Blog/What is a system prompt?
Definition

What is a system prompt?

By Samuel Seidel · Published September 9, 2026

A system prompt is the block of instruction text supplied to a language model before any user input, telling it how to behave, what role to take, what to avoid, and how to format its answers. The model reads it first, treats it as higher-priority guidance than the user's messages, and applies it to every turn that follows.

How it fits into a request

Most chat-style model APIs accept a conversation as a sequence of messages, and the system prompt is typically the first one, tagged separately from user and assistant turns. It might set a persona ("you are a customer support agent for X"), a constraint ("never provide medical advice"), an output format ("respond only in valid JSON"), or background context the model should treat as given. It's set once and applies for the whole session, so the user never has to repeat it.

Who controls it depends on who's hosting

On a self-hosted deployment, the operator sets the system prompt directly, as a config value or file passed to the inference server, and it contains exactly what they wrote. A hosted API product often adds a second, invisible layer: the vendor's own system instructions, placed ahead of whatever the developer supplies. Those instructions usually aren't published in full and can change without notice, so a developer on a hosted API is building on a baseline they can't see or control.

Why that difference matters for self-hosted deployments

When you run your own model, the system prompt you write is the whole system prompt. If you need predictable behavior, that matters: what you write is what runs, and no vendor update to hidden instructions can change your deployment's behavior without you knowing. You can also audit and version the exact instructions in use, which matters when behavior has to be reproduced or explained later.

What it doesn't do

A system prompt shapes behavior; it doesn't enforce it the way a hard-coded rule in application logic would. A well-written system prompt makes deviation less likely, but a model can still be pushed off its instructions by adversarial or unusual input, particularly with weaker or smaller models. Anything that must never happen, whatever a user types, needs a check outside the model; a more forceful sentence in the prompt won't guarantee it.

System prompts vs other kinds of instruction

A system prompt is distinct from a user message, which comes from whoever is chatting with the model in that turn, and from a developer-supplied function or tool definition, which describes what actions the model can take rather than how it should behave generally. Some APIs also distinguish a "developer" role from a "system" role, splitting instructions from the application builder and the end user's own custom instructions into separate priority tiers. The naming varies between providers and inference engines; the idea of instructions that apply to the whole conversation, not a single turn, is the same in all of them.

Where GPUwerk fits

On a GPUwerk Spark you run and configure the inference engine yourself, so the system prompt is whatever you set, with no hidden vendor layer on top. See the glossary for related terms, or the private LLM hosting page for how deployments are set up.

Related pages

First top-up: pay $20, get $30 in credit

Set your own instructions, with nothing hidden underneath.

One dedicated Spark, $0.79/hour, deployed in minutes in Europe, North America or East Asia.

Create account