Oh My Pi now supports custom model integration with multiple inference backends, including vLLM, llama.cpp, and SGLang, enabling users to run locally hosted LLMs with flexible deployment options. The update expands compatibility for various model formats and optimizes performance across hardware configurations.