System Prompt Generator
Fill in a role, task, audience, tone, constraints, and output format to assemble a clean system prompt. Everything runs in your browser, so nothing you type is uploaded.
About the System Prompt Generator
The System Prompt Generator builds a clean, well-structured system prompt from a few fields: role, task, audience, tone, constraints, and output format. Fill in what you need, skip what you don't, and the prompt assembles itself as plain text you can paste into ChatGPT, Claude, the API, or any agent setup. Everything runs in your browser, so nothing you type is uploaded or stored.
How it works
- Fill in the fields you care about: role or persona, task, audience, tone, and output format. Empty fields are left out of the result.
- Add constraints one at a time to list rules the model must follow, and remove any you don't want.
- Optionally add a few example inputs and outputs to show the model the style you expect.
- Read the assembled prompt in the output box and click Copy to use it.
Features
- Live assembly: the prompt updates as you type, with clear sections for Role, Task, Audience, Tone, Constraints, and Output format.
- Repeatable constraints list so you can add and remove rules without editing raw text.
- Optional examples section for showing input/output pairs.
- Empty fields are skipped, so the result stays clean and only includes what you set.
- One-click copy, and everything stays in your browser.
Frequently asked questions
What is a system prompt?
A system prompt is the instruction that sets up how a model behaves before the conversation starts. It defines the role, the task, the tone, and any rules to follow. This tool helps you write one in a consistent, readable structure.
Which models does the output work with?
The result is plain text with labelled sections, so it works with any chat model or API that accepts a system or developer message, including Claude, ChatGPT, and most agent frameworks.
Do I have to fill in every field?
No. Fill in only what you need. Any field you leave blank is left out of the assembled prompt, so the output never contains empty headings.
Is my input sent anywhere?
No. The prompt is assembled entirely in your browser with JavaScript. Nothing you type is uploaded, saved, or logged.
How should I write constraints?
Keep each constraint short and specific, one rule per line, such as 'Answer in under 200 words' or 'Cite a source for every claim'. The tool lists them as a numbered set the model can follow point by point.