app/llm/openai_chat.py

"""Throttled OpenAI chat completions for auxiliary LLM call sites."""

from __future__ import annotations

from typing import Any

from app.async_executor import run_sync_in_executor
from app.config import settings
from app.llm.llm_throttle import throttled_llm_call


async def chat_completions_create_raw(
    *,
    messages: list[Any],
    model: str | None = None,
    max_tokens: int = 800,
    temperature: float = 0.0,
    response_format: dict[str, str] | None = None,
    phase: str = "chat",
    section_id: str | None = None,
    api_key: str | None = None,
    tenant_id: str | None = None,
) -> str:
    """Chat completion with arbitrary message content (text or multimodal)."""
    model_name = model or settings.chat_model
    key = (api_key or settings.openai_api_key or "").strip()
    if not key:
        return ""

    kwargs: dict[str, Any] = {
        "model": model_name,
        "messages": messages,
        "max_tokens": max_tokens,
        "temperature": temperature,
    }
    if response_format is not None:
        kwargs["response_format"] = response_format

    if settings.enable_async_pipeline:
        from openai import AsyncOpenAI

        from app.llm.llm_throttle import make_cache_hit_slot
        from app.llm.prompt_cache import (
            log_openai_cache_usage,
            normalize_messages_for_caching,
            openai_extra_kwargs,
            prompt_caching_active,
        )

        client = AsyncOpenAI(api_key=key)
        if prompt_caching_active() and isinstance(messages, list):
            kwargs["messages"] = normalize_messages_for_caching(messages)
            kwargs.update(
                openai_extra_kwargs(
                    phase=phase, model=model_name, tenant_id=tenant_id
                )
            )

        cache_slot = make_cache_hit_slot()

        async def _call() -> str:
            resp = await client.chat.completions.create(**kwargs)
            if prompt_caching_active():
                cache_slot[0] = log_openai_cache_usage(
                    resp, phase=phase, section_id=section_id
                )
            return (resp.choices[0].message.content or "").strip()

        return await throttled_llm_call(
            phase=phase,
            section_id=section_id,
            cache_hit_out=cache_slot,
            call=_call,
        )

    def _sync() -> str:
        from openai import OpenAI

        from app.llm.llm_throttle import throttled_sync_llm_call

        client = OpenAI(api_key=key)

        def _call() -> str:
            resp = client.chat.completions.create(**kwargs)
            return (resp.choices[0].message.content or "").strip()

        return throttled_sync_llm_call(phase=phase, section_id=section_id, call=_call)

    return await run_sync_in_executor(_sync)


async def chat_completions_create(
    *,
    messages: list[dict[str, str]],
    model: str | None = None,
    max_tokens: int = 800,
    temperature: float = 0.0,
    response_format: dict[str, str] | None = None,
    phase: str = "chat",
    section_id: str | None = None,
    api_key: str | None = None,
    tenant_id: str | None = None,
) -> str:
    """Return assistant message content from a text-only chat completion."""
    return await chat_completions_create_raw(
        messages=messages,
        model=model,
        max_tokens=max_tokens,
        temperature=temperature,
        response_format=response_format,
        api_key=api_key,
        phase=phase,
        section_id=section_id,
        tenant_id=tenant_id,
    )