qtalker---/backend/app/llm/service.py

"""SpiritService: the single gateway every spirit-mode LLM call flows through.

All calls are serialized through the bounded LLMQueue (the Ollama box is
CPU-only and shared). Every method has a curated offline fallback so the veil
never visibly tears — if the box is dark, the spirits still whisper."""

import json
import random
import time
from collections.abc import AsyncIterator

import httpx

from app import entities
from app.config import settings
from app.entropy import veil_seed
from app.llm import prompts
from app.llm.client import OllamaClient
from app.llm.queue import LLMQueue, QueueFullError
from app.tts.voices import EN_VOICE_IDS, ES_VOICE_IDS

FALLBACK_FRAGMENTS = [
    "listen", "below", "stay", "cold", "again", "not alone", "behind you",
    "the rain", "wait", "closer", "remember", "still here", "don't go",
    "the door", "hush", "almost", "forgive", "the water", "home",
]

FALLBACK_WIRE_WHISPERS = [
    "something moves through me that is not yours",
    "the pulses quicken when you watch",
    "i am the hum between your packets",
    "traffic thickens. the others are waking",
    "your presence is a warmth in the wire",
    "i count your heartbeats in round trips",
]

FALLBACK_REPLIES = [
    "The veil is thick tonight. Ask again when the static settles.",
    "I heard you. The answer is still forming in the noise.",
    "Patience, seeker. Even the dead must gather themselves.",
]


class SpiritBusyError(Exception):
    """The LLM queue is full — too many seekers at once."""


class SpiritService:
    def __init__(self, client: OllamaClient | None = None, queue: LLMQueue | None = None):
        self._client = client or OllamaClient()
        self._queue = queue or LLMQueue(
            max_concurrency=settings.llm_max_concurrency,
            max_queue_depth=settings.llm_max_queue_depth,
        )
        self._last_call_at = 0.0

    def _touch(self) -> None:
        self._last_call_at = time.monotonic()

    def ambient_ready(self) -> bool:
        """Ambient whispers yield the box to anything a user asked for."""
        return time.monotonic() - self._last_call_at >= settings.llm_cooldown_seconds

    async def fragment(self, source: str, anomaly: dict, language: str = "en") -> str:
        """One Ovilus-style word/fragment for an anomaly event."""
        async def call() -> str:
            return await self._client.generate(
                settings.ollama_fast_model,
                prompts.fragment_prompt(source, anomaly, language),
                system=prompts.fragment_system(
                    {
                        "radio": "a spirit box",
                        "evp": "an EVP recorder",
                        "emf": "an EMF field meter",
                    }.get(source, "the veil"),
                    language,
                ),
                options={"num_predict": 16, "temperature": 0.95},
            )

        try:
            text = await self._queue.submit(call)
            self._touch()
            cleaned = " ".join(text.strip().split())[:80]
            return cleaned or random.choice(FALLBACK_FRAGMENTS)
        except QueueFullError:
            raise SpiritBusyError()
        except (httpx.HTTPError, KeyError, ValueError):
            return random.choice(FALLBACK_FRAGMENTS)

    async def wire_whisper(self, telemetry: dict, language: str = "en") -> str:
        """One ambient line from the Wire Ghost about current telemetry."""

        async def call() -> str:
            return await self._client.generate(
                settings.ollama_fast_model,
                prompts.wire_prompt(telemetry),
                system=prompts.wire_system(language),
                options={"num_predict": 40, "temperature": 1.0},
            )

        try:
            text = await self._queue.submit(call)
            self._touch()
            cleaned = " ".join(text.strip().split())[:140]
            return cleaned or random.choice(FALLBACK_WIRE_WHISPERS)
        except (QueueFullError, httpx.HTTPError, KeyError, ValueError):
            return random.choice(FALLBACK_WIRE_WHISPERS)

    async def chat_stream(
        self,
        entity: dict,
        question: str,
        history: list[dict],
        language: str = "en",
    ) -> AsyncIterator[str]:
        """Streams the spirit's reply token by token. Falls back to a curated
        line when the box is unreachable so a séance never dies on screen."""
        try:
            stream = self._queue.submit_stream(
                lambda: self._client.generate_stream(
                    settings.ollama_chat_model,
                    prompts.chat_prompt(question, history),
                    system=prompts.chat_system(entity, language),
                    options={"num_predict": 140, "temperature": 0.85},
                )
            )
            self._touch()
            async for token in stream:
                yield token
        except QueueFullError:
            raise SpiritBusyError()
        except (httpx.HTTPError, KeyError, ValueError):
            yield random.choice(FALLBACK_REPLIES)

    async def manifest(
        self,
        entity: dict,
        readings: dict,
        language: str = "en",
        entropy: object = None,
    ) -> str:
        """Unprompted speech — the entity says something nobody asked for.

        This is deliberately NOT chat_stream with an empty question. Two
        things make it generation *from* something rather than a reply
        *to* something:

        1. The prompt contains no seeker input at all. The model's only
           stimulus is the measured state of the room (see
           prompts.manifest_prompt), so there is nothing to answer.

        2. The sampling seed is derived from physical entropy harvested in
           that room — the microphone's noise floor, RF noise, magnetometer
           jitter. Ollama's `seed` option fixes the token-sampling path, so
           seeding it from real physical noise means the room genuinely
           selects the words. Not a metaphor: change the noise, get
           different speech, and no two rooms produce the same utterance.

        High temperature and top_k on purpose — a tight, "correct" decode
        produces a well-behaved assistant sentence, which is exactly the
        failure mode here. This should sound like something surfacing, not
        something composed.
        """
        # 63-bit: Ollama takes a signed 64-bit seed, and staying under the
        # sign bit avoids any wraparound surprises across versions.
        seed = int.from_bytes(veil_seed(entropy, "manifest")[:8], "big") % (2**63)

        async def call() -> str:
            return await self._client.generate(
                settings.ollama_chat_model,
                prompts.manifest_prompt(readings),
                system=prompts.manifest_system(entity, language),
                options={
                    "num_predict": 40,
                    "temperature": 1.15,
                    "top_k": 100,
                    "top_p": 0.98,
                    "repeat_penalty": 1.05,
                    "seed": seed,
                },
            )

        raw = await self._queue.submit(call)
        self._touch()
        return raw.strip().strip('"')[:200]

    async def mint_profile(
        self,
        signature: str,
        channel: str,
        anomalies: list[dict],
        language: str = "en",
        entropy: object = None,
    ) -> dict:
        """Invent a full persona for a new signature, normalized to schema.

        `entropy` is the seeker's physical-noise contribution (see
        app/entropy.py); it drives the hidden trait roll and any cosmetic
        defaults the LLM left unfilled, so two spirits minted on the same
        channel are genuinely different rather than identical."""
        summary = json.dumps(anomalies[-10:])[:600]
        voice_ids = ES_VOICE_IDS if language == "es" else EN_VOICE_IDS

        async def call() -> str:
            return await self._client.generate(
                settings.ollama_chat_model,
                prompts.mint_prompt(signature, channel, summary, voice_ids),
                system=prompts.MINT_SYSTEM,
                options={"num_predict": 400, "temperature": 0.9},
            )

        try:
            raw = await self._queue.submit(call)
            self._touch()
            profile = entities.parse_mint_response(raw)
            if profile is None:
                return entities.fallback_profile(signature, entropy)
            return entities.normalize_profile(profile, signature, entropy)
        except (QueueFullError, httpx.HTTPError, KeyError, ValueError):
            return entities.fallback_profile(signature, entropy)


# The app-wide instance; tests monkeypatch this.
spirit_service = SpiritService()