agents.py

import os
import json
import requests

# Set your API Key securely
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
LLAMA_MODEL = "llama3-8b-8192"
SLM_MODEL = "llama3-8b-8192"

# ---------------------------
# Groq API Call Wrapper
# ---------------------------
def call_groq(prompt, model=LLAMA_MODEL):
    response = requests.post(
        "https://api.groq.com/openai/v1/chat/completions",
        headers={"Authorization": f"Bearer {GROQ_API_KEY}"},
        json={
            "model": model,
            "messages": [{"role": "user", "content": prompt}],
            "temperature": 0.7
        }
    )
    if response.status_code != 200:
        return f"[Groq API Error {response.status_code}]: {response.text}"

    try:
        data = response.json()
        return data.get("choices", [{}])[0].get("message", {}).get("content", "[No valid response]")
    except Exception as e:
        return f"[Groq Parsing Error]: {str(e)}"

# ---------------------------
# Empathetic First-Aider Chat Agent
# ---------------------------
def run_first_aider(message, mood):
    prompt = f"""
You're a warm and respectful AI listener. Respond to this user's message kindly and briefly, in 1-2 sentences.
- Never give advice.
- Be supportive, clear, emotionally kind, and safe.
- Maintain boundaries.

User mood: {mood}
Message: "{message}"
"""
    return call_groq(prompt, model=SLM_MODEL)

# ---------------------------
# Context-Aware Introspect Agent
# ---------------------------
def get_user_context():
    context = ""

    if os.path.exists("chat_log.json"):
        with open("chat_log.json") as f:
            chats = json.load(f)[-3:]  # last 3 chat messages
            context += "\nRecent Conversations:\n"
            for c in chats:
                context += f"User: {c['user']}\nAI: {c['ai']}\n"

    if os.path.exists("journal_log.json"):
        with open("journal_log.json") as f:
            logs = json.load(f)[-2:]
            context += "\nJournal Entries:\n"
            for j in logs:
                context += f"Mood: {j['mood']}\nEntry: {j['entry']}\nAI Response: {j['response']}\n"

    return context.strip() if context.strip() else None

def run_introspect(message, mood):
    context = get_user_context()
    if not context:
        return "Let's talk a bit or journal first so I can help you reflect better."

    prompt = f"""
You're a calm, thoughtful AI helping a user gently reflect on their emotional patterns.

Context from past chats and journals:
{context}

User's new message: "{message}"
Mood: {mood}

Instructions:
- Highlight any potential recurring emotional themes (gently).
- Suggest a new way to think about the situation (without naming therapy techniques).
- End with a kind affirmation or journaling suggestion.

Be kind, short, helpful, and never mention psychology, CBT, or NLP.
"""
    return call_groq(prompt, model=LLAMA_MODEL)

# ---------------------------
# Journaling Agent (Context-Aware, CBT/NLP Masked)
# ---------------------------
def get_mood_context():
    if not os.path.exists("journal_log.json"):
        return []
    with open("journal_log.json") as f:
        logs = json.load(f)
        return [x["mood"] for x in logs[-3:]]

def run_journaling_pipeline(mood, entry, mode):
    recent_moods = get_mood_context()
    mood_summary = f"Recent moods: {', '.join(recent_moods)}." if recent_moods else "No mood history."
    reflective_prompt = "What patterns have you noticed in how you've been feeling lately?" if recent_moods else "What stood out to you emotionally today?"

    prompt = f"""
You're a reflective journaling assistant. Help the user explore their thoughts kindly and safely.

User wrote:
"{entry}"
Journaling Mode: {mode}
Mood: {mood}
{mood_summary}
Prompt: {reflective_prompt}

Instructions:
- Reflect what the user might be feeling or thinking.
- Suggest gentle rephrasing or ways to understand the situation better.
- End with a kind affirmation or open-ended journaling question.
- Do not mention therapy, CBT, NLP, psychology, or diagnosis.

Be gentle and warm.
"""
    response = call_groq(prompt, model=LLAMA_MODEL)
    return {
        "entry": entry,
        "mood": mood,
        "mode": mode,
        "response": response
    }