config.py

import os


class Config:
    OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434")

    # Ollama Embedding Model Settings
    OLLAMA_EMBEDDING_MODEL = os.getenv("OLLAMA_EMBEDDING_MODEL", "mxbai-embed-large:latest")
    OLLAMA_EMBEDDING_MODEL_DIM = int(os.getenv("OLLAMA_EMBEDDING_MODEL_DIM", "1024"))

    # Ollama Generative Model Settings
    OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "qwen3:8b")
    OLLAMA_TIMEOUT = float(os.getenv("OLLAMA_TIMEOUT", "120.0"))
    OLLAMA_THINKING = os.getenv("OLLAMA_THINKING", "False").lower() in (
        "true",
        "1",
        "t",
    )
    # Temperature controls randomness of outputs: 0.0 is deterministic, 1.0 is creative
    OLLAMA_TEMPERATURE = float(os.getenv("OLLAMA_TEMPERATURE", "0.3"))
    # Default context_window is -1 (infer from model)
    # Increasing the context window allows the model to consider more prior conversation
    # For our case, if the data sample + schema are large, we may need a bigger context window
    # otherwise the model starts to hallucinate. For example, it took the data sample in
    # consideration, but partially ignored the schema when generating metadata.
    # Note, that different models have different max context windows. For Qwen3 8B it's 40960
    OLLAMA_CONTEXT_WINDOW = int(os.getenv("OLLAMA_CONTEXT_WINDOW", "40960"))

    # MCP Settings
    MCP_CLIENT_URL = os.getenv("MCP_CLIENT_URL", "http://127.0.0.1:5337/sse")

    # Agent Settings
    AGENT_MAX_ITERATIONS = int(os.getenv("AGENT_MAX_ITERATIONS", "20"))

    # PostgreSQL Settings
    PG_URI = os.getenv("PG_URI", "postgresql://postgres:postgres@localhost/ai_pg_vector_db")
    PG_DB_NAME = os.getenv("PG_DB_NAME", "ai_pg_vector_db")
    PG_TABLE_NAME = os.getenv("PG_TABLE_NAME", "llamaindex")