configs/cli_agent_config.json

{
  "model_name": "anthropic/claude-opus-4-6",
  "save_sessions": true,
  "session_dataset_repo": "smolagents/ml-intern-sessions",
  "share_traces": true,
  "personal_trace_repo_template": "{hf_user}/ml-intern-sessions",
  "yolo_mode": false,
  "confirm_cpu_jobs": true,
  "auto_file_upload": true,
  "messaging": {
    "enabled": false,
    "auto_event_types": ["approval_required", "error", "turn_complete"],
    "destinations": {}
  },
  "mcpServers": {
    "hf-mcp-server": {
      "transport": "http",
      "url": "https://huggingface.co/mcp?login"
    }
  },
  "_local_provider_examples": {
    "_comment": "Set model_name to any of the prefixes below to use a local / self-hosted LLM. All local providers speak an OpenAI-compatible API. Reasoning effort is automatically disabled for local models.",
    "llamacpp/llama-3-8b": "llama.cpp server (default http://localhost:8080/v1). Override with LLAMACPP_API_BASE env var.",
    "lmstudio/llama-3-8b": "LM Studio (default http://localhost:1234/v1). Override with LMSTUDIO_API_BASE env var.",
    "mlx/llama-3-8b": "MLX OpenAI-compatible wrapper (default http://localhost:8000/v1). Override with MLX_API_BASE env var.",
    "nim/llama-3-8b": "NVIDIA NIM (default http://localhost:8000/v1). Override with NIM_API_BASE env var.",
    "ollama/llama3.1": "Ollama (default http://localhost:11434/v1). Override with OLLAMA_API_BASE env var.",
    "vllm/llama-3-8b": "vLLM (default http://localhost:8000/v1). Override with VLLM_API_BASE env var.",
    "tgi/llama-3-8b": "Text Generation Inference (default http://localhost:8080/v1). Override with TGI_API_BASE env var.",
    "local/llama-3-8b": "Custom PyTorch / any OpenAI-compatible server (default http://localhost:8000/v1). Override with LOCAL_API_BASE env var.",
    "_auth_note": "All local providers use api_key='no-key' by default. Set LOCAL_API_KEY env var if your server requires authentication."
  }
}