cores/onnx/__init__.py

"""
ONNX Core — centralized ONNX Runtime model management.

This is the SINGLE place where ONNX models are loaded.  Every provider
that needs ONNX inference calls `get_session()` or `get_model()` here.

Design guarantees:
  1. Each model loads exactly ONCE per process (via EmbeddingCache).
  2. Models are lazy — only loaded when first requested.
  3. CPU thread count is capped (configurable, default 2) for low-RAM.
  4. Models are auto-downloaded on first use (if enabled in settings).
  5. No provider imports onnxruntime directly — they go through here.

This keeps onnxruntime as an OPTIONAL dependency: if it's not installed,
providers that need it gracefully report `is_available() -> False`.
"""

from cores.onnx.session import (
    is_onnx_available,
    get_session,
    run_inference,
    ONNXModel,
)
from cores.onnx.downloader import download_model, ensure_model, model_path

__all__ = [
    "is_onnx_available",
    "get_session",
    "run_inference",
    "ONNXModel",
    "download_model",
    "ensure_model",
    "model_path",
]