"""
Configuration for A-MEM System
"""
import os
from pathlib import Path
from dotenv import load_dotenv
# Lade .env Datei
load_dotenv()
class Config:
# Paths - Use absolute path based on file location, not cwd
# This ensures the paths work regardless of where the script is called from
_config_file = Path(__file__).resolve()
BASE_DIR = _config_file.parent.parent.parent # Go up from src/a_mem/config.py to project root
DATA_DIR = BASE_DIR / "data"
CHROMA_DIR = DATA_DIR / "chroma"
GRAPH_DIR = DATA_DIR / "graph"
GRAPH_PATH = GRAPH_DIR / "knowledge_graph.json"
# LLM Provider Selection
LLM_PROVIDER = os.getenv("LLM_PROVIDER", "ollama").lower() # "ollama" oder "openrouter"
# Ollama Settings
OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434")
OLLAMA_LLM_MODEL = os.getenv("OLLAMA_LLM_MODEL", "qwen3:4b")
OLLAMA_EMBEDDING_MODEL = os.getenv("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text:latest")
# OpenRouter Settings
OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY", "")
OPENROUTER_BASE_URL = os.getenv("OPENROUTER_BASE_URL", "https://openrouter.ai/api/v1")
OPENROUTER_LLM_MODEL = os.getenv("OPENROUTER_LLM_MODEL", "openai/gpt-4o-mini")
OPENROUTER_EMBEDDING_MODEL = os.getenv("OPENROUTER_EMBEDDING_MODEL", "openai/text-embedding-3-small")
# Model Settings (kompatibel mit altem Code)
@property
def LLM_MODEL(self):
"""Returns the current LLM model (provider-dependent)"""
if self.LLM_PROVIDER == "openrouter":
return self.OPENROUTER_LLM_MODEL
return self.OLLAMA_LLM_MODEL
@property
def EMBEDDING_MODEL(self):
"""Returns the current embedding model (provider-dependent)"""
if self.LLM_PROVIDER == "openrouter":
return self.OPENROUTER_EMBEDDING_MODEL
return self.OLLAMA_EMBEDDING_MODEL
# Retrieval Settings
MAX_NEIGHBORS = int(os.getenv("MAX_NEIGHBORS", "5"))
MIN_SIMILARITY_SCORE = float(os.getenv("MIN_SIMILARITY_SCORE", "0.4"))
# Concurrency
LOCK_FILE = GRAPH_DIR / "graph.lock"
# TCP Server Settings (optional, für Script-Zugriff)
TCP_SERVER_ENABLED = os.getenv("TCP_SERVER_ENABLED", "false").lower() == "true"
TCP_SERVER_HOST = os.getenv("TCP_SERVER_HOST", "127.0.0.1")
TCP_SERVER_PORT = int(os.getenv("TCP_SERVER_PORT", "42424"))
# Researcher Agent Settings
RESEARCHER_ENABLED = os.getenv("RESEARCHER_ENABLED", "false").lower() == "true"
RESEARCHER_CONFIDENCE_THRESHOLD = float(os.getenv("RESEARCHER_CONFIDENCE_THRESHOLD", "0.5"))
RESEARCHER_MAX_SOURCES = int(os.getenv("RESEARCHER_MAX_SOURCES", "5"))
RESEARCHER_MAX_CONTENT_LENGTH = int(os.getenv("RESEARCHER_MAX_CONTENT_LENGTH", "10000"))
# Jina Reader Settings (local Docker)
JINA_READER_ENABLED = os.getenv("JINA_READER_ENABLED", "true").lower() == "true"
JINA_READER_HOST = os.getenv("JINA_READER_HOST", "localhost")
JINA_READER_PORT = int(os.getenv("JINA_READER_PORT", "2222"))
JINA_READER_URL = f"http://{JINA_READER_HOST}:{JINA_READER_PORT}"
# Unstructured Settings (for PDF extraction)
UNSTRUCTURED_ENABLED = os.getenv("UNSTRUCTURED_ENABLED", "true").lower() == "true"
UNSTRUCTURED_API_URL = os.getenv("UNSTRUCTURED_API_URL", "http://localhost:8000")
UNSTRUCTURED_API_KEY = os.getenv("UNSTRUCTURED_API_KEY", "")
UNSTRUCTURED_USE_LIBRARY = os.getenv("UNSTRUCTURED_USE_LIBRARY", "true").lower() == "true" # Use library directly (default: true, falls back to API if library fails)
# Google Search API Settings (for web search)
GOOGLE_SEARCH_ENABLED = os.getenv("GOOGLE_SEARCH_ENABLED", "true").lower() == "true"
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY", "YOUR_KEY")
GOOGLE_SEARCH_ENGINE_ID = os.getenv("GOOGLE_SEARCH_ENGINE_ID", "c7bf0393f031a4691")
# Cache file for amem_stats --diff
AMEM_STATS_CACHE_FILE = DATA_DIR / "amem_stats_cache.json"
# Graph Backend Selection (networkx, rustworkx, or falkordb)
GRAPH_BACKEND = os.getenv("GRAPH_BACKEND", "networkx").lower() # "networkx", "rustworkx", or "falkordb"
def __init__(self):
self.DATA_DIR.mkdir(parents=True, exist_ok=True)
self.CHROMA_DIR.mkdir(parents=True, exist_ok=True)
self.GRAPH_DIR.mkdir(parents=True, exist_ok=True)
settings = Config()
