# Cloud-adapted __init__.py — lazy imports only.
# Only llm_model and tts_model are used on the cloud server.
# Other models (ocr, speech, vision) are not available.