app/llm_factory.py

"""
llm_factory.py — Creates LangChain ChatOpenAI instances pointing at OpenRouter.
All LLMs in this project are built through this single factory so config
stays in one place.
"""
from langchain_openai import ChatOpenAI

from app.config import settings


def make_llm(
    model: str,
    temperature: float | None = None,
    max_tokens: int | None = None,
) -> ChatOpenAI:
    """
    Return a LangChain ChatOpenAI client configured for OpenRouter.

    Args:
        model:       OpenRouter model string  e.g. "arcee-ai/trinity-large-preview:free"
        temperature: Override; defaults to settings.PANEL_TEMPERATURE
        max_tokens:  Override; defaults to settings.PANEL_MAX_TOKENS
    """
    if not settings.OPENROUTER_API_KEY:
        raise ValueError(
            "OPENROUTER_API_KEY is not set. "
            "Add it to your .env file or HF Space Secrets."
        )

    return ChatOpenAI(
        model=model,
        openai_api_key=settings.OPENROUTER_API_KEY,
        openai_api_base=settings.BASE_URL,
        temperature=temperature if temperature is not None else settings.PANEL_TEMPERATURE,
        max_tokens=max_tokens if max_tokens is not None else settings.PANEL_MAX_TOKENS,
        default_headers={
            "HTTP-Referer": "https://codedebug.local",
            "X-Title":      "CodeDebug Multi-LLM Assistant",
        },
    )


def make_panel_llm(model: str, temperature: float | None = None) -> ChatOpenAI:
    """Panel LLM with panel token budget."""
    return make_llm(model, temperature=temperature, max_tokens=settings.PANEL_MAX_TOKENS)


def make_judge_llm() -> ChatOpenAI:
    """Judge LLM with larger token budget and near-zero temperature."""
    return make_llm(
        settings.JUDGE_MODEL,
        temperature=settings.JUDGE_TEMPERATURE,
        max_tokens=settings.JUDGE_MAX_TOKENS,
    )