LLM.config.reference.json

{
  "_guide": {
    "purpose": "Reference-only multi-source LLM config. Copy the profiles you need into LLM.config.json and replace placeholder keys.",
    "default_profile_id": "Selects the active profile id at startup. It can be changed in WebUI later.",
    "response_stream": "Global default for visible assistant response streaming. Missing or false means disabled. Internal JSON/tool/classifier calls are not shown as live bubbles.",
    "thinking_stream": "Global default for live thinking stream when a provider/model supports it.",
    "request_timeout": "Per LLM request timeout in seconds. Increase for slow local models or long generation tasks.",
    "temperature": "Default sampling temperature. Use 0.1-0.3 for coding/reasoning, higher only for creative writing.",
    "effort": "Optional reasoning effort: off|low|medium|high|max. Leave empty to let the task level decide (L1-L2 low, L3-L5 max). Maps to Claude extended-thinking budget, OpenAI/DeepSeek reasoning_effort, and GLM thinking. Per-profile 'effort' overrides this global default.",
    "max_effort": "Optional ceiling for reasoning effort (off|low|medium|high|max). Caps the level chosen by task level or 'effort' so a low-tier profile never exceeds it. Per-profile 'max_effort' overrides this global default.",
    "capabilities": "Set per-profile multimodal support when auto detection is wrong.",
    "custom_http": "Only enable response_stream for custom_http when the endpoint returns OpenAI-compatible SSE or JSONL chunks."
  },
  "default_profile_id": "local-ollama",
  "thinking_stream": false,
  "response_stream": false,
  "temperature": 0.2,
  "request_timeout": 3600,
  "max_output_tokens": 8192,
  "context_token_limit": 200000,
  "auto_model_switch": false,
  "execution_mode": "sync",
  "profiles": [
    {
      "id": "local-ollama",
      "label": "Local Ollama",
      "provider": "ollama",
      "base_url": "http://127.0.0.1:11434",
      "model": "qwen3-coder:latest",
      "thinking_stream": true,
      "response_stream": false,
      "request_timeout": 3600,
      "temperature": 0.2,
      "capabilities": {
        "text": true,
        "image": false,
        "audio": false,
        "video": false
      },
      "_notes": "Best for local/offline work. Turn response_stream on only after the model endpoint is stable."
    },
    {
      "id": "local-vllm",
      "label": "Local vLLM",
      "provider": "vllm",
      "base_url": "http://127.0.0.1:8000/v1",
      "model": "auto",
      "api_key": "",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 3600,
      "temperature": 0.2,
      "_notes": "OpenAI-compatible local server. model='auto' is useful when the server exposes exactly one model."
    },
    {
      "id": "local-lmstudio",
      "label": "Local LM Studio",
      "provider": "lmstudio",
      "base_url": "http://127.0.0.1:1234/v1",
      "model": "auto",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 3600,
      "_notes": "Use after starting LM Studio's OpenAI-compatible server."
    },
    {
      "id": "openai-main",
      "label": "OpenAI Compatible",
      "provider": "openai_compat",
      "base_url": "https://api.openai.com/v1",
      "model": "gpt-4o-mini",
      "api_key": "REPLACE_WITH_OPENAI_API_KEY",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 600,
      "temperature": 0.2,
      "_notes": "Use any OpenAI-compatible cloud by changing base_url/model/api_key."
    },
    {
      "id": "siliconflow-main",
      "label": "SiliconFlow",
      "provider": "siliconflow",
      "base_url": "https://api.siliconflow.cn/v1",
      "model": "Qwen/Qwen3-Next-80B-A3B-Instruct",
      "api_key": "REPLACE_WITH_SILICONFLOW_API_KEY",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 900,
      "_notes": "OpenAI-compatible provider. Keep request_timeout higher for large coding tasks."
    },
    {
      "id": "anthropic-main",
      "label": "Anthropic",
      "provider": "anthropic",
      "base_url": "https://api.anthropic.com",
      "model": "claude-sonnet-4-20250514",
      "api_key": "REPLACE_WITH_ANTHROPIC_API_KEY",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 900,
      "effort": "",
      "max_effort": "max",
      "_notes": "Uses Anthropic Messages API streaming, not OpenAI SSE parsing. effort empty = task level decides (L3-L5 run extended thinking at max budget)."
    },
    {
      "id": "glm-main",
      "label": "GLM",
      "provider": "glm",
      "base_url": "https://open.bigmodel.cn/api/paas/v4",
      "model": "glm-4-flash",
      "api_key": "REPLACE_WITH_GLM_API_KEY",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 900,
      "effort": "",
      "max_effort": "max",
      "_notes": "OpenAI-compatible GLM route. If using a proxy endpoint, replace base_url. effort maps to GLM thinking={type:enabled}; empty = task level decides."
    },
    {
      "id": "kimi-main",
      "label": "KIMI Moonshot",
      "provider": "kimi",
      "base_url": "https://api.moonshot.cn/v1",
      "model": "moonshot-v1-32k",
      "api_key": "REPLACE_WITH_MOONSHOT_API_KEY",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 900,
      "_notes": "Use moonshot-v1-128k when long context is more important than latency."
    },
    {
      "id": "openrouter-main",
      "label": "OpenRouter",
      "provider": "openrouter",
      "base_url": "https://openrouter.ai/api/v1",
      "model": "anthropic/claude-sonnet-4",
      "api_key": "REPLACE_WITH_OPENROUTER_API_KEY",
      "headers": {
        "HTTP-Referer": "http://127.0.0.1:8080",
        "X-Title": "Clouds Coder"
      },
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 900,
      "_notes": "Model ids must match OpenRouter's model slug format."
    },
    {
      "id": "custom-openai-sse",
      "label": "Custom OpenAI SSE",
      "provider": "custom_http",
      "endpoint": "https://your-api.example.com/v1/chat/completions",
      "model": "your-model-name",
      "api_key": "REPLACE_WITH_CUSTOM_API_KEY",
      "headers": {
        "Authorization": "Bearer ${api_key}"
      },
      "payload_template": "",
      "thinking_stream": false,
      "response_stream": true,
      "request_timeout": 900,
      "_notes": "response_stream=true requires the endpoint to accept stream=true and return OpenAI-compatible data chunks."
    },
    {
      "id": "custom-nonstream",
      "label": "Custom Non-stream JSON",
      "provider": "custom_http",
      "endpoint": "https://your-api.example.com/chat",
      "model": "your-model-name",
      "api_key": "REPLACE_WITH_CUSTOM_API_KEY",
      "headers": {
        "Authorization": "Bearer ${api_key}",
        "Content-Type": "application/json"
      },
      "payload_template": "{\"model\":\"${model}\",\"messages\":${messages_json},\"temperature\":${temperature},\"max_tokens\":${max_tokens},\"stream\":false}",
      "thinking_stream": false,
      "response_stream": false,
      "request_timeout": 900,
      "_notes": "Use this shape when the endpoint returns one final JSON response instead of streaming chunks."
    }
  ],
  "model_capabilities": {
    "gpt-4o-mini": {
      "text": true,
      "image": true,
      "audio": false,
      "video": false
    },
    "claude-sonnet-4-20250514": {
      "text": true,
      "image": true,
      "audio": false,
      "video": false
    }
  },
  "media_endpoints": {
    "openai": {
      "image": "https://api.openai.com/v1/images/generations",
      "audio": "https://api.openai.com/v1/audio/speech"
    }
  }
}