import os # --- Directory Configuration --- _BASE_DIR = os.path.dirname(os.path.dirname(__file__)) DATA_DIR = os.environ.get("DATA_DIR", _BASE_DIR) MARKDOWN_DIR = os.path.join(DATA_DIR, "markdown_docs") PARENT_STORE_PATH = os.path.join(DATA_DIR, "parent_store") QDRANT_DB_PATH = os.path.join(DATA_DIR, "qdrant_db") # --- Qdrant Configuration --- CHILD_COLLECTION = "document_child_chunks" SPARSE_VECTOR_NAME = "sparse" # --- Model Configuration --- DENSE_MODEL = os.environ.get("DENSE_MODEL", "sentence-transformers/all-MiniLM-L6-v2") SPARSE_MODEL = "Qdrant/bm25" LLM_MODEL = os.environ.get("GROQ_MODEL", "openai/gpt-oss-20b") JUDGE_MODEL = "ministral-3:3b-instruct-2512-q8_0" LLM_TEMPERATURE = 0 LLM_SEED = 42 LLM_TIMEOUT = 60 LLM_MAX_RETRIES = 2 # --- Retrieval Configuration --- RETRIEVAL_SCORE_THRESHOLD = 0.4 DEFAULT_RETRIEVAL_K = 7 CHILD_CHUNK_SEPARATOR = "\n\n\n\n" # --- Agent Configuration --- MAX_TOOL_CALLS = 8 MAX_ITERATIONS = 10 GRAPH_RECURSION_LIMIT = 50 MAIN_HISTORY_MESSAGES_TO_KEEP = 4 BASE_TOKEN_THRESHOLD = 2000 TOKEN_GROWTH_FACTOR = 0.9 # --- Terminal Execution Logging --- EXECUTION_LOGGING_ENABLED = False EXECUTION_LOG_MAX_CHARS = 1200 EXECUTION_LOG_USE_COLOR = True # --- Text Splitter Configuration --- CHILD_CHUNK_SIZE = 500 CHILD_CHUNK_OVERLAP = 100 MIN_PARENT_SIZE = 2000 MAX_PARENT_SIZE = 4000 HEADERS_TO_SPLIT_ON = [ ("#", "H1"), ("##", "H2"), ("###", "H3") ] # --- Langfuse Observability --- LANGFUSE_ENABLED = os.environ.get("LANGFUSE_ENABLED", "false").lower() == "true" LANGFUSE_PUBLIC_KEY = os.environ.get("LANGFUSE_PUBLIC_KEY", "") LANGFUSE_SECRET_KEY = os.environ.get("LANGFUSE_SECRET_KEY", "") LANGFUSE_BASE_URL = os.environ.get("LANGFUSE_BASE_URL", "http://localhost:3000")