"""Create one real agent with our exact config, read it back, and print the
latency-relevant fields — then delete it. Verifies whether turn_timeout, the
TTS model, and the custom-LLM model actually landed (vs. being silently ignored
and falling back to ElevenLabs defaults like the 7s turn timeout).

Run (from backend/):  python -m scripts.agent_inspect
"""
from __future__ import annotations

import asyncio
import json

import httpx
from dotenv import load_dotenv

load_dotenv(override=True)  # .env wins over any stale shell var

from app.config import get_settings  # noqa: E402
from app.services import elevenlabs_client as E  # noqa: E402


async def main() -> None:
    settings = get_settings()
    if not (settings.voice_configured and settings.llm_configured):
        print("Need both ELEVENLABS_API_KEY and OPENROUTER_API_KEY in .env.")
        return

    print(f"Sending turn_timeout={settings.TURN_TIMEOUT}, "
          f"tts={settings.TTS_MODEL_ID}\n")

    agent_id = await E.create_session_agent(
        name="vion-inspect-temp",
        system_prompt="You are a person on a phone call. Reply briefly.",
        voice_id=E.resolve_voice_id("P1"),
    )
    print(f"Created agent {agent_id}; reading it back...\n")

    url = f"{settings.ELEVENLABS_BASE_URL.rstrip('/')}/v1/convai/agents/{agent_id}"
    headers = {"xi-api-key": settings.ELEVENLABS_API_KEY}
    try:
        async with httpx.AsyncClient(timeout=30) as client:
            resp = await client.get(url, headers=headers)
        cfg = resp.json().get("conversation_config", {})
        turn = cfg.get("turn", {})
        tts = cfg.get("tts", {})
        prompt = cfg.get("agent", {}).get("prompt", {})

        print("--- What ElevenLabs actually stored ---")
        print("turn:", json.dumps(turn))
        print("tts.model_id:", tts.get("model_id"))
        print("prompt.llm:", prompt.get("llm"))
        print("custom_llm.model_id:", prompt.get("custom_llm", {}).get("model_id"))
        print("prompt.max_tokens:", prompt.get("max_tokens"))
        print()

        stored = turn.get("turn_timeout")
        if stored == settings.TURN_TIMEOUT:
            print(f"OK: turn_timeout={stored} applied. The 10s is NOT the turn "
                  f"setting — chase the LLM (run latency_probe / dashboard).")
        else:
            print(f"PROBLEM: we sent {settings.TURN_TIMEOUT} but ElevenLabs has "
                  f"{stored!r}. The field nesting is wrong and it's defaulting. "
                  f"Paste this output back and I'll fix the structure.")
    finally:
        await E.delete_agent(agent_id)
        print(f"\nDeleted temp agent {agent_id}.")


if __name__ == "__main__":
    asyncio.run(main())