API Key Model Visibility

This opt-in feature makes the desktop model picker trust the non-hidden model catalog returned by Codex CLI for API-key authenticated hosts.

The upstream desktop UI can apply a separate Statsig available_models allowlist after model/list succeeds. That allowlist is useful for ChatGPT rollouts, but it can hide models exposed by an OpenAI-compatible provider even when the provider marks them as visible. This feature bypasses that extra allowlist only when the active host reports authMethod = "apikey".

Enable

Add the feature id to linux-features/features.json:

{
  "enabled": [
    "api-key-model-visibility"
  ]
}

Then rebuild the app:

./install.sh

Behavior

• API-key hosts show every model/list entry whose hidden field is false.
• ChatGPT, Copilot, and Amazon Bedrock hosts keep the upstream filtering rules.
• Models that the CLI marks as hidden remain hidden.
• The feature does not grant model access. The configured provider must accept the selected model id.
• Reasoning choices such as Max or Ultra appear only when the selected model advertises those efforts and the desktop UI enables them.

Risks

The CLI can list a model that the configured provider or current API key cannot actually use. In that case the model will appear in the picker but the provider may reject requests at runtime.

Test

node --test linux-features/api-key-model-visibility/test.js