ERP/bot/app/config.py

"""Bot configuration.

All settings can be overridden with environment variables prefixed with
``PROCUREMENT_BOT_`` (e.g. ``PROCUREMENT_BOT_BOT_TOKEN=...``) or a local
``.env`` file. Docker/compose passes them in directly (same convention as
the api service's ``PROCUREMENT_*`` prefix).
"""
from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        env_prefix="PROCUREMENT_BOT_",
        env_file=".env",
        env_file_encoding="utf-8",
        extra="ignore",
    )

    # Bale Bot API — Telegram-compatible clone hosted at tapi.bale.ai.
    BALE_API_BASE: str = "https://tapi.bale.ai"
    BOT_TOKEN: str = ""  # from Bale's BotFather
    POLL_TIMEOUT_SECONDS: int = 25  # getUpdates long-poll window
    POLL_ERROR_SLEEP_SECONDS: float = 3.0

    # ERP API (in-network service URL inside Docker).
    API_BASE_URL: str = "http://localhost:8000/api/v1"
    API_TIMEOUT_SECONDS: float = 30.0
    # Optional bearer for the day procurement endpoints start enforcing auth.
    SERVICE_TOKEN: str = ""

    # Frontend URL (for opening Wizard from Bale bot)
    FRONTEND_URL: str = "http://localhost:8081"

    # Security token for standalone wizard review URL
    DRAFT_TOKEN_SECRET: str = "procurement-draft-secret-change-in-prod-v7"
    DRAFT_TOKEN_EXPIRE_SECONDS: int = 7200  # 2 hours

    # Persistent storage directory for draft attachments (proformas)
    ATTACHMENTS_DIR: str = "data/attachments/proformas"

    # OpenAI-compatible AI provider (LLM), e.g. OpenAI / GLM / DeepSeek / OpenRouter.
    AI_BASE_URL: str = ""  # e.g. https://api.openai.com/v1
    AI_API_KEY: str = ""
    LLM_MODEL: str = "gpt-4o-mini"
    LLM_TIMEOUT_SECONDS: float = 60.0

    # STT providers — support multiple models/APIs in parallel.
    # 1. JSON list: [{"name": "m1", "base_url": "...", "api_key": "...", "model": "..."}, ...]
    STT_PROVIDERS_JSON: str = ""

    # 2. Individual provider variables (up to 3 distinct STT models)
    STT_PROVIDER_1_NAME: str = "model-1"
    STT_PROVIDER_1_BASE_URL: str = ""
    STT_PROVIDER_1_API_KEY: str = ""
    STT_PROVIDER_1_MODEL: str = ""
    STT_PROVIDER_1_LANGUAGE: str = "fa"

    STT_PROVIDER_2_NAME: str = "model-2"
    STT_PROVIDER_2_BASE_URL: str = ""
    STT_PROVIDER_2_API_KEY: str = ""
    STT_PROVIDER_2_MODEL: str = ""
    STT_PROVIDER_2_LANGUAGE: str = "fa"

    STT_PROVIDER_3_NAME: str = "model-3"
    STT_PROVIDER_3_BASE_URL: str = ""
    STT_PROVIDER_3_API_KEY: str = ""
    STT_PROVIDER_3_MODEL: str = ""
    STT_PROVIDER_3_LANGUAGE: str = "fa"

    # Legacy single STT provider fallback:
    STT_BASE_URL: str = ""
    STT_API_KEY: str = ""
    STT_MODEL: str = "whisper-1"
    STT_LANGUAGE: str = "fa"

    # Audio Enhancement / Preprocessing (denoising & speech quality enhancement)
    AUDIO_ENHANCE_ENABLED: bool = False
    AUDIO_ENHANCE_BASE_URL: str = ""
    AUDIO_ENHANCE_API_KEY: str = ""
    AUDIO_ENHANCE_MODEL: str = ""
    AUDIO_ENHANCE_TIMEOUT_SECONDS: float = 60.0
    AUDIO_ENHANCE_TARGET_FORMAT: str = ""  # empty = pass raw bytes directly; e.g. "wav" / "mp3"
    AUDIO_ENHANCE_METHOD: str = "auto"     # "auto" | "raw_stream" | "multipart"

    @property
    def stt_base_url(self) -> str:
        return self.STT_BASE_URL or self.AI_BASE_URL

    @property
    def stt_api_key(self) -> str:
        return self.STT_API_KEY or self.AI_API_KEY

    @property
    def stt_providers(self) -> list[dict]:
        import json
        if self.STT_PROVIDERS_JSON:
            try:
                data = json.loads(self.STT_PROVIDERS_JSON)
                if isinstance(data, list) and len(data) > 0:
                    return data
            except Exception:
                pass

        # Check explicit STT_PROVIDER_1 / 2 / 3
        custom_list: list[dict] = []
        for idx in (1, 2, 3):
            b_url = getattr(self, f"STT_PROVIDER_{idx}_BASE_URL")
            key = getattr(self, f"STT_PROVIDER_{idx}_API_KEY")
            model = getattr(self, f"STT_PROVIDER_{idx}_MODEL")
            name = getattr(self, f"STT_PROVIDER_{idx}_NAME")
            lang = getattr(self, f"STT_PROVIDER_{idx}_LANGUAGE") or "fa"
            if b_url and key:
                custom_list.append({
                    "name": name or f"model-{idx}",
                    "base_url": b_url,
                    "api_key": key,
                    "model": model or self.STT_MODEL,
                    "language": lang,
                    "timeout": 30.0,
                })
        if custom_list:
            return custom_list

        # Fallback to single legacy STT provider if set
        if self.stt_base_url and self.stt_api_key:
            return [{
                "name": "default",
                "base_url": self.stt_base_url,
                "api_key": self.stt_api_key,
                "model": self.STT_MODEL,
                "language": self.STT_LANGUAGE,
                "timeout": 30.0,
            }]
        return []

    # PostgreSQL — same instance/database as the api service.
    DATABASE_URL: str = "postgresql+asyncpg://postgres:postgres@localhost:5432/procurement"

    # Chats that may use the bot without Casdoor linking (comma separated).
    ADMIN_CHAT_IDS: str = ""

    # Drafts older than this are marked expired by the sweeper loop.
    DRAFT_TTL_HOURS: int = 24

    # Observability (phase 1) — silent pipeline recorder (DB only; no external
    # tools, no extra LLM tokens). Data is NEVER deleted.
    OBS_ENABLED: bool = True
    OBS_VOICE_DIR: str = "data/voices"   # persistent volume (./bot/data) — keep forever
    OBS_PROMPT_VERSION: str = ""         # manual label for the NLU prompt in use
    # Read-only DB role for the Grafana datasource (dashboard phase 1); empty =
    # role provisioning is skipped (local dev can point the datasource at the
    # postgres user instead). Rotating = change env + restart the bot.
    OBS_READER_PASSWORD: str = ""

    # Jev (TypeSafe) entity matching — phases 2-5. Model version is PINNED
    # (phase 0 resolved jev-1.13.0) so upstream updates cannot silently change
    # decisions; every response's actual model is recorded for drift detection.
    TYPESAFE_API_KEY: str = ""
    JEV_MODEL: str = "jev-1.13.0"
    JEV_ENABLED: bool = True
    JEV_TIMEOUT: float = 8.0
    JEV_CONF_EXACT: float = 0.85
    JEV_CONF_FLOOR: float = 0.65   # phase-6 calibration (2026-10-12): 0.60 let one
                                   # wrong candidate (جعبه→بسته @0.61) through; the
                                   # correct picks it drops are covered by difflib.
    JEV_MAX_QUESTIONS_PER_FIELD: int = 15

    LOG_LEVEL: str = "INFO"

    @property
    def admin_chat_ids(self) -> set[int]:
        out: set[int] = set()
        for part in self.ADMIN_CHAT_IDS.replace(";", ",").split(","):
            part = part.strip()
            if part.isdigit():
                out.add(int(part))
        return out


settings = Settings()