Scepsy is a serving system that orchestrates multiple large language models (LLMs) and tools for agentic workflows, achieving up to 2.5x higher throughput and 1.0-3.3x lower latency compared to existing systems.