Anthropic Claude

KhunQuant connects to Anthropic via the official anthropic-sdk-go. The anthropic/ provider implements the LLMProvider interface and additionally implements ThinkingCapable for extended thinking support.

Configuration

Add to ~/.khunquant/config.json:

{
  "model_list": [
    {
      "model_name": "claude",
      "model": "claude-sonnet-4-6",
      "api_key": "ANTHROPIC_API_KEY"
    }
  ]
}

Add the API key to ~/.khunquant/.security.yml:

model_list:
  - model_name: claude
    api_key: "sk-ant-..."

Extended thinking

Claude models support extended thinking, which lets the model reason step-by-step before producing a final response. Configure the thinking level per model:

thinking_level	Description
"none"	No extended thinking (default)
"low"	Brief thinking budget
"high"	Maximum thinking budget

{
  "model_list": [
    {
      "model_name": "claude-think",
      "model": "claude-opus-4-7",
      "api_key": "ANTHROPIC_API_KEY",
      "thinking_level": "high"
    }
  ]
}

Fallback chain example

Configure a primary model with a cheaper fallback:

{
  "agents": {
    "defaults": {
      "model_name": "claude",
      "model_fallbacks": ["claude-haiku"]
    }
  },
  "model_list": [
    { "model_name": "claude",       "model": "claude-sonnet-4-6", "api_key": "ANTHROPIC_API_KEY" },
    { "model_name": "claude-haiku", "model": "claude-haiku-4-5-20251001",  "api_key": "ANTHROPIC_API_KEY" }
  ]
}

The fallback chain automatically retries on rate limits, overload, or context overflow errors.