import os
from dotenv import load_dotenv

load_dotenv()

class Config:
    # DeepSeek API - PRIMARY (OpenAI-compatible)
    DEEPSEEK_API_KEY = os.getenv('DEEPSEEK_API_KEY', '')
    DEEPSEEK_BASE_URL = 'https://api.deepseek.com/v1'
    DEEPSEEK_MODEL = 'deepseek-chat'  # DeepSeek V3 model
    
    # OpenRouter API - For OCR and fallback LLM
    OPENROUTER_API_KEY = os.getenv('OPENROUTER_API_KEY', '')
    OPENROUTER_BASE_URL = 'https://openrouter.ai/api/v1'
    
    # AI Models (OpenRouter fallback)
    MODEL_MAP = {
        'gemma': 'google/gemma-3-4b-it:free',
        'mistral': 'mistralai/mistral-small-3.1-24b-instruct:free',
    }
    
    # Use DeepSeek first, then OpenRouter models
    USE_DEEPSEEK = True  # Set to False to use OpenRouter instead
    FALLBACK_ORDER = ['gemma', 'mistral']
    
    # ChromaDB Cloud Configuration
    CHROMA_API_KEY = os.getenv('CHROMA_API_KEY', '')
    CHROMA_TENANT = os.getenv('CHROMA_TENANT', 'default_tenant')
    CHROMA_DATABASE = os.getenv('CHROMA_DATABASE', 'default_database')
    CHROMA_HOST = 'api.trychroma.com'  # ChromaDB Cloud endpoint
    
    # JWT Configuration
    JWT_SECRET = os.getenv('JWT_SECRET', 'your-secret-key-change-in-production')
    JWT_EXPIRY_HOURS = 24
    
    # Upload Configuration
    UPLOAD_FOLDER = os.path.join(os.path.dirname(__file__), 'uploads')
    MAX_CONTENT_LENGTH = 200 * 1024 * 1024  # 200MB max file size
    ALLOWED_EXTENSIONS = {'pdf', 'doc', 'docx', 'ppt', 'pptx', 'xls', 'xlsx', 'txt', 'md', 'png', 'jpg', 'jpeg', 'gif', 'webp'}
 
    CHUNK_SIZE = 500  # Smaller chunks for higher precision with many documents
    CHUNK_OVERLAP = 150  # Good overlap to avoid losing info at boundaries
    TOP_K_RESULTS = 100  # High - comprehensive retrieval for 3000+ docs across buckets
    AI_TEMPERATURE = 0.0  # Zero temperature for maximum determinism and accuracy
    RELEVANCE_THRESHOLD = 3.0  # Higher threshold - include all potentially relevant
    MAX_CONVERSATION_HISTORY = 20  # Remember more conversation for pronoun context
    AI_MAX_TOKENS = 4096  # Maximum tokens for detailed responses
    AI_TIMEOUT = 90  # More time for complex multi-document queries