config.py

"""
VoiceVault — Centralized Configuration
======================================
Single source of truth for all project settings.
Loaded from environment variables / .env file.
Never import os.environ directly elsewhere — always use `cfg` from here.

Usage:
    from config import cfg
    print(cfg.groq_api_key)
    print(cfg.data_dir / "my_kb" / "chroma")
"""

from pathlib import Path
from pydantic import Field
from pydantic_settings import BaseSettings, SettingsConfigDict


class VoiceVaultConfig(BaseSettings):
    """
    Pydantic-settings config model.
    All fields are loaded from environment variables (case-insensitive).
    Defaults are safe, production-ready values.
    """

    model_config = SettingsConfigDict(
        env_file=".env",
        env_file_encoding="utf-8",
        case_sensitive=False,
        extra="ignore",
    )

    # ------------------------------------------------------------------ #
    # API Keys (required for LLM generation — optional for ingestion)     #
    # ------------------------------------------------------------------ #
    groq_api_key: str = Field(default="", alias="GROQ_API_KEY")
    gemini_api_key: str = Field(default="", alias="GEMINI_API_KEY")

    # ------------------------------------------------------------------ #
    # Model Identifiers                                                    #
    # ------------------------------------------------------------------ #
    whisper_model: str = Field(
        default="openai/whisper-large-v3",
        alias="WHISPER_MODEL",
    )
    distil_whisper_model: str = Field(
        default="distil-whisper/distil-large-v3",
        alias="DISTIL_WHISPER_MODEL",
    )
    embedding_model: str = Field(
        default="sentence-transformers/all-MiniLM-L6-v2",
        alias="EMBEDDING_MODEL",
    )
    cross_encoder_model: str = Field(
        default="cross-encoder/ms-marco-MiniLM-L12-v2",
        alias="CROSS_ENCODER_MODEL",
    )
    groq_llm_model: str = Field(
        default="llama-3.1-70b-versatile",
        alias="GROQ_LLM_MODEL",
    )
    gemini_llm_model: str = Field(
        default="gemini-1.5-flash",
        alias="GEMINI_LLM_MODEL",
    )

    # ------------------------------------------------------------------ #
    # File System Paths                                                    #
    # ------------------------------------------------------------------ #
    data_dir: Path = Field(default=Path("data"), alias="DATA_DIR")

    @property
    def uploads_dir(self) -> Path:
        """Sandboxed upload directory — all user files land here first."""
        return self.data_dir / "uploads"

    @property
    def models_cache_dir(self) -> Path:
        """Local model cache to avoid re-downloading on each restart."""
        return Path("models")

    def kb_dir(self, kb_name: str) -> Path:
        """Per-knowledge-base root directory."""
        return self.data_dir / kb_name

    def kb_chroma_dir(self, kb_name: str) -> Path:
        """ChromaDB persistence directory for a knowledge base."""
        return self.kb_dir(kb_name) / "chroma"

    def kb_bm25_path(self, kb_name: str) -> Path:
        """Serialized BM25 index path for a knowledge base."""
        return self.kb_dir(kb_name) / "bm25.pkl"

    def kb_db_path(self, kb_name: str) -> Path:
        """SQLite metadata database path for a knowledge base."""
        return self.kb_dir(kb_name) / "voicevault.db"

    # ------------------------------------------------------------------ #
    # Retrieval Parameters                                                 #
    # ------------------------------------------------------------------ #
    bm25_top_k: int = Field(default=20, alias="BM25_TOP_K")
    vector_top_k: int = Field(default=20, alias="VECTOR_TOP_K")
    rrf_k: int = Field(default=60, alias="RRF_K")
    rerank_top_k: int = Field(default=20, alias="RERANK_TOP_K")
    final_top_k: int = Field(default=5, alias="FINAL_TOP_K")
    max_chunks_per_page: int = Field(default=2, alias="MAX_CHUNKS_PER_PAGE")

    # ------------------------------------------------------------------ #
    # Chunking Parameters                                                  #
    # ------------------------------------------------------------------ #
    chunk_size_min: int = Field(default=100, alias="CHUNK_SIZE_MIN")
    chunk_size_max: int = Field(default=600, alias="CHUNK_SIZE_MAX")
    chunk_overlap: int = Field(default=50, alias="CHUNK_OVERLAP")
    semantic_similarity_threshold: float = Field(
        default=0.5, alias="SEMANTIC_SIMILARITY_THRESHOLD"
    )

    # ------------------------------------------------------------------ #
    # Generation Parameters                                                #
    # ------------------------------------------------------------------ #
    max_answer_tokens: int = Field(default=500, alias="MAX_ANSWER_TOKENS")
    llm_temperature: float = Field(default=0.1, alias="LLM_TEMPERATURE")
    conversation_window: int = Field(default=5, alias="CONVERSATION_WINDOW")

    # ------------------------------------------------------------------ #
    # Knowledge Base Limits                                                #
    # ------------------------------------------------------------------ #
    max_docs_per_kb: int = Field(default=500, alias="MAX_DOCS_PER_KB")
    max_chunks_per_kb: int = Field(default=100_000, alias="MAX_CHUNKS_PER_KB")
    kb_storage_warn_threshold: float = Field(
        default=0.80, alias="KB_STORAGE_WARN_THRESHOLD"
    )

    # ------------------------------------------------------------------ #
    # Security                                                             #
    # ------------------------------------------------------------------ #
    bcrypt_rounds: int = Field(default=12, alias="BCRYPT_ROUNDS")
    share_link_expiry_days: int = Field(default=7, alias="SHARE_LINK_EXPIRY_DAYS")

    # ------------------------------------------------------------------ #
    # Server                                                               #
    # ------------------------------------------------------------------ #
    host: str = Field(default="0.0.0.0", alias="HOST")
    port: int = Field(default=7860, alias="PORT")
    debug: bool = Field(default=False, alias="DEBUG")

    # ------------------------------------------------------------------ #
    # Supported Upload Extensions (security whitelist)                     #
    # ------------------------------------------------------------------ #
    allowed_extensions: frozenset[str] = frozenset(
        {".pdf", ".html", ".htm", ".docx", ".md", ".mdx", ".txt"}
    )
    max_upload_size_mb: int = Field(default=50, alias="MAX_UPLOAD_SIZE_MB")

    def ensure_directories(self) -> None:
        """Create all required runtime directories if they don't exist."""
        self.data_dir.mkdir(parents=True, exist_ok=True)
        self.uploads_dir.mkdir(parents=True, exist_ok=True)
        self.models_cache_dir.mkdir(parents=True, exist_ok=True)

    def has_groq_key(self) -> bool:
        """True if a Groq API key is configured."""
        return bool(self.groq_api_key)

    def has_gemini_key(self) -> bool:
        """True if a Gemini API key is configured."""
        return bool(self.gemini_api_key)

    def has_any_llm_key(self) -> bool:
        """True if at least one LLM key is available."""
        return self.has_groq_key() or self.has_gemini_key()


# ------------------------------------------------------------------ #
# Singleton — import this everywhere                                   #
# ------------------------------------------------------------------ #
cfg = VoiceVaultConfig()