Learn
What is a system prompt?
Last reviewed July 16, 2026
A system prompt is the instruction layer an application sends a model before any user message: who the model is acting as, what rules it must follow, what tone to use, and how to use its tools. The model treats it as higher-priority context for the whole conversation. Every serious AI product has one; the visible chat is always riding on instructions you do not see.
What belongs in a system prompt
Role and scope (what the agent is for), behavioral rules (what it must always or never do), output conventions (format, language, tone), and tool guidance (when to search, when to ask). Facts that change per conversation belong in messages; knowledge that should persist across sessions belongs in memory, not in an ever-growing prompt.
System prompts count as context
The system prompt is resent with every request, so its tokens are billed on every message and consume context window. Long prompts also dilute attention. The craft is compression: the strongest system prompts are short, specific, and testable, with details moved to documents or skills the model pulls in when needed.
idapt agents make the instruction layer explicit: each agent carries editable instructions, with memory and skills layered on top, so behavior is configured once rather than retyped per chat.
Frequently asked
See it in practice
The fastest way to understand it is to use it. Start free, no credit card.
Try idapt free