"""Application configuration via environment variables."""

from pathlib import Path

from pydantic_settings import BaseSettings


class Settings(BaseSettings):
    # Default LLM: Groq-hosted model via its OpenAI-compatible API. Fast and
    # capable enough to plan messy multi-file reshapes. Set LLM_API_KEY in .env.
    # NOTE: raw cell grids are sent to Groq's API, so this is not on-prem private.
    # Override LLM_BASE_URL / LLM_MODEL in .env to point at any other endpoint.
    llm_model: str = "llama-3.3-70b-versatile"
    llm_api_key: str = ""
    llm_base_url: str = "https://api.groq.com/openai/v1"

    # Secure LLM — used when reference files are attached (may contain sensitive content)
    # Falls back to default LLM if not configured
    secure_llm_model: str = ""
    secure_llm_api_key: str | None = None
    secure_llm_base_url: str | None = None

    # Upload limits
    max_upload_size_mb: int = 200  # max file size per uploaded file (MB)

    # LLM timeout
    llm_timeout_seconds: int = 60  # max seconds to wait for LLM response

    # Session
    session_ttl_hours: int = 4  # reduced from 24 to save memory on HF Spaces

    # Directories
    upload_dir: Path = Path("./uploads")
    output_dir: Path = Path("./output")
    log_dir: Path = Path("./audit_logs")

    model_config = {
        "env_file": ".env",
        "env_file_encoding": "utf-8",
        "env_file_ignore_missing": True,
    }

    def ensure_dirs(self) -> None:
        for d in (self.upload_dir, self.output_dir, self.log_dir):
            d.mkdir(parents=True, exist_ok=True)

    @property
    def has_secure_llm(self) -> bool:
        """Check if a secure LLM is configured for sensitive content."""
        return bool(self.secure_llm_model and self.secure_llm_api_key)


settings = Settings()