Skip to main content
Zed is a high-performance code editor with a built-in AI Agent Panel and Inline Assist. Both features use the language_models.openai_compatible provider mechanism, which routes requests through the Orq.ai AI Gateway and makes the full model catalog available, including Anthropic Claude models.

Prerequisites

Setup

Two paths produce equivalent results. The GUI writes to the same settings.json as the manual approach.
Open Zed settingsPress , on macOS or Ctrl , on Linux / Windows to open settings.json.Add the provider blockMerge the following into the top-level object:
"language_models": {
  "openai_compatible": {
    "Orq": {
      "api_url": "https://api.orq.ai/v3/router",
      "available_models": [
        {
          "name": "anthropic/claude-sonnet-4-6",
          "display_name": "Claude Sonnet 4.6 (Orq)",
          "max_tokens": 200000,
          "capabilities": {
            "tools": true,
            "images": true,
            "parallel_tool_calls": true,
            "prompt_cache_key": false
          }
        },
        {
          "name": "openai/gpt-4o",
          "display_name": "GPT-4o (Orq)",
          "max_tokens": 128000,
          "capabilities": {
            "tools": true,
            "images": true,
            "parallel_tool_calls": true,
            "prompt_cache_key": false
          }
        }
      ]
    }
  }
}
Add as many models as needed. Use the provider/model_id format for name (e.g. anthropic/claude-opus-4-7, openai/gpt-4o-mini).
The capabilities object requires all four fields (tools, images, parallel_tool_calls, prompt_cache_key). Zed’s settings schema rejects model entries that omit any of them.
Add the API keyOpen the Agent Panel → Configure Providers → click Orq → paste the <ORQ_API_KEY>.Alternatively, export it as an environment variable before launching Zed. Zed derives the variable name from the provider name, so a provider named Orq reads ORQ_API_KEY automatically:
export ORQ_API_KEY="<ORQ_API_KEY>"
Start a sessionOpen the Agent Panel, pick a model from the dropdown, and send a message.

Caveats

Zed generates thread titles via a separate LLM call that may route to any model in available_models. Traces will show token usage against that model regardless of which model was active in the Agent Panel.

Troubleshooting

Verify that settings.json is valid JSON and that the capabilities object for each model includes all four fields: tools, images, parallel_tool_calls, and prompt_cache_key. Zed silently drops malformed model entries.
Confirm that ORQ_API_KEY is set in the environment Zed was launched from, or that the key was pasted directly in the Configure Providers panel. The environment variable name is derived from the provider name: a provider named Orq reads ORQ_API_KEY.
Verify the api_url in settings.json (or the GUI field) is set to https://api.orq.ai/v3/router. Requests sent to the provider’s own endpoint bypass the AI Gateway and produce no Traces.
Edit Predictions is not currently supported through the AI Gateway.