System prompts

System prompts are the foundational, persistent instructions that define an LLM's persona, constraints, and overall behavioral alignment.

System prompts function as the 'BIOS' configuration for a Large Language Model (LLM): they are a foundational instruction set, loaded once, that dictates the model's core conduct (e.g., 'You are a concise, trustworthy summarization engine'). Unlike a user prompt, the system prompt remains constant across all interactions, ensuring consistency. Developers use this mechanism to set specific guardrails, assign a role (e.g., 'expert TypeScript engineer'), and enforce output formatting (e.g., 'always use Markdown for code blocks'). This steerability is critical for production-grade AI applications, directly impacting response reliability and brand voice adherence.

https://www.promptingguide.ai