memory_bank/summarize_memory.py

import os
import json
import re
from openai import OpenAI

# --- Configuration & Client Setup ---
api_key = os.environ.get("GAPGPT_API_KEY")
base_url = "https://api.gapgpt.app/v1"

if not api_key:
    print("⚠️ Warning: 'GAPGPT_API_KEY' not found. Checking for 'OPENAI_API_KEY'...")
    api_key = os.environ.get("OPENAI_API_KEY")
    if api_key:
        print("ℹ️ Using 'OPENAI_API_KEY' instead. Switching base_url to default.")
        base_url = None
    else:
        print("❌ CRITICAL ERROR: No API Key found. LLM calls will fail.")

if base_url:
    client = OpenAI(api_key=api_key, base_url=base_url)
else:
    client = OpenAI(api_key=api_key)


def get_llm_response(prompt):
    """Helper to call LLM and handle basic errors."""
    if not client.api_key:
        return "{}"

    try:
        response = client.chat.completions.create(
            model="gpt-4o",
            messages=[
                {"role": "system", "content": "You are a JSON generator. Output ONLY valid JSON. No markdown, no explanations."},
                {"role": "user", "content": prompt}
            ],
            temperature=0.3,
            max_tokens=1000
        )
        content = response.choices[0].message.content
        if not content:
            print("⚠️ LLM returned empty content.")
            return "{}"
        return content
    except Exception as e:
        print(f"❌ Error calling LLM: {e}")
        return "{}"

def clean_json_text(text):
    """
    Robustly extracts JSON object from text using Regex.
    Finds the substring starting with '{' and ending with '}'.
    """
    if not text: return ""
    
    # First, try to remove markdown code blocks if present
    text = text.replace('```json', '').replace('```', '')
    
    # Use Regex to find the outer-most JSON object
    match = re.search(r'\{.*\}', text, re.DOTALL)
    if match:
        return match.group(0)
    
    return text.strip()

def update_episodic_memory_only(memory_data, user_name):
    """
    Analyzes the LATEST session to extract: Topics, Emotions, Insights.
    """
    print(f"⚡ Processing Episodic Memory for session end: {user_name}")
    
    if user_name not in memory_data:
        return memory_data

    user_data = memory_data[user_name]
    sessions = user_data.get("sessions", [])

    if not sessions:
        return memory_data

    last_session = sessions[-1]
    
    if last_session.get("analyzed", False):
        print(f"   Session {last_session.get('session_id')} already analyzed.")
        return memory_data
    
    conversation = last_session.get("conversation", [])
    if not conversation:
        return memory_data

    print("   ... ⚡ Generating Episodic Memory (Calling LLM) ...")

    conversation_text = json.dumps(conversation, ensure_ascii=False)

    prompt = f"""
    Analyze this chat session between AI (Emma) and user ({user_name}).
    
    Conversation:
    {conversation_text}

    Return a JSON object with these exact keys:
    {{
        "topics": ["list", "of", "topics"],
        "emotion": "single string emotion",
        "key_insights": ["list", "of", "insights"]
    }}
    """

    response_text = get_llm_response(prompt)
    cleaned_text = clean_json_text(response_text)

    try:
        analysis_result = json.loads(cleaned_text)
    except json.JSONDecodeError as e:
        print(f"   ⚠️ JSON Error: {e}")
        print(f"   📝 Raw LLM Response was: {response_text}") # Debug print
        analysis_result = {"topics": [], "emotion": "Unknown", "key_insights": []}

    episode_entry = {
        "session_id": last_session.get("session_id"),
        "date": last_session.get("date"),
        "topics": analysis_result.get("topics", []),
        "emotion": analysis_result.get("emotion", "Neutral"),
        "insights": analysis_result.get("key_insights", [])
    }

    if "episodic_memory" not in user_data:
        user_data["episodic_memory"] = []

    user_data["episodic_memory"].append(episode_entry)
    last_session["analyzed"] = True
    
    print(f"   ✅ Episodic memory successfully updated.")
    return memory_data

def update_semantic_memory_only(memory_data, user_name):
    """
    Analyzes ALL episodic memories to update the Semantic Profile.
    """
    print(f"🧠 Processing Semantic Memory for: {user_name}")

    if user_name not in memory_data:
        return memory_data

    user_data = memory_data[user_name]
    episodes = user_data.get("episodic_memory", [])

    if not episodes:
        print("   No episodic memory found.")
        return memory_data

    episodes_text = json.dumps(episodes[-10:], ensure_ascii=False)

    prompt = f"""
    Based on these past session analyses for user '{user_name}', update their profile.

    History:
    {episodes_text}

    Current Profile:
    {user_data.get('semantic_memory', 'None')}

    Return JSON:
    {{
        "big_five": {{
            "openness": "Low/Medium/High",
            "conscientiousness": "Low/Medium/High",
            "extraversion": "Low/Medium/High",
            "agreeableness": "Low/Medium/High",
            "neuroticism": "Low/Medium/High"
        }},
        "core_values": ["value1", "value2"],
        "summary": "Concise personality summary."
    }}
    """

    response_text = get_llm_response(prompt)
    cleaned_text = clean_json_text(response_text)

    try:
        semantic_result = json.loads(cleaned_text)
        if "summary" in semantic_result:
            user_data["semantic_profile"] = semantic_result
            user_data["semantic_memory"] = semantic_result["summary"]
            print(f"   ✅ Semantic memory updated.")
        else:
            print("   ⚠️ Semantic update failed: 'summary' key missing.")
            
    except json.JSONDecodeError as e:
        print(f"   ⚠️ JSON Error: {e}")
        print(f"   📝 Raw LLM Response was: {response_text}")

    return memory_data

def summarize_memory(memory_data, user_name):
    data = update_episodic_memory_only(memory_data, user_name)
    data = update_semantic_memory_only(data, user_name)
    return data