Models

About Models in Nuwa

TODO: update this doc

Choosing a Model

The Nuwa Client resolves and runs your Cap’s model via the Nuwa LLM Gateway by default. You select a provider and a model ID in the Cap:

core: {
  model: {
    providerId: 'openrouter' | 'anthropic' | 'google' | 'groq' | 'togetherai' | 'azure' | 'deepseek' | 'mistral' | 'openai_chat_completion' | 'openai_responses',
    modelId: 'openai/gpt-4o-mini',
    parameters: { temperature: 0.3 },
    supportedInputs: ['text', 'image'], // must include 'text'
    contextLength: 128000,
    // Optional: override default LLM Gateway
    // customGatewayUrl: 'https://your-gateway.example.com/api/v1'
  }
}

Providers

• OpenRouter compatible (default via Nuwa LLM Gateway)
• Anthropic, Google, Azure, Together, Groq, Mistral, DeepSeek
• OpenAI (Chat Completions, Responses)

The Nuwa Client attaches DID authentication and payment metadata to every request; your gateway can enforce both without extra code in the Cap.

Streaming & Usage

The LLM Gateway supports both non-streaming and streaming (SSE/NDJSON) responses. When streaming, payment frames are injected in-band and filtered out for your application automatically.

Custom Gateway

Set customGatewayUrl if your Cap must talk to a specialized backend. The client will use that base URL with the same auth and payment behavior.

If your gateway is not OpenRouter-compatible, implement the usage fields your Cap expects and verify your content type and payment headers match the Nuwa Payment Kit.