; otherwise use Chat Completions.

4. Start with the Zed Agent

   Save the provider and select its model in a new Zed Agent conversation. Do not select an external ACP agent for this check. If you changed an environment-provided key, restart Zed so the local app receives the updated environment.

API URL

https://api.lumaos.cloud/v1

Model ID

MODEL_ID_FROM_CATALOG

Key storage

Provider settings → system keychain

{
  "language_models": {
    "openai_compatible": {
      "luma-cloud": {
        "api_url": "https://api.lumaos.cloud/v1",
        "available_models": [
          {
            "name": "MODEL_ID_FROM_CATALOG",
            "display_name": "Luma Cloud",
            "max_tokens": "REPLACE_WITH_CONTEXT_WINDOW_INTEGER",
            "capabilities": {
              "tools": false,
              "images": false,
              "chat_completions": true
            }
          }
        ]
      }
    }
  }
}

Reply with exactly: Luma Cloud connection works.

Read README.md in this test folder and summarize its first paragraph. Do not edit files, run shell commands, or access the network.

Check the connection

1. Send a short text request, then check Usage → API Wallet for Wallet usage, or subscription Usage for Builder keys. A populated model list alone does not confirm a successful response.
2. Before using an agent on your project, try one read-only file task in a disposable folder. Confirm the tool result and final answer, keeping action approval enabled.

For errors or a request that stops, see troubleshooting. A visible model list alone does not confirm that a chat or editing task can complete.

What to expect

• Zed's edit prediction has its own configuration. This connection covers model-backed chat, Inline Assistant, and similar Zed features.
• External agents configure their own API connection; they do not inherit this provider automatically.
• Do not place API keys in settings.json. Existing provider environment variables can override saved keys.

Troubleshooting Zed