mcp.json

{
  "version": "1.0.0",
  "engine": {
    "name": "Mohawk-Inference-Engine",
    "max_concurrent_requests": 10,
    "default_backend": "llama-cpp"
  },
  "models": [
    {
      "id": "llama-3.2-3b-instruct",
      "name": "Llama 3.2 3B Instruct",
      "path": "/models/llama-3.2-3b-instruct.Q4_K_M.gguf",
      "context_length": 8192,
      "embedding": false,
      "system_prompt": "You are a helpful, harmless, and honest AI assistant developed by Mohawk.",
      "hardware_config": {
        "n_gpu_layers": 30,
        "main_gpu": 0,
        "tensor_split": [0.8, 0.2],
        "offload_kqv": true,
        "flash_attn": true
      }
    },
    {
      "id": "mistral-7b-v0.3",
      "name": "Mistral 7B v0.3",
      "path": "/models/mistral-7b-v0.3.Q5_K_M.gguf",
      "context_length": 32768,
      "embedding": true,
      "system_prompt": "You are Mistral, an advanced AI capable of complex reasoning.",
      "hardware_config": {
        "n_gpu_layers": 24,
        "main_gpu": 0,
        "tensor_split": [1.0],
        "offload_kqv": true,
        "flash_attn": false
      }
    }
  ],
  "system_prompts": {
    "default": "You are a helpful AI assistant.",
    "coder": "You are an expert software engineer. Provide clean, efficient, and well-commented code.",
    "creative": "You are a creative writer. Focus on storytelling, imagery, and emotional depth.",
    "analyst": "You are a data analyst. Focus on facts, logic, and structured outputs."
  },
  "metrics": {
    "enabled": true,
    "refresh_rate_ms": 500,
    "expose_hardware": true,
    "log_level": "info"
  },
  "security": {
    "api_keys": ["mk_live_1234567890"],
    "cors_origins": ["http://localhost:3000", "http://localhost:5173"]
  }
}