Update config.py

config.py

@@ -47,16 +47,10 @@ BASE_DIR = Path(__file__).parent
 # ============================================================================
 # DATABASE PATHS
 # ============================================================================
-# These directories store different types of data:
-# - learning_data: Text files with information about the user (personal data, preferences, etc.)
-# - chats_data: JSON files containing past conversation history
-# - vector_store: FAISS index files for fast similarity search
 LEARNING_DATA_DIR = BASE_DIR / "database" / "learning_data"
 CHATS_DATA_DIR = BASE_DIR / "database" / "chats_data"
 VECTOR_STORE_DIR = BASE_DIR / "database" / "vector_store"
 
-# Create directories if they don't exist so the app can run without manual setup.
-# parents=True creates parent folders; exist_ok=True avoids error if already present.
 LEARNING_DATA_DIR.mkdir(parents=True, exist_ok=True)
 CHATS_DATA_DIR.mkdir(parents=True, exist_ok=True)
 VECTOR_STORE_DIR.mkdir(parents=True, exist_ok=True)
@@ -64,87 +58,47 @@ VECTOR_STORE_DIR.mkdir(parents=True, exist_ok=True)
 # ============================================================================
 # GROQ API CONFIGURATION
 # ============================================================================
-# Groq is the LLM provider we use for generating responses.
-# You can set one key (GROQ_API_KEY) or multiple keys for fallback:
-#   GROQ_API_KEY, GROQ_API_KEY_2, GROQ_API_KEY_3, ... (no upper limit).
-# PRIMARY-FIRST: Every request tries the first key first. If it fails (rate limit,
-# timeout, etc.), the server tries the second, then third, until one succeeds.
-# If all keys fail, the user receives a clear error message.
-# Model determines which AI model to use (llama-3.3-70b-versatile is latest).
-
 def _load_groq_api_keys() -> list:
-    """
-    Load all GROQ API keys from the environment.
-    Reads GROQ_API_KEY first, then GROQ_API_KEY_2, GROQ_API_KEY_3, ... until
-    a number has no value. There is no upper limit on how many keys you can set.
-    Returns a list of non-empty key strings (may be empty if GROQ_API_KEY is not set).
-    """
     keys = []
-    # First key: GROQ_API_KEY (required in practice; validated when building services).
     first = os.getenv("GROQ_API_KEY", "").strip()
     if first:
         keys.append(first)
-
-    # Additional keys: GROQ_API_KEY_2, GROQ_API_KEY_3, GROQ_API_KEY_4, ...
     i = 2
     while True:
         k = os.getenv(f"GROQ_API_KEY_{i}", "").strip()
         if not k:
-            # No key for this number; stop (no more keys).
             break
         keys.append(k)
         i += 1
-
     return keys
 
 GROQ_API_KEYS = _load_groq_api_keys()
-# Backward compatibility: single key name still used in docs; code uses GROQ_API_KEYS.
 GROQ_API_KEY = GROQ_API_KEYS[0] if GROQ_API_KEYS else ""
-
 GROQ_MODEL = os.getenv("GROQ_MODEL", "llama-3.3-70b-versatile")
 
 # ============================================================================
 # TAVILY API CONFIGURATION
 # ============================================================================
-# Tavily is a fast, AI-optimized search API designed for LLM applications
-# Get API key from: https://tavily.com (free tier available)
-# Tavily returns English-only results by default and is faster than DuckDuckGo
 TAVILY_API_KEY = os.getenv("TAVILY_API_KEY", "")
 
 # ============================================================================
 # BRAIN MODEL (Query Classification — Jarvis Mode)
 # ============================================================================
-# The brain classifies each query as "general" or "realtime" using Groq.
-# Uses the same GROQ_API_KEYS with rotation (brain and chat never use the same key).
 GROQ_BRAIN_MODEL = os.getenv("GROQ_BRAIN_MODEL", "llama-3.1-8b-instant")
 
 # ============================================================================
 # TTS (TEXT-TO-SPEECH) CONFIGURATION
 # ============================================================================
-# edge-tts uses Microsoft Edge's free cloud TTS. No API key needed.
-# Voice list: run `edge-tts --list-voices` to see all available voices.
-# Default: en-GB-RyanNeural (male British voice, fitting for JARVIS).
-# Override via TTS_VOICE in .env (e.g. TTS_VOICE=en-US-ChristopherNeural).
 TTS_VOICE = os.getenv("TTS_VOICE", "en-GB-RyanNeural")
 TTS_RATE = os.getenv("TTS_RATE", "+22%")
 
 # ============================================================================
 # EMBEDDING CONFIGURATION
 # ============================================================================
-# Embeddings convert text into numerical vectors that capture meaning
-# We use HuggingFace's sentence-transformers model (runs locally, no API needed)
-# CHUNK_SIZE: How many characters to split documents into
-# CHUNK_OVERLAP: How many characters overlap between chunks (helps maintain context)
 EMBEDDING_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
-CHUNK_SIZE = 1000  # Characters per chunk
-CHUNK_OVERLAP = 200  # Overlap between chunks
-
-# Maximum conversation turns (user+assistant pairs) sent to the LLM per request.
-# Older turns are kept on disk but not sent to avoid context/token limits.
+CHUNK_SIZE = 1000  
+CHUNK_OVERLAP = 200  
 MAX_CHAT_HISTORY_TURNS = 20
-
-# Maximum length (characters) for a single user message. Prevents token limit errors
-# and abuse. ~32K chars ≈ ~8K tokens; keeps total prompt well under model limits.
 MAX_MESSAGE_LENGTH = 32_000
 
 # ============================================================================
@@ -235,16 +189,9 @@ def load_user_context() -> str:
     """
     Load and concatenate the contents of all .txt files in learning_data.
     Reads every .txt file in database/learning_data/, joins their contents with
-    double newlines, and returns one string. Used by code that needs the raw
-    learning text (e.g. optional utilities). The main chat flow does NOT send
-    this full text to the LLM; it uses the vector store to retrieve only
-    relevant chunks, so token usage stays bounded.
-
-    Returns:
-        str: Combined content from all .txt files, or "" if none exist or all fail to read.
+    double newlines, and returns one string.
     """
     context_parts = []
-    # Sorted by path so the order is always the same across runs.
     text_files = sorted(LEARNING_DATA_DIR.glob("*.txt"))
     for file_path in text_files:
         try:
@@ -254,6 +201,4 @@ def load_user_context() -> str:
                     context_parts.append(content)
         except Exception as e:
             logger.warning("Could not load learning data file %s: %s", file_path, e)
-
-    # Join all file contents with double newline; empty string if no files or all failed.
     return "\n\n".join(context_parts) if context_parts else ""