backend/app/config.py

from pydantic_settings import BaseSettings


class Settings(BaseSettings):
    groq_api_key: str = ""
    groq_model: str = "llama-3.1-8b-instant"
    groq_base_url: str = "https://api.groq.com/openai/v1"
    # Speech-to-text for the voice chat panel — same Groq account/credential,
    # different endpoint (audio/transcriptions instead of chat/completions).
    groq_whisper_model: str = "whisper-large-v3-turbo"
    # Comma-separated fallback models tried in order when primary hits rate
    # limit (decommissioned entries are skipped automatically at runtime)
    groq_fallback_models: str = (
        "llama-3.3-70b-versatile,"
        "openai/gpt-oss-20b,"
        "openai/gpt-oss-120b,"
        "qwen/qwen3-32b"
    )

    # ── Optional extra LLM providers (OpenAI-compatible) ─────────────────────
    # Tried in order after Groq when its models are rate-limited/unavailable.
    # A provider is skipped entirely when its API key is empty.
    cerebras_api_key: str = ""
    cerebras_base_url: str = "https://api.cerebras.ai/v1"
    cerebras_models: str = "llama-3.3-70b,llama3.1-8b"

    gemini_api_key: str = ""
    gemini_base_url: str = "https://generativelanguage.googleapis.com/v1beta/openai"
    gemini_models: str = "gemini-2.0-flash"

    openrouter_api_key: str = ""
    openrouter_base_url: str = "https://openrouter.ai/api/v1"
    # All current free models on OpenRouter (ordered best → smallest).
    # Audio/image/safety/vision-only and sub-2B models are excluded.
    # Decommissioned entries are skipped automatically at runtime.
    openrouter_models: str = (
        "openai/gpt-oss-120b:free,"
        "nvidia/nemotron-3-ultra-550b-a55b:free,"
        "nousresearch/hermes-3-llama-3.1-405b:free,"
        "nvidia/nemotron-3-super-120b-a12b:free,"
        "meta-llama/llama-3.3-70b-instruct:free,"
        "qwen/qwen3-next-80b-a3b-instruct:free,"
        "qwen/qwen3-coder:free,"
        "moonshotai/kimi-k2.6:free,"
        "google/gemma-4-31b-it:free,"
        "google/gemma-4-26b-a4b-it:free,"
        "nvidia/nemotron-3-nano-30b-a3b:free,"
        "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free,"
        "openai/gpt-oss-20b:free,"
        "cognitivecomputations/dolphin-mistral-24b-venice-edition:free,"
        "nvidia/nemotron-nano-9b-v2:free,"
        "meta-llama/llama-3.2-3b-instruct:free,"
        "poolside/laguna-m.1:free,"
        "poolside/laguna-xs.2:free,"
        "nex-agi/nex-n2-pro:free"
    )

    # Comma-separated list of allowed browser origins (frontend URLs).
    cors_origins: str = "http://localhost:3000,http://127.0.0.1:3000"

    # ── Abuse protection / cost control ──────────────────────────────────────
    # Required to call POST /ingest over HTTP; empty = endpoint disabled
    # (local ingestion via scripts/ingest.py is unaffected).
    admin_token: str = ""
    rate_limit_chat_per_minute: int = 20   # per client IP
    rate_limit_voice_per_minute: int = 15  # per client IP, voice transcription
    voice_max_upload_mb: int = 8           # ~60s of webm/opus, generous ceiling
    email_max_per_session: int = 5
    email_max_per_day: int = 50            # global, resets at UTC midnight
    response_cache_ttl_seconds: int = 3600
    response_cache_max_entries: int = 200
    session_idle_ttl_seconds: int = 6 * 3600
    max_sessions: int = 1000

    chroma_persist_directory: str = "./chroma_db"
    chroma_collection_name: str = "negoptim_knowledge"

    embedding_model: str = "intfloat/multilingual-e5-small"

    retrieval_top_k: int = 5
    chunk_size: int = 700
    chunk_overlap: int = 100

    max_history_turns: int = 10
    knowledge_base_path: str = "../knowledge"

    # ── Email ─────────────────────────────────────────────────────────────────
    # Priority: Resend (HTTP, works on any host) → SMTP (blocked on HF free tier)
    # Either absent → simulation mode (logs + reports success, no real delivery).
    resend_api_key: str = ""
    resend_from: str = "Negoptim AI <onboarding@resend.dev>"

    smtp_host: str = "smtp.gmail.com"
    smtp_port: int = 587
    smtp_user: str = ""
    smtp_password: str = ""
    smtp_from: str = ""            # defaults to smtp_user when empty
    smtp_from_name: str = "Negoptim AI"
    # All outgoing mail is force-routed here while testing (per spec).
    # Set to "" to deliver to the real recipient.
    email_override_to: str = "samirmajzoub9@gmail.com"

    @property
    def fallback_model_list(self) -> list[str]:
        return [m.strip() for m in self.groq_fallback_models.split(",") if m.strip()]

    @property
    def cerebras_model_list(self) -> list[str]:
        return [m.strip() for m in self.cerebras_models.split(",") if m.strip()]

    @property
    def gemini_model_list(self) -> list[str]:
        return [m.strip() for m in self.gemini_models.split(",") if m.strip()]

    @property
    def openrouter_model_list(self) -> list[str]:
        return [m.strip() for m in self.openrouter_models.split(",") if m.strip()]

    @property
    def cors_origin_list(self) -> list[str]:
        return [o.strip() for o in self.cors_origins.split(",") if o.strip()]

    @property
    def resend_configured(self) -> bool:
        return bool(self.resend_api_key)

    @property
    def smtp_configured(self) -> bool:
        return bool(self.smtp_host and self.smtp_user and self.smtp_password)

    @property
    def smtp_sender(self) -> str:
        return self.smtp_from or self.smtp_user

    class Config:
        env_file = ".env"


settings = Settings()