"""
backend/agent/router.py — HF Cloud Version
Full intelligent routing with NVIDIA fallback, Emergency Replay bypass, and Omni-Memory sync.
"""
import asyncio
import logging
from backend.agent.personas import inject_persona
from backend.ws.ws_manager import ws_manager


class LLMRouter:
    async def process_transcript(self, transcript: str, active_persona: str, task_type: str = "general"):
        logging.info(f"Router: Received transcript for {active_persona}: '{transcript[:80]}...'")

        # ── EMERGENCY REPLAY BYPASS ──────────────────────────────────────────
        # If the transcript is a direct replay command, bypass LLM entirely.
        replay_triggers = ["replay", "what did you just say", "repeat that", "say that again"]
        if any(t in transcript.lower() for t in replay_triggers):
            await ws_manager.emit("agent:emergency_replay", {"trigger": transcript})
            logging.info("Router: Emergency Replay triggered — bypassing LLM.")
            return

        # ── PERSONA CONTEXT INJECTION ────────────────────────────────────────
        context = inject_persona(active_persona, transcript)

        # ── OMNI-MEMORY HIVE MIND CONTEXT ────────────────────────────────────
        try:
            from backend.services.memory_service import GlobalOmniMemory
            omni_context = GlobalOmniMemory.get_global_context_stream()
            if omni_context:
                context = omni_context + context
        except Exception as e:
            logging.warning(f"Router: OmniMemory unavailable: {e}")

        # ── INTELLIGENT TASK CLASSIFICATION ──────────────────────────────────
        transcript_lower = transcript.lower()
        if any(k in transcript_lower for k in ["code", "debug", "function", "class", "python", "rust", "fix bug"]):
            task_type = "heavy_compute"
        elif any(k in transcript_lower for k in ["plan", "research", "schedule", "remind", "analyse"]):
            task_type = "agentic"
        elif any(k in transcript_lower for k in ["look at", "see", "image", "camera", "screenshot", "video"]):
            task_type = "multimodal"
        elif any(k in transcript_lower for k in ["password", "private", "secure", "pii", "personal"]):
            task_type = "specialized"

        # ── TOKEN MANAGER ROUTING (Google Primary → NVIDIA Fallback) ─────────
        try:
            from backend.services.token_manager import run_task
            response_text = await asyncio.to_thread(run_task, context, task_type=task_type)
        except Exception as e:
            logging.error(f"Router: Token Manager failed: {e}")
            response_text = "I encountered an issue processing that request, sir."

        # ── STREAM RESPONSE TO FRONTEND ──────────────────────────────────────
        words = response_text.split()
        for i, word in enumerate(words):
            token = word + (" " if i < len(words) - 1 else "")
            await ws_manager.emit("agent:token", {"token": token, "persona": active_persona})
            await asyncio.sleep(0.02)

        await ws_manager.emit("agent:finished", {"full_text": response_text})

        # ── TTS SYNTHESIS ─────────────────────────────────────────────────────
        try:
            from backend.voice.tts import synthesize
            await synthesize(response_text)
        except Exception as e:
            logging.warning(f"Router: TTS unavailable on HF: {e}")

        # ── RECORD TO OMNI MEMORY ─────────────────────────────────────────────
        try:
            from backend.services.memory_service import GlobalOmniMemory
            GlobalOmniMemory.record_action("CLOUD_HF", "router", task_type, response_text[:100].replace('\n', ' ') + "...")
        except Exception as e:
            logging.warning(f"Router: OmniMemory record failed: {e}")


router = LLMRouter()