Assemble a system prompt from eight modular sections (role, context, constraints, output format, examples and more) starting from one of five persona templates, with the full prompt previewed live as you build.
Build structured system prompts with modular sections. Toggle sections on/off, use templates as starting points, then copy the compiled prompt for any LLM.
Quick Start Templates
Enable sections and add content to see the compiled prompt here...
Good system prompts are structured, not long. The eight sections exist to prompt you for the things people habitually omit. What the model must refuse, what format the output takes, what it should do when it lacks information. Those omissions cause most of the behavior people describe as the model ignoring instructions.
Length has a real cost. A system prompt is sent with every request, so a bloated one is a recurring charge on every call you make. Build it deliberately, then cut anything that has not changed an output, and enable prompt caching, which makes a stable system prompt dramatically cheaper at volume.