100% local · 5 frameworks

Build mega-prompts that actually work

Fill in a few simple fields, pick a proven prompt-engineering framework, and copy a structured mega-prompt ready for ChatGPT, Claude, Gemini, or any LLM. 100% client-side — nothing leaves your browser.

CARE · PREP · RACE · STAR · CRISPE Persona selector Live preview Save favorites

1. Choose a framework

Each framework shapes the structure of your mega-prompt. Pick the one that fits your task.

2. Pick a persona

Sets the role and expertise the LLM should adopt.

3. Fill in the task

The topic is required — everything else is optional but recommended.

Live preview

0 chars · ~0 tokens
Enter a topic to generate your prompt…

How to use

Four simple steps — your prompt is assembled locally and never sent anywhere.

1

Pick a framework & persona

Choose a prompt-engineering structure (CARE, PREP, RACE, STAR, or CRISPE) and a persona that matches the expertise you need.

2

Fill in the fields

Type your topic, audience, tone, output format, and length. Add examples or constraints to sharpen the output.

3

Watch the live preview

The mega-prompt assembles itself in real time on the right, with a token estimate so you can stay under your model's limit.

4

Copy, download, or save

Copy the prompt to your clipboard, download it as a .txt file, or save it to favorites for later reuse — all stored locally.

The 5 frameworks

Each framework is a proven structure for organising a prompt. Pick the one whose flow matches what you are asking for.

Frequently asked questions

No. The AI Prompt Formatter runs entirely in your browser. Your inputs, generated prompts, and saved favorites never leave your device. You can verify this in your browser's DevTools → Network tab — no requests are made when you type or save.

It depends on the task. Use CARE for task-driven requests where context shapes the result. PREP is great for persuasive, point-first communication. RACE works well for role-and-context tasks. STAR is ideal for situational, scenario-based prompts. CRISPE gives the most structure when you need explicit personality and experimentation.

The estimate uses the common heuristic of roughly 4 characters per token (characters ÷ 4, rounded up). This is only an approximation — actual token counts vary by tokenizer (GPT, Claude, Gemini all differ) and by language. Use it as a ballpark to avoid blowing past context limits, not as an exact number.

Favorites are saved in your browser's localStorage under the key promptfmt-favorites. They persist across page reloads on the same browser, but they are not synced anywhere and will be lost if you clear browser storage or use a different device/browser.

Favorites are read-only snapshots. To edit, load the favorite back into the form with the "Load" button, modify the fields, and save it again (optionally under a new name). You can then delete the old one.

Yes. The generated prompt is plain text that works with ChatGPT, Claude, Gemini, Mistral, Llama, and any other chat-based LLM. Just paste it into the model's input box. Some models may respond better to certain frameworks, so feel free to experiment.

Yes. The app is a single self-contained HTML file with no external dependencies. After the first load you can disconnect from the internet and keep using it. You can also save the page (Ctrl+S) and run it from your disk.