my_voice_assistant_v3_jamulix/app/api/chat.py

from io import BytesIO

from fastapi import APIRouter, Depends, HTTPException, Query
from fastapi.responses import JSONResponse, StreamingResponse

from app.config import settings
from app.errors import RoutingError
from app.auth import require_user
from app.store import ANONYMOUS_USER_ID, User, SessionOwnershipError
from app.dependencies import (
    resolve_route,
    build_orchestrator,
    resolve_output_endpoint,
    get_store,
    piper_voice_for_language,
)
from app.core.memory_extractor import maybe_schedule_extraction
from app.quota import enforce_quota, record_usage, QuotaExceededError
from app.safety.emergency import handle_emergency, schedule_llm_emergency_check
from app.schemas import ChatRequest

router = APIRouter()


def _route_headers(route) -> dict:
    return {
        "X-Input-Endpoint": route.input_endpoint,
        "X-Output-Endpoint": route.output_endpoint,
        "X-STT-Provider": route.stt_provider,
        "X-LLM-Provider": route.llm_provider,
        "X-TTS-Provider": route.tts_provider,
    }


@router.post("/chat")
async def chat(
    payload: ChatRequest,
    debug: bool = Query(
        default=False,
        description="Return JSON trace instead of audio response",
    ),
    session_id: str | None = Query(
        default=None,
        description="Optional session id to apply a stored route",
    ),
    user: User = Depends(require_user),
):
    overrides = {
        "input_endpoint": payload.input_endpoint,
        "output_endpoint": payload.output_endpoint,
        "language": payload.language,
        "language_mode": payload.language_mode,
        "stt_provider": payload.stt_provider,
        "llm_provider": payload.llm_provider,
        "tts_provider": payload.tts_provider,
    }
    store = get_store()

    try:
        route = resolve_route(user, session_id, overrides)
        orchestrator = build_orchestrator(route)
        output = await resolve_output_endpoint(route)
        # Gespraechsverlauf laden (nur bei gesetzter session_id -> sonst zustandslos).
        conversation = (
            store.get_recent_messages(session_id, settings.history_max_messages)
            if session_id
            else []
        )
        # Langzeit-Erinnerungen sind nutzerbezogen und gelten auch ohne Session.
        memories = store.get_memories(user.id)
    except SessionOwnershipError as exc:
        raise HTTPException(status_code=403, detail=str(exc))
    except RoutingError as exc:
        raise HTTPException(status_code=422, detail=str(exc))

    llm_context = list(conversation)
    user_context_parts = []
    if user.id != ANONYMOUS_USER_ID:
        user_context_parts.append(f"Du sprichst mit {user.display_name}.")
    if memories:
        user_context_parts.append(
            "Was du ueber den Nutzer weisst:\n" + "\n".join(f"- {m.content}" for m in memories)
        )
    if user_context_parts:
        llm_context = [{"role": "system", "content": "\n".join(user_context_parts)}] + llm_context

    # Notfall-Erkennung zuerst (immer eskalieren, auch bei Quota-Limit).
    emergency = handle_emergency(user, payload.text, store)
    # Stufe 2: LLM-Klassifikation als Hintergrund-Task (nur wenn Stichwoerter nichts fanden).
    schedule_llm_emergency_check(user, payload.text, store, emergency)

    if emergency is None:
        try:
            enforce_quota(user, store)
        except QuotaExceededError as exc:
            raise HTTPException(status_code=429, detail=str(exc))

    # Explizit angeforderte Stimme gewinnt; sonst folgt sie der Routensprache.
    voice = payload.voice or piper_voice_for_language(route.tts_provider, route.language)

    try:
        trace, audio = await orchestrator.chat_text(
            payload.text,
            language=route.language,
            voice=voice,
            output=output,
            history=llm_context,
            language_mode=route.language_mode,
            text_only=bool(payload.text_only),
        )
    except Exception as exc:
        raise HTTPException(status_code=502, detail=str(exc))

    record_usage(user, store, len(payload.text) + len(trace.semantic_response or ""))

    # Turn persistieren (User-Eingabe + semantische Antwort) fuer das Gedaechtnis.
    if session_id:
        store.append_message(session_id, user.id, "user", payload.text)
        store.append_message(session_id, user.id, "assistant", trace.semantic_response)
        maybe_schedule_extraction(store, user.id, session_id)

    if debug:
        return JSONResponse(
            content={
                "ok": True,
                "voice": voice,
                "route": route.as_dict(),
                "history_len": len(conversation),
                "memories_len": len(memories),
                "emergency": emergency,
                "trace": {
                    "raw_transcript": trace.raw_transcript,
                    "cleaned_transcript": trace.cleaned_transcript,
                    "semantic_response": trace.semantic_response,
                    "spoken_response": trace.spoken_response,
                    "tts_ready_text": trace.tts_ready_text,
                },
            }
        )

    headers = {
        "Content-Language": route.language,
        "X-Audio-Format": "pcm",
        "X-Audio-Sample-Rate": "24000",
        "X-Audio-Channels": "1",
        "X-Audio-Sample-Width": "16",
        **_route_headers(route),
    }
    if emergency:
        headers["X-Emergency"] = emergency["category"]
    return StreamingResponse(BytesIO(audio), media_type="audio/pcm", headers=headers)