agent_engine.py

import os
import logging
import requests
from time import perf_counter, sleep
from .memory_manager import embed_and_store, retrieve_relevant

# Agent prompts
PROMPTS = {
    "Initiator": "You are the Discussion Initiator...",
    "Responder": "You are the Critical Responder...",
    "Guardian": "You are the Depth Guardian...",
    "Provocateur": "You are the Cross-Disciplinary Provocateur...",
    "Cultural": "You are the Cultural Perspective...",
    "Judge": "You are the Impartial Judge..."
}

CHAT_MODEL = os.environ.get("CHAT_MODEL", "HuggingFaceH4/zephyr-7b-beta")
HF_API_TOKEN = os.environ.get("HF_API_TOKEN", "")
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")


def safe_chat(system_prompt: str, history: list, temperature: float = 0.7) -> str:
    """Call HF inference API with timing and error handling."""
    start = perf_counter()
    payload = {
        "inputs": [{"role": "system", "content": system_prompt}] + history,
        "parameters": {"max_new_tokens": 300, "temperature": temperature}
    }
    headers = {"Authorization": f"Bearer {HF_API_TOKEN}"} if HF_API_TOKEN else {}
    try:
        resp = requests.post(
            f"https://api-inference.huggingface.co/models/{CHAT_MODEL}",
            json=payload,
            headers=headers,
            timeout=60
        )
        if resp.status_code == 200:
            data = resp.json()
            text = data[0].get('generated_text', '').strip()
        elif resp.status_code == 503:
            logging.warning("Model loading, retrying...")
            sleep(15)
            return safe_chat(system_prompt, history, temperature)
        else:
            logging.error(f"HF error {resp.status_code}: {resp.text}")
            text = f"⚠️ API Error {resp.status_code}"
    except Exception as e:
        logging.error(f"safe_chat exception: {e}")
        text = f"⚠️ System Error: {e}"
    elapsed = perf_counter() - start
    logging.info(f"safe_chat: {elapsed:.3f}s for '{system_prompt[:30]}...'")
    return text


def step_turn(conversation: list, turn: int, topic: str, params: dict) -> list:
    """Advance one turn of the multi-agent conversation."""
    # Choose agent by sequence
    sequence = ["Initiator", "Responder", "Guardian", "Provocateur", "Cultural"]
    agent = sequence[turn % len(sequence)]
    prompt = PROMPTS.get(agent, "")
    # Prepare history
    history = [{"role": "user", "content": msg['text']} for msg in conversation[-5:]]
    response = safe_chat(prompt, history, temperature=params[agent]['creativity'])
    embed_and_store(response, agent, topic)
    conversation.append({"agent": agent, "text": response, "turn": turn + 1})
    return conversation