System Prompt

---
sidebar_position: 1
---

# System Prompt

---

## What it is

Think of a system prompt like the briefing a manager gives a new employee on their first day — before any customer walks in, before any task begins. It establishes who the employee is, what they are allowed to do, and how they should behave in every interaction that follows.

A system prompt is a message sent to a language model before any user turn, using a reserved message role (`system`, `developer`, or an equivalent top-level field). It primes the model's operating context: its persona, constraints, output contract, and available background knowledge. The model is trained to treat instructions in this role with higher authority than instructions from the user turn.

It is not stored on the server between requests. Your application is responsible for prepending the system prompt to the full message array on every single API call.

---

## How it works

### Position in the context window

The 10-second version: the context window is an ordered sequence of tokens. The system prompt always occupies the lowest-indexed positions — it comes first, every time.

The full payload for any request looks like this: