rag/providers.py

from llama_index.core.llms import LLM

from rag.config import MAX_ANSWER_TOKENS, PROVIDERS


def make_llm(provider: str, api_key: str, model: str, **kwargs) -> LLM:
    """Build an LLM client for a single request.

    Args:
        provider: a slug from `config.PROVIDERS` ("openai" | "gemini" | "anthropic").
        api_key: the key the user pasted into the UI.
        model: a model ID, one of the provider's `models`.
        **kwargs: passed through to the underlying LlamaIndex class.

    Raises:
        ValueError: unknown provider, or a blank API key.
    """
    spec = PROVIDERS.get(provider)
    if spec is None:
        raise ValueError(
            f"Unknown provider {provider!r}. Expected one of: {', '.join(PROVIDERS)}"
        )

    api_key = (api_key or "").strip()
    if not api_key:
        raise ValueError(f"No API key provided. Please paste your {spec.key_label}.")

    # We leave temperature at each provider's default, because the gpt-5 family
    # rejects any value other than 1.
    kwargs.setdefault("max_tokens", MAX_ANSWER_TOKENS)

    if provider == "openai":
        from llama_index.llms.openai import OpenAI

        return OpenAI(model=model, api_key=api_key, **kwargs)

    if provider == "gemini":
        from llama_index.llms.google_genai import GoogleGenAI

        return GoogleGenAI(model=model, api_key=api_key, **kwargs)

    if provider == "anthropic":
        from llama_index.llms.anthropic import Anthropic

        return Anthropic(model=model, api_key=api_key, **kwargs)

    raise ValueError(f"Provider {provider!r} is configured but not implemented.")