Prompt workspace
Turn rough intent into a reusable prompt with modes, patterns, guardrails, schemas, source rules, examples, variables, and separate system/user output. Free, runs in your browser, nothing uploaded.
01 / mode
Pick the closest job, then tune the pattern and controls.
02 / task
03 / pattern
01
Choose the job — research, coding, writing, data, agents, extraction, support, visual, and more. Each mode loads sensible defaults for pattern, tone, source policy, and guardrails.
02
Fill in the task, audience, context, and any source input or examples. The coverage checklist shows which parts of a strong prompt you still have missing.
03
Toggle behavior rules like no invented facts, uncertainty flagging, and safe tool use. Add must-includes, things to avoid, schemas, and eval cases for reusable prompts.
04
Copy the full prompt, or just the system or user half, into ChatGPT, Claude, Gemini, Cursor, or your own agent stack. Download it as a text file to keep it in your prompt library.
A good prompt is specific about four things: who the model should be, what you want done, what material it is working with, and what a good answer looks like. Vague prompts produce generic answers; the builder forces the specifics that change output quality.
Yes. The generated prompt is plain text with labeled sections, which every major chat model and coding assistant understands. Use the Full output for chat apps, or the separate System and User outputs when wiring prompts into an API or agent framework.
A pattern is a reusable instruction style, such as few-shot (learn from examples), plan-then-answer (reason privately, show a short rationale), critique-and-refine, rubric scoring, extraction, or grounded research. Patterns shape how the model approaches the task, not just what it produces.
Use a schema whenever the answer feeds another system — JSON for APIs, pipelines, and extraction tasks. The builder flags schema as missing whenever you pick a JSON output format or the extraction mode, so parseable output is never an afterthought.
Yes. Everything runs in your browser, nothing is uploaded, and there is no account or usage limit. Build as many prompts as you want and keep them in your own library.