Oh My Pi (omp) supports custom models through local inference servers like vLLM, llama.cpp, SGLang, and gateways by configuring ~/.omp/agent/models.yml. Recent updates require renaming custom providers and adding Qwen template settings for reasoning effort compatibility.