wtf is prometheus-ai?

Prometheus AI is a web application that takes a short, rough prompt from the user and rewrites it into a longer, more structured prompt aimed at a specific AI model. The README frames the problem as people writing prompts the way they write text messages, vague and missing context, and then being disappointed by the results. The app tries to fix that by applying prompt engineering rules automatically. The user supplies their own API key for an AI provider. The key is kept in the browser, not on any server, and the app does not require a sign up or store data on its side. Supported providers listed in the README include OpenRouter, Anthropic Claude, OpenAI, Mistral, and Google AI. OpenRouter is suggested as a way to reach many models through a single key. When rewriting a prompt, Prometheus AI tailors the output to the target model. The README spells out different focuses per model, such as XML structure and explicit constraints for Claude, direct task framing for ChatGPT, multimodal clarity for Gemini, stable prefixes for Grok, compact instructions for Mistral, and research scoping for Perplexity. The user can also pick from ten use case categories, including text, image, code, websites and apps, AI agents, research, marketing, learning, and documents. Each category has more specific subcases, with the README claiming over sixty in total. Files such as text documents, PDFs, and images can be attached alongside the prompt. The project is a Next.js 16 application written in TypeScript, with Tailwind CSS for styling, Framer Motion for animation, Zod for validation, and Vitest for tests. The README shows a directory layout where the prompt engineering logic lives under apps/web/lib/transform, with files for a rule compiler, provider API calls, intent detection, model configurations, and use case definitions. Running it locally needs Node.js 18 or newer. The README gives the standard clone, npm install, and npm run dev steps, with the app available at localhost:3000. No environment file is needed because the API key is entered in the browser. For production it can be deployed to Vercel, Railway, Render, Fly.io, Coolify, or any Node compatible host. The license is MIT.