Zed Editor Custom API Setup

Zed supports both its Agent Settings UI and settings.json for OpenAI-compatible endpoints. Let Zed store the TokenHub API key in the system keychain, or supply it through the environment variable derived from the provider name. Do not place the key in settings.json.

Add a provider in Agent Settings

1. Open the command palette and run agent: open settings.
2. Go to Settings → AI → LLM Providers.
3. Click Add Provider and choose an OpenAI-compatible provider.
4. Enter tokenhub as the provider name.
5. Enter https://us-api.tokenhub.com/v1 as the API URL.
6. Enter the exact model ID shown in the TokenHub model catalog.
7. Enter that model's real context window.
8. Enter your TokenHub API key in the provider settings so Zed stores it in the system keychain.

Edit settings.json when you need multiple models

Zed supports providers under language_models.openai_compatible. In this schema, max_tokens means the context window, not maximum output. Replace both limits with the values documented for your model:

{
  "language_models": {
    "openai_compatible": {
      "tokenhub": {
        "api_url": "__API_BASE_URL__/v1",
        "available_models": [
          {
            "name": "YOUR_TOKENHUB_MODEL_ID",
            "display_name": "TokenHub / YOUR_TOKENHUB_MODEL_ID",
            "max_tokens": 128000,
            "max_output_tokens": 8192,
            "capabilities": {
              "tools": true,
              "images": false,
              "parallel_tool_calls": false,
              "prompt_cache_key": false,
              "chat_completions": true,
              "interleaved_reasoning": false,
              "max_tokens_parameter": false
            }
          }
        ]
      }
    }
  }
}

If capabilities is omitted, Zed uses the OpenAI-compatible defaults shown above. Set images, parallel_tool_calls, or another capability to true only when both the model and the TokenHub endpoint actually support it.

Follow Zed's API-key lookup rule

Zed converts the provider ID to uppercase snake case and appends _API_KEY. The provider ID tokenhub therefore maps to:

export TOKENHUB_API_KEY="sk-..."

A non-empty environment variable takes precedence over the keychain. If you updated the key in the UI but Zed still sends an old credential, unset this variable and restart Zed.

Do not infer Reasoning or Responses support from a model name

• When a TokenHub model supports OpenAI-style reasoning_effort, set only a value the endpoint accepts. Zed exposes none, minimal, low, medium, high, xhigh, and max, but a third-party endpoint may support fewer values.
• TokenHub currently recommends Chat Completions for this connection, so keep capabilities.chat_completions: true.
• Change it to false only for a model that requires the Responses API. Do not switch APIs merely to enable reasoning.
• Set max_tokens_parameter: true only when the endpoint requires max_tokens instead of max_completion_tokens.

Verify inside Zed

Select TokenHub / YOUR_TOKENHUB_MODEL_ID in Agent or Inline Assistant and ask it to explain the current file. If the model is absent, check the JSON, provider ID, and available_models. For a 401, check the TOKENHUB_API_KEY visible to the Zed process. If tools or images fail, return to capabilities and make sure the declaration does not exceed the model's real support.

Official reference

• Zed: Use API Access