Skip to main content
ChatJS provides access to 120+ models from multiple providers through the Vercel AI Gateway.

​
Model Registry Architecture

​
Auto-Update (Runtime)

The app fetches available models from the AI Gateway API at runtime. Results are cached for 1 hour. 
// lib/ai/models.ts
export const fetchModels = unstable_cache(
  async (): Promise<ModelData[]> => {
    const models = await fetchModelsRaw();
    return models.map(toModelData);
  },
  ["ai-gateway-models"],
  { revalidate: 3600, tags: ["ai-gateway-models"] }
);
When no API key is available, the app falls back to a static snapshot in models.generated.ts.

​
Manual Update (Build-time)

To refresh the fallback snapshot, run: 
bun fetch:models
This fetches current models from the gateway and writes them to lib/ai/models.generated.ts. Run this periodically to keep the fallback fresh.

​
Model Configuration

All model settings are configured in chat.config.ts under the models key:
chat.config.ts
models: {
  providerOrder: ["openai", "google", "anthropic", "xai"],
  disabledModels: ["morph/morph-v3-large", "morph/morph-v3-fast"],
  curatedDefaults: [
    "openai/gpt-5-nano",
    "openai/gpt-5-mini",
    "google/gemini-2.5-flash-lite",
    "anthropic/claude-sonnet-4.5",
    // ...
  ],
  anonymousModels: [
    "google/gemini-2.5-flash-lite",
    "openai/gpt-5-mini",
    // ...
  ],
  defaults: {
    chat: "openai/gpt-5-nano",
    title: "google/gemini-2.5-flash-lite",
    pdf: "openai/gpt-5-mini",
    artifact: "openai/gpt-5-nano",
    // ...
  },
},
SettingDescription
providerOrderProvider sort order in model selector
disabledModelsModels hidden from all users
curatedDefaultsModels enabled by default for new users
anonymousModelsModels available to anonymous users
defaults.*Default model for each task type
Image generation uses a separate defaults.image setting. Language models with the image-generation tag can also generate images inline. See Image Generation for details.

​
Reasoning Variants

Models that support extended thinking are automatically split into two variants:
  • {model-id} (standard mode)
  • {model-id}-reasoning (extended thinking enabled)
This happens automatically in buildAppModels() for any model with reasoning: true.

​
Visibility Pipeline

Model visibility is determined through a pipeline:

​
Authenticated Users

  1. Remove disabled models (models.disabledModels)
  2. Apply defaults (models.curatedDefaults + any new API models not in models.generated.ts)
  3. Apply user overrides from saved preferences

​
Anonymous Users

  1. Remove disabled models
  2. Filter to models.anonymousModels only

​
User Settings

Users configure their model preferences at /settings/models. This page lets users toggle individual models on or off. A link to AI Registry provides detailed model information.

​
Environment Variables

The AI Gateway handles authentication with all providers. You do not need individual API keys for OpenAI, Anthropic, Google, etc. Instead, you authenticate with the gateway using one of these:
  • VERCEL_OIDC_TOKEN (automatically provided on Vercel deployments)
  • AI_GATEWAY_API_KEY (for non-Vercel deployments, obtain from Vercel AI Gateway)
Without either key, the app uses the fallback model list and cannot make API calls to providers.