The extraction pipeline accesses LLMs through a LiteLLM proxy that is OpenAI-compatible so that model routing can be handled behind a unified interface.

LLM access: Extraction pipeline uses a LiteLLM proxy

— PYRANA_PLATFORM_ARCHITECTURE_REFERENCE.md

(OpenAI-compatible)

— PYRANA_PLATFORM_ARCHITECTURE_REFERENCE.md

for model routing

— PYRANA_PLATFORM_ARCHITECTURE_REFERENCE.md