Model Configuration

Configure AI providers and model selection for NullClaw

NullClaw supports 50+ AI providers through a unified configuration interface. Configure provider credentials, default models, and temperature settings.

Basic Model Configuration

Provider Setup

Configure providers in the models.providers section:

{
  "models": {
    "providers": {
      "openrouter": {
        "api_key": "YOUR_OPENROUTER_API_KEY"
      },
      "openai": {
        "api_key": "YOUR_OPENAI_API_KEY"
      }
    }
  }
}

models.providers · object

Map of provider name to provider configuration. Each key is a provider identifier (e.g., openrouter, openai, anthropic).

models.providers.<name>.api_key · string · required

API key for the provider. Can also be set via environment variables like OPENROUTER_API_KEY.

models.providers.<name>.base_url · string

Optional custom base URL for API requests. Useful for proxies or compatible providers.

models.providers.<name>.native_tools · boolean · default: "true"

Whether the provider supports native OpenAI-style tool calls. Set to false to use XML tool format via system prompt.

Default Model Selection

Configure the default model used by agents:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "openrouter/anthropic/claude-sonnet-4"
      }
    }
  },
  "default_provider": "openrouter"
}

agents.defaults.model.primary · string · default: "openrouter/anthropic/claude-sonnet-4"

Primary model identifier. Format: provider/model-name or just model-name if using the default provider.

default_provider · string · default: "openrouter"

Default provider when model identifier doesn’t specify one.

default_model · string

Legacy fallback model name (prefer agents.defaults.model.primary).

Temperature Configuration

Control model creativity and randomness:

{
  "default_temperature": 0.7
}

default_temperature · number · default: "0.7"

Sampling temperature (0.0 to 1.0). Lower values are more deterministic, higher values more creative.

reasoning_effort · string

Optional reasoning effort level for compatible models (e.g., low, medium, high).

Multiple Provider Configuration

You can configure multiple providers for fallback and redundancy:

{
  "models": {
    "providers": {
      "openrouter": {
        "api_key": "sk-or-v1-..."
      },
      "openai": {
        "api_key": "sk-..."
      },
      "anthropic": {
        "api_key": "sk-ant-..."
      }
    }
  },
  "reliability": {
    "fallback_providers": ["openai", "anthropic"],
    "provider_retries": 2,
    "provider_backoff_ms": 500
  }
}

reliability.fallback_providers · array

Ordered list of fallback provider names to try if the primary fails.

reliability.provider_retries · number · default: "2"

Number of retry attempts per provider before moving to fallback.

reliability.provider_backoff_ms · number · default: "500"

Milliseconds to wait between retry attempts (exponential backoff).

Supported Providers

NullClaw supports these providers out of the box:

• openrouter — Multi-provider router (recommended)
• openai — GPT-4, GPT-3.5, and embedding models
• anthropic — Claude 3/4 family
• google — Gemini models
• mistral — Mistral AI models
• groq — Fast inference for open models
• ollama — Local model hosting
• cohere — Cohere models
• together — Together AI models
• And 40+ more compatible providers

Environment Variable Overrides

You can set API keys via environment variables instead of config file:

export OPENROUTER_API_KEY="sk-or-v1-..."
export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."

Environment variables take precedence over config file values.

Example: Local Model with Ollama

Run models locally using Ollama:

{
  "models": {
    "providers": {
      "ollama": {
        "base_url": "http://localhost:11434"
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "ollama/llama3"
      }
    }
  },
  "default_provider": "ollama"
}

No API key is required for Ollama. Just ensure the Ollama server is running locally.