How it works
Modern LLM APIs accept a system prompt as a separate field, applied to every request in the conversation. It's where you set the model's persona, give it tools, define output format, and specify constraints. A well-written system prompt is often the difference between a good agent and a bad one.
Example
A B2B SaaS support agent's system prompt: 'You are a senior customer support specialist for [Product]. Answer in 2-4 sentences max. Use only the tools provided. If unsure, escalate. Never invent product features.' Plus tool definitions and brand voice guidelines.