Foundation Models (FMs) such as GPT, LLaMA and diffusion models have been transformative in AI, demonstrating remarkable versatility across tasks. However, compared to traditional cloud computing, the full potential of edge computing (such as edge servers, PCs, smartphones, and even IoTs) - with its inherent cost, latency, and privacy benefits - remains untapped for deploying these models. In this project, we will develop a new system involving both software and hardware for Edge FM-as-a-Service that harnesses the power of distributed, diverse, and collaborative edge devices, providing user-friendly access to FM services at the edge without the burden of expensive deployment or complex management.