20 complete expert prompts, one tap to copy. Or just type up top and we build one for you.
Already have a draft? Paste it — the engine scores it live and shows exactly what's missing.
Your draft
Live score
Stronger version · added lines highlighted
Everything above is instant and runs on any device. Power users can optionally load a small AI model that runs on your own GPU to rewrite prompts even further — nothing leaves your machine.
Load an on-device model (optional, advanced)
Agent trace
Why this is free — and private
- It's just HTML/CSS/JS on GitHub Pages. No server, no database, no API bill — so it can't run out of credits.
- The prompt engine is pure JavaScript. A tiny dual-encoder ("two-tower", the CLIP-family shape) embeds your words and every intent into the same space and cosine-scores them live on each keystroke — blended with a rule engine, softmaxed into the confidence bars you see. The quality rubric and the 20 expert scaffolds run the same way: in your browser, instantly, on any phone.
- Personalization is a computation about you, for you. A small two-tower model learns how you like your answers and lives entirely in your browser's local storage. It never leaves your machine, respects Do-Not-Track, and you can reset or turn it off any time.
- Optional on-device AI. Load a model and it runs on your GPU via WebGPU — the site owner pays $0, your text never leaves your device.
Built by Sugeerth Murugesan. The thesis, demonstrated by this page: useful AI tools don't need expensive infrastructure — they need good scaffolding and a little personalization.